MASTER THESIS
Gibrat’s Law for Cities and
Structural Transformation of Asian Countries
Author:
Diana Sekarayu Karunia
Student number: S2844591
Email:
d.s.karunia@student.rug.nl
Supervisor:
Prof. Dr. Steven Brakman
Co-Assessor:
Dr. Raquel Ortega Argiles
MSc International Economics and Business
Faculty of Economics and Business
2 Abstract
Gibrat’s law states that the growth of a city is independent of its size or that there is a proportionate growth of all city size. However, a country’s spatial distribution is changing following the development and importance of the sector during structural transformation due to the urbanization process. This examines whether Gibrat’s law applies to the cities in East Asia and Southeast Asia and whether a city spatial distribution depends on the structural transformation process. We find that Gibrat’s law does not hold for aggregate municipalities and urban areas. Three country development level groups display mixed results and demonstrate that city spatial distribution depends on the structural transformation process. Decreasing share of manufacturing activities correlates to a faster growth of large cities. Moreover, we highlight the different result concerning the use of two spatial units (municipalities and urban areas) to test Gibrat’s law. Finally, the results will be used to formulate policy recommendation for the respective countries.
3 Table of Contents
Abstract ... 2
1. Introduction ... 4
2. Literature Review and Hypothesis ... 6
2.1 Gibrat’s Law and Agglomeration Economies ... 6
2.2 Gibrat’s Law and Structural Transformation ... 9
2.3 Hypothesis ... 12
3. Methods and Data ... 13
3.1 Methods ... 13
3.2 Data ... 14
4. Empirical Results ... 16
4.1 Municipalities ... 16
4.2 Urban Areas ... 18
4.3 Country Development Level ... 20
4.4 Analysis of Gibrat’s law and Structural Transformation ... 23
4.5 Policy Recommendations ... 25
5. Conclusion ... 27
References ... 29
Appendix 1 ... 30
4 1. Introduction
As countries experience economic development, they also tend to experience increasing urbanization. United Nations data shows that urbanization in developing countries is going faster than urbanization process in developed countries. The annual growth rate of the urbanized population in developing countries ranged between 2.7 percent and 4.2 percent between 1950 and 2010, outpacing the urban growth rates in developed countries which ranged between 0.6 percent and 2.4 percent in the same period. Urbanization in Asia occurs with structural transformation and cities are “production cities” where labor mixed between tradable and non-tradable work. In these countries, urbanization is closely related to the share of manufacturing and services in GDP, and standard models in structural transformation. In contrast, urbanization without structural transformation occurs in Sub-Saharan Africa due to a higher share of natural resources exports (Gollin et al., 2013).
Over the past decade, the global economy has entered a new era in East Asia Pacific region. From the late 1960s to the late 1980s, global emergence of international division of labor occurred marked by the shift of labor-intensive assembly operations to a newly industrializing economies, first in the four “dragon” economies of South Korea, Taiwan, Hong Kong, and Singapore, then into the ASEAN countries of Malaysia, Thailand, Indonesia and to a lesser extent, the Philippines and Vietnam and recently into China. Globalization of production, commerce and finance capital requires a physical geography of cities, urban networks and transport and communications linkages to affect its expanding spatial reach, thus encouraging urbanization. It takes East Asia Pacific only 10 years to urbanize the equivalent number of people that moved to urban areas in Europe in 50 years. Although the urban population in Asia is large, only 36 percent of its total population living in urban areas, indicating that urban growth still continues in the future (World Bank, 2015).
5
The previous findings of city size distribution are facing two robust empirical regularities: Zipf’s law and Gibrat’s law. Zipf’s law or also known as the rank-size rule applies if city sizes follow a power law with exponent 1. If this holds, the largest city would have twice the population of the second-largest city, three times the population of the third-largest city, and so on. The second empirical regularity, Gibrat’s law, states that the growth of a city is independent of its size. All cities, either big or small should be expectedly grow at the same average rate or there is a proportionate growth. Gabaix (1999) shows that if Gibrat’s law holds for city growth, then the city size distribution must follow Zipf’s law. To accept Gibrat’s law, cities facing increasing return to scale should equalize their agglomeration force and their spreading forces. However, it seems obvious that for any urban system to develop, some places must attract more people and grow faster than others. Otherwise, a homogenous population distribution across space would have occurred. Moreover, it is generally acknowledged that the largest cities and metropolitan areas tend to grow faster than their smaller counterparts (Black and Henderson, 2003). Thus, it can be assumed that if Gibrat’s law applies, the agglomeration forces of the large cities might not be strong enough to offset the spreading forces so that the growth does not exceed the smaller locations.
Urbanization in East Asia Pacific is an essential phenomenon as part of structural transformation. This process can either bring advantages or disadvantages for each country and might affect the lives of hundreds of millions of people during the coming decades. Although the growth of urban areas provides opportunities for the poor, urban expansion, if not well planned, can also exacerbate inequality in access to services, employment, and housing. Hence, it is appealing to examine city size distribution for East Asia Pacific countries during the massive urbanization process at this moment, in addition to little previous studies about this region. Previous literature testing Gibrat’s law find various results. Klein and Leunig (2015) find evidence of Gibrat’s law during the British industrial revolution. Meanwhile, Michaels et al. (2012) find a violation of Gibrat’s law during the United States structural transformation.
6
law, large cities might not have strong agglomeration forces required to capture the benefit of urbanization. Whether or not a country’s city size distribution follows Gibrat’s law, Portnov et al. (2012) affirms that the result can help formulate effective development policies aimed at enhancing city growth in priority areas or at restricting it in overpopulated areas. Thus, the results of this study will help to conclude policy suggestions for cities in East Asia Pacific.
Our findings show mixed results of Gibrat’s law from the ten countries observed. First, aggregate municipalities data and urban areas data of all countries show rejection of Gibrat’s law. Further investigation shows that while larger locations in high-income countries grow faster than smaller locations, the other two groups show city size distribution reflecting urbanization process. People move to more urbanized areas reflected by smaller locations exhibit dispersion while medium locations exhibit faster growth. However, largest locations show dispersion again indicating agglomeration economies of large areas may be offset by higher spreading forces. Moreover, we find correlation between decreasing share of manufacturing activities to faster growth of large cities. Finally, policy recommendations are made concerning these results.
This paper set up as follows: the next chapter will discuss the theories and previous findings of Gibrat’s law and structural transformation followed with the hypothesis; the third chapter deals with the research method and data used in this study; the fourth chapter consists of the result and analysis; and lastly the conclusion is provided.
2. Literature Review and Hypothesis
2.1 Gibrat’s Law and Agglomeration Economies
City population is distributed unevenly with a strong clustering or agglomeration of economic activity in various important centers. This phenomenon repeats itself at various levels of aggregation and the distribution across space is not random but follows a remarkable pattern known as rank size distribution. When observing this phenomenon, two empirical regularities are found (Eeckhout, 2004). The first is known as Zipf’s law which is a special case of the rank size distribution. In 1949, George Kingsley Zipf showed that within a country, the size of the largest cities is inversely proportional to their rank of the size of the city. According to Zipf’s law, the distribution of cities should follow a Pareto distribution or power distribution with exponent equals to one, with R(n) = An-a or in a log function:
7
for city i in time t, where R is the city rank and n is city population size. The rank size rule is obtained for at = 1. For at well below one indicating urban concentration, and at more than 1
indicating dispersion of the city distribution. In other words, Zipf’s law holds if the largest city is precisely k times as large as the kth largest city. For example, in the United States, New York City is roughly twice the size of the second largest city, Los Angeles, and about three time the size of the third largest city, Chicago. To explain Zipf’s law, one refers to Gibrat’s law which is also the second empirical regularity stated by Eeckhout.
Gabaix (1999) derives Zipf’s law from Gibrat’s law. Gibrat’s Law states the growth rate of a city is independent of its size, or smaller cities on average do not grow faster or slower than larger cities. The simplest formulation of Gibrat’s law applied to city size is as follows:
ln Xit – Xit-1 = Δln(Xit) = α0 + α1 ln(Xit-1) + εit (2)
where Xt is the population size of a locality at time t. The null hypothesis, implied by Gibrat’s
law, is α1 = 0.
The proportionate growth process generates the lognormal distribution, not the Pareto distribution as necessary for Zipf’s law. Gabaix (1999), however, shows that proportionate growth processes can generate Zipf’s law at the upper tail of the distribution, and while considering all cities and not just the upper tail, it results in lognormal distribution or Gibrat’s law. Gabaix elaborates two different routes to Zipf’s law on the upper tail distribution: either cities behave like constant-return-to-scale economies or due to endogenously counterbalancing effects of negative externalities in increasing return to scale environment. In the CRS economy, the differences of externalities between large cities and small cities are bounded. In increasing return to scale economy, large cities experience equilibrium condition between positive and negative externalities. Thus, cities follow Gibrat’s law if they are characterized either by constant return to scale or external economies of scale of which its positive externalities compensate its negative externalities.
8
called pure external economies of scale. The most common example is information spillovers from industry-wide production increase that accumulate the knowledge of each firm to further increase its own production. In contrast to pure external economies, pecuniary external economies do not alter technological production function of individual firm. It is transmitted by the market through price effects for individual firm. This type of external economies also exist in geographical economies through a love-of-variety effect in a large local market. Consumer’s utility will be higher the more number of varieties of manufactured goods are available. It can be concluded that the existence of positive externalities leads to geographical concentration of economic activities in the city.
There are two types of agglomeration economies that can be distinguished, localization economies and urbanization economies. In localization economies, city growth is fostered by specialization because economies of scale only occur to firms within the same industry. With urbanization economies, firms regardless of the industry will stimulate city growth because of their collective existence in the city. However, which type of agglomeration economies is more important still can be debated since previous studies find different evidences.
New economic geography theory explains the emergence of large agglomerations by the combination of internal increasing returns to scale and transportation cost and emphasizes linkages between firms and suppliers as well as between firms and customers (Schmutzler, 1999). At first, a region enjoys first-nature advantage of natural resources that encourages agglomeration. Increasing return to scale then fosters geographical concentration of production. Transportation cost makes certain locations attractive because the proximity to market and suppliers. Finally, concentration of production attracts mobile factors of production. This process is called second-nature geography. It results in more jobs and consumption opportunities that attracts more and more firms and workers which stimulates further agglomeration or circular causation effect. These agglomeration forces have a countervailing effect, the spreading forces. Concentration of activities results in higher prices of land and housing, increased pollution, and congestion. These spreading forces may cause a city to spread to its peripheral areas and may cause smaller cities to emerge.
9
competition, high transport costs and land use pattern. Therefore, firms are likely to cluster within large metropolitan areas when they sell differentiated products and transport cost are low. As a result of firms locate together, cities provide a wide array of final goods and specialized labor markets that, furthermore, make them more attractive to consumer or workers. To conclude, agglomerations are the outcome of cumulative processes involving both the supply and demand sides.
The distribution of people and economic activity across space is uneven, and while continuously changing, is not random. Economic factors are clearly play part as the determinant of the dynamics of city populations. For example, San Francisco Bay area experienced higher-than-average population growth during the high-technology industry boom. In contrast, Detroit in the last decade experienced a decrease in population as the manufacturing industry there suffered a severe decline. This confirms that agglomeration and mobility of the population between locations are connected to economic activity.
2.2 Gibrat’s Law and Structural Transformation
Structural transformation relates to the spatial distribution of economic activities as explained by Desmet and Henderson (2014). A country’s spatial organization is changing following the development and importance of sector. A country that moves from agriculture to manufacturing will be expected to show different spatial distribution to a country that has moved to service sector. Hence, we can expect varying results of Gibrat’s law regarding each country and country groups.
post-10
industrialization era since 1900 is characterized by declining of industrial sector and increasing share of service sector. The negative externalization such as congestion becomes an important factor.
City size distribution during such industrialization has been studied before, for example Klein and Leunig (2013) who examine Gibrat’s law during British Industrial Revolution. They find the evidence of Gibrat’s law as small cities and large cities grow proportionately. Although small cities are not growing fast, high commuting cost in larger cities reduce the positive externalities of high productivity industries, thus resulting in the proportionate growth. During this era, there were no planning laws to separate industrial and residential housing to maximize the extent of agglomeration economies in large cities. Hence, it results in tight urban areas with mixed use land patterns.
Michaels et al. (2012) find interesting results from their study observing locations’ growth during structural transformation of the United States in both rural and urban areas. They observe three level of location density: low, medium, and high. In low density areas, there is negative association between initial population density and density growth that contrast to positive association in intermediate density areas. Yet, in high density areas, population growth is uncorrelated with its initial population density or size leading to acceptance of Gibrat’s law. This pattern is the result of the differences in agriculture’s initial share of employment. In medium density locations, the share of agriculture in initial employment is decreasing in population density, structural transformation raises population growth at higher densities with lower shares of agriculture employment. It means that non agriculture sector generates agglomeration economies from high productivity industries that makes larger cities grow faster. In contrast, in the high density locations non agriculture sector productivity is assumed to be constant that leads to constant growth of the cities.
11
Ishii, 1998) and France (Duranton and Puga, 2000) also suggest the urban product-cycle hypothesis where product development occurs in major metro areas with diverse industrial-occupational environments. Once products and production processes are standardized, production is decentralized to smaller or more specialized cities with lower wage and lower land rent.
There are major differences between cities in advanced countries and developing countries as highlighted by Duranton (2015). The urban system of many developing countries often holds back economic growth. Although the largest cities in developing countries are the centers of innovation, there is little in terms of relocation of the production of mature products to specialized cities. This situation makes these cities larger than they should be and increases congestion. Smaller cities are stuck with the production of the most backward products. Ensuring the urban system acts as drivers of economic growth will be a major challenge for these countries. This lack of differentiation in urban functionality may limit the dynamism of cities in developing countries.
It is interesting to observe Gibrat’s law during the structural transformation in East Asia Pacific countries as there are only limited previous studies about this region. As countries undergo structural change, labor is relocated from rural agriculture to urban manufacturing and services. Massive urbanization may lead to negative externalities dominate over large cities so that they do not reap benefit from agglomeration economies. Moreover, different country groups can be observed: the more advanced countries in East Asia and developing countries in Southeast Asia. Different results may be found in these groups leading to different spatial policies recommendation.
12 2.3 Hypothesis
For this study, we will observe ten countries in Asia for period 2000-2010 in two different spatial units: municipalities (administrative unit) and urban areas. Countries observed are China, Japan, South Korea, Taiwan, Indonesia, Malaysia, Thailand, Philippines, Vietnam, and Cambodia.
For municipalities, we expect a positive relationship between initial density of the city and its growth. The reason is that urbanization process is still going rapidly in this region. Bigger cities with larger initial densities are more attractive to workers and firms because they offer more amenities, for example labor pooling. Workers can find suitable job easier when there are many firms located in larger cities. Love-of-variety effect also plays here, where it increases utility of a person who lives in a large city by offering more variety of products.
The result for urban areas is expected to be different: either Gibrat’s law holds or there will be negative relationship between initial density of the city and its growth rate following previous studies using urban areas as spatial unit (Soo, 2007; Black and Henderson, 2003). As urban areas consist of high density locations, spreading effect might dominate the agglomeration force as locations are getting denser. Smaller urban areas may have less spreading forces as negative external economies of scale within a city is a function of the overall size of the city (Henderson, 1974).
Since our literature review shows the correlation between city size distribution and a country’s structural transformation, we predict to see any difference of each country depending of its development level. Less developed countries that undergo the process of decreasing agriculture share and increasing manufacture share may follow similar pattern as previous study in Michaels et al. (2012). Structural transformation is followed by massive urbanization from rural areas to urban areas. Small locations exhibit dispersion, and in contrast, medium locations shows increasing growth as denser locations become more attractive to firms and workers. Larger locations may show proportional growth or negative growth if diseconomies of scale dominate the agglomeration forces. Meanwhile, high-income countries in Asia have moved to services sector. As service sector requires to locate in denser areas, large cities might exhibit concentration and faster growth.
13 3. Methods and Data
3.1 Methods
To test whether city growth follows Gibrat’s law, this study follows Black and Henderson (2003) and Soo (2007) in estimating the following regression
Δln(Xit) = α0 + α1 ln(Xit-1) + εit
where the dependent variable is city population growth and the independent variable is the initial city size at period t. Gibrat’s law implies that α1 = 0 because there is no correlation
between initial city size and city growth. Negative α1 indicates decreasing population size in
the time interval, while positive value implies increasing size (Portnov et al., 2012). The equation above can be estimated using simple OLS.
In this study, variable city density is used instead of city population to represent city size and city growth. Ciccone and Hall (1996) show that density is often favored relative to population because it appears to yield more reliable results. The reason is that density-based measures of agglomeration are more robust to zoning idiosyncrasies. Since agglomeration economies come from the benefit of locating near each other, a large population cannot say anything without determine if it locates in a wide or narrow area. Density is usually used to measure the agglomeration economies of a location, such as density-productivity relationship (Ciccone and Hall, 1996; Glaeser, 2010). However, since several previous studies used population to test Gibrat’s law, we include the regression using population growth as the dependent variable and initial population size as the independent variable in Appendix 2.
To explore whether there is any difference in city size distribution pattern, we regress data for each country and each specific development level. The countries are divided into four development levels based on the World Bank classification (see Appendix 1). High-income countries are Japan, South Korea, and Taiwan. Upper-middle-income countries are China, Malaysia and Thailand. Lower-middle-income countries are Indonesia, Philippines, and Vietnam. Meanwhile, Cambodia is the only low-income country. After we regress the aggregate data of all municipalities and urban areas, we run three regressions for each country development level: low & low-middle-income countries, upper-middle-income countries, and high-income countries. Finally, we run a regression of each individual country to see every country characteristics. We expect different pattern of city size distribution respective to the country and development level.
14
following the same method of Portnov et al. (2012), to test whether there is difference in the growth rates of different groups of cities. To choose which ANOVA method will be used, first, we should check the normality of distribution of the data. If the data does not follow normal distribution, the Levene test will be used, otherwise, the Bartlett’s test will be applied. These tests are used to examine the differences of the variances across group of countries with the null hypothesis is the variances of all groups are equal. The Levene test is less sensitive than the Bartlett’s test to departures from the normality distribution. However, after testing for the normality of distribution, it is shown that the density growth data does not follow normal distribution, so that the Levene test will be more appropriate to use.
3.2 Data
In observing Gibrat’s law, it is important to determine the data used either using data of municipalities or urban areas. On the one hand, using municipalities gives advantage as development disparities between municipalities may have a profound effect on their attractiveness and population growth rates in general. High-income region may provide better services and facilities that attract firms and workers, in contrast to poorly developed region which has lower facilities and service, thus less attractive to potential newcomers (Vaturi et al., 2004). On the other hand, using urban areas data give different analysis point. Urban area is an economically integrated unit formed by individual municipalities and function as an area. However, urban areas are likely to be heterogeneous due to diversity of population and uneven proximity of locations. Aggregating diverse geographic units imply substantial loss of information (Salway and Wakefield, 2005). Thus, in this study, two different spatial areas are used to find out which one is the best explaining the behavior of city distribution and to avoid the bias of the population density.
15
compared to larger cities. Therefore, if the sample size is low one may tend to accept the relative fulfilment of Gibrat’s law even when the behavior of the entire distribution may be different.
In this study, ten countries will be examined: China, Japan, South Korea, Taiwan, Indonesia, Malaysia, Thailand, Philippines, Vietnam, and Cambodia using 5782 municipalities and 1001 urban areas for period 2000-2010. These ten countries are chosen because they represent different level of economic development: the advanced and emerging parts of East Asia Pacific. Furthermore, it addresses to the availability of urban areas data from the World Bank database. First, area of municipality is based on census of each country Data available for each country cover different level of municipality. In regard to uniformity, second level administrative unit data will be used for all countries. Second, since the definition of urban area is different in each country, we rely on a single source that performs a measurement of urban areas. We obtain data of built-up areas from maps created by Moderate Resolution Imaging Spectroradiometer (MODIS) satellite data that has been used for the World Bank (2015) report before available in puma.worldbank.org. These maps rely on a geophysical definition of built-up area: built-up land refers to places dominated by the “built environment,” which includes all non-vegetative, human-constructed elements (roads, buildings, and the like) with greater than 50 percent coverage of a landscape. These built-up areas measure urban expansion within the municipality. Furthermore, with the World Bank unique identifier of agglomeration, built-up areas within the same urban agglomeration can be identified, and hence aggregated so that ones can see the extent of the urban agglomeration.
16
Previous studies of Gibrat’s law investigate its applicability to examine city growth using relatively short-term data, mainly for past decades, using data for either urban areas or municipalities (Portnov et al., 2012; Resende, 2004). Furthermore, this study objection is to give policy recommendation according to the results whether cities follow Gibrat’s law. Thus, examining city distribution for the last decade is sufficient information.
4. Empirical Results
4.1 Municipalities
Figure 1. (a) population 2000 to population growth of municipalities 2000-2010; (b) density 2000 to density
growth of municipalities 2000-2010
Firstly, to overview the correlation between initial population and population growth, scatter plot is drawn. From the scatter plot shown in Figure 1, a positive correlation between initial population of the cities in 2000 and population growth can be observed. Using untruncated data of all small and large locations, thus, the hypothesis of positive dependency of size and growth can be proven. However, for initial density and density growth (Figure 1b) the positive pattern is not seen clearly. Further, statistical test is needed to test the significance of the independency of city growth.
17
municipalities data, there is dependency of population growth to its initial density, rejecting Gibrat’s law.
Table 1. Regression result of municipalities data (N=5782) with density growth (ln) as the dependent variable
and density 2000 (ln) as the independent variable
Model Density 2000 (ln) Constant F R2
(1) All Municipalities 0.0119*** -0.0098 59.10*** 0.01 (0.002) (0.009) (2) Low& lower-middle-income -0.0158*** 0.2564*** 43.73*** 0.0197 (0.002) (0.013) (3) Upper-middle-income 0.0121*** -0.0227 8.43*** 0.0061 (0.004) (0.021) (4) High-income 0.0379*** -0.2630*** 1067.04*** 0.3245 (0.001) (0.007) (5) Cambodia -0.0577*** 0.4312 30.65*** 0.1512 (0.01) (0.047) (6) China -0.0031 0.08162*** 0.42 0.0013 (0.005) (0.027) (7) Indonesia -0.0377*** 0.3898*** 78.79*** 0.1932 (0.004) (0.023) (8) Japan 0.0400*** -0.2787*** 1696.25*** 0.5009 (0.001) (0.006) (9) Malaysia 0.0139* 0.0919*** 3.72* 0.0278 (0.007) (0.033) (10) Philippines 0.0068** 0.1143*** 5.94** 0.0037 (0.003) (0.016) (11) South Korea 0.0746*** -0.4426*** 55.71*** 0.2571 (0.010) (0.057) (12) Taiwan 0.0142*** -0.0977*** 17.71*** 0.0461 (0.003) (0.023) (13) Thailand 0.0220*** -0.0970*** 13.37*** 0.0143 (0.006) (0.031) (14) Vietnam -0.0827*** 0.8959*** 7.29*** 0.0826 (0.031) (0.192)
Low & lower-middle-income countries: Cambodia, Indonesia, Philippines, Vietnam. Upper-middle-income countries: China, Malaysia, Thailand. High-income countries: Japan, South Korea, Taiwan.
***0.01 significance level, **0.05 significance level, *0.1 significance level.
18
group grow faster while in the other two groups larger cities grow faster. In other words, agglomeration economies of large cities in upper-middle-income and high-income countries are strong enough to offset the spreading forces so that larger cities grow faster than smaller locations.
Two variables regression results for individual country is displayed in model 5 to 14. In this individual regression, it shows that among all countries, China municipalities grow proportionately and thus follow Gibrat’s law. All of the low & lower-middle-income countries, except Philippines, show negative and significant correlation between initial density and density growth. Meanwhile, the rest of the countries have positive and significant result. From these models, different city size distribution among countries can be seen, particularly among different country development levels. The analysis will be discussed in the later section.
4.2 Urban Areas
Figure 2. (a) population 2000 to population growth of urban areas 2000-2010; (b) density 2000 to density
growth of urban areas 2000-2010
Figure 2 displays the scatter plot to give overview of the correlation between initial populations to its population growth of urban areas in all countries. In contrast to the Figure 1, urban area data exhibit a negative relationship between those two variables. Although the scatter plots do not show clear negative signs, the fitted values line clearly plot a downward sloping line. Moreover, statistical test is needed to confirm the significance of the correlation between initial density and density growth which is displayed in table 2.
19
while individual country regressions are shown in model 5 to 14. Model 1 shows that the correlation between initial density and density growth is negative and significant, confirming the downward sloping pattern in the scatter plot. The hypothesis drawn in the previous section can be proven correct that larger urban areas experience stronger spreading effect than agglomeration effect. Meanwhile, smaller urban areas still have strong agglomeration effect, thus have faster growth rates. This result leads to the rejection of Gibrat’s law for aggregate urban agglomeration areas.
Table 2. Regression result of urban areas data (N=1001) with density growth (ln) as the dependent variable and
density 2000 (ln) as the independent variable
Model Density 2000 (ln) Constant F R2
(1) All Urban Areas -0.0446*** 0.3992*** 23.64*** 0.0231
(0.009) (0.078) (2) Low& lower-middle-income -0.0092 0.2851** 0.49 0.0029 (0.013) (0.116) (3) Upper-middle-income -0.0956*** 0.7804*** 87.64*** 0.1092 (0.010) (0.086) (4) High-income -0.0558*** 0.5113*** 37.52*** 0.2509 (0.009) (0.074) (5) Cambodia -0.167 1.5738 0.11 0.0994 (0.503) (4.48) (6) China -0.0802*** 0.6411*** 60.83*** 0.0818 (0.010) (0.087) (7) Indonesia -0.027 0.4622*** 2.31 0.0277 (0.018) (0.157) (8) Japan -0.004 0.1182 0.2 0.0026 (0.009) (0.072) (9) Malaysia -0.0283 0.4729 0.35 0.0191 (0.048) (0.373) (10) Philippines -0.0313* 0.4335*** 3.95* 0.0732 (0.016) (0.135) (11) South Korea -0.0228** 0.2099** 4.74** 0.1648 (0.010) (0.089) (12) Taiwan -0.109*** 0.8863** 10.46** 0.5376 (0.034) (0.288) (13) Thailand -0.0237 0.3282 0.11 0.0112 (0.071) (0.576) (14) Vietnam 0.0305 -0.0407 0.37 0.0121 (0.050) (0.447)
Low & lower-middle-income countries: Cambodia, Indonesia, Philippines, Vietnam. Upper-middle-income countries: China, Malaysia, Thailand. High-income countries: Japan, South Korea, Taiwan.
20
The regression results for the three country development levels and individual countries are displayed in model 2 to 14. Low & lower-middle-income group exhibit proportionate growth of urban areas leading to the acceptance of Gibrat’s law. Meanwhile, the other two groups show negative and significant correlation between initial density and density growth. However, when we observe per individual country, almost all of them follow Gibrat’s law except China and Taiwan that violate the law by displaying negative and significant result at 1 percent.
Furthermore, this result demonstrates clear difference of using urban areas and municipalities data, showing the importance of choosing spatial units in testing Gibrat’s law. Choosing urban areas as units of the analysis states the fact that urban areas function as a whole integrated economic unit. In contrast, using municipalities gives profound effect on population growth patterns as there are development disparities among municipalities in providing different quality of services and facilities. Gonzalez-Val et al. (2013), furthermore, examine that choosing small sample locations of largest cities mostly leads to not rejecting Gibrat’s law. Hence, the larger the sample size, the greater the evidence to validate Gibrat’s law.
4.3 Country Development Level
Table 3. ANOVA test of differences in density growth of three development level groups using both
municipalities and urban areas data.
Municipalities Urban Areas
N Mean SD N Mean SD
Low & lower-middle-income 2180 0.1717 0.1631 170 0.2045 0.0876 Upper-middle-income 1379 0.0375 0.2036 717 -0.0265 0.1591 High-income 2223 -0.0505 0.1176 114 0.0578 0.0570 Levene test W0 125.16*** 41.3*** W50 120.67*** 41.86*** W10 122.03*** 41.92***
***0.01 significance level, **0.05 significance level, *0.1 significance level.
21
and urban areas. The three indicators report the Levene’s robust test statistic (W0) for the
equality of variances between the groups and two alternative formulations that use more robust estimators when dealing with skewed populations. The first alternative (W50) replaces
the mean with the median. The second alternative (W10) replaces the mean with the 10%
trimmed mean. All of three indicators show significant differences between the groups.
According to Table 3, the ANOVA test yields significant differences among the groups, which leads us to further investigate the city size distribution of each of the country group. Since this study is using untruncated data, we can analyze the growth pattern of continuum location, from smaller locations and larger locations. This further investigation is based on the assumption of either smaller locations grow faster (Gonzalez-Val et al., 2013) or larger areas grow faster (Black and Henderson, 2003) and are offset by the slower growth of other locations that may lead to the acceptance of Gibrat’s law. Following Michaels et al. (2012)1 and Klein and Leunig (2013)2, we draw the population growth to each log initial population density in 2000 bin as shown in Figure 3.
Figure 3. Mean of population growth per population density 2000 bin in three development level groups
Figure 3 displays population growth of each population density 2000 bin. The x-axes are population density bins, defined by rounding log initial population density of each
1
Michaels et al., (2012) Panel A figure II: Employment share and regression predictions
22
municipality to every 0.5 decimal point. The y-axes shows mean of population growth of each bin. From this figure, there are apparent differences of population growth in every level of initial population density among the three groups. For low & lower-middle-income curve, a downward sloping pattern can be seen. Small locations ranging from 1 to 5 log density (5-150 people per km2) follow negative trend. In contrast, medium locations ranging from 5 to 8 log density (150-3000 people per km2) exhibits positive trend followed by another negative trend for larger locations more than 3000 people per km2.
Based on Table 1, both upper-middle-income countries and high-income countries show positive correlations between initial density and density growth. However, when looking into detail of initial population density, they show apparent pattern difference. Upper-middle-income countries trend is more similar to low & lower-Upper-middle-income countries but with more distinct reversion in trend. From the smallest locations until log density 5 (150 people per km2), there is decreasing growth as locations are getting denser. From there on, the curve turns upward and reach a peak at log density 8.5 (5000 people per km2). In large locations above 5000 people per km2, there is significant decrease as locations are getting denser.
In contrast, high-income countries display continuous positive trend from smallest locations to largest locations, or in other words, larger locations grow faster than smaller locations. Nonetheless, the growth of all locations in high-income countries is on average lower than other country groups. The reason is that a high percentage of urban population is accompanied by low population growth, and even negative growth for Japan.
23
4.4 Analysis of Gibrat’s law and Structural Transformation
In this section, analysis will be made concerning the regression results and structural transformation that occurred in each country. Structural transformation taken place in East Asia Pacific may explain why Gibrat’s law is violated in each individual country and three groups of development levels. Moreover, the disaggregated initial density of small, medium, and large locations will be analyzed underlying the policy recommendation given in the subsequent section.
More advanced countries in East Asia such as Japan first emerged as manufacturing power in the 1960s when it started exporting electronics and consumer goods, followed by South Korea and Taiwan. Starting from 1980s, Japan builds plants across Southeast Asia, and China starts to open up and becomes the biggest player until today. United Nations state that Asia accounts for 26.5 percent global manufacturing output in 1990 and increases to 46.5 percent in 2013. As China’s working-age population has peaked in 2012 and the wage keeps increasing, Southeast Asian countries offer a big labor pool with lower wages and mostly market-friendly policy environments. Moreover, the implementation of ASEAN Economic Community (AEC) starting the end of 2015 will abolish trade barriers among the ten ASEAN countries and open up labor market with burgeoning middle class. Due to this reason, Japan, South Korea, and Taiwan have been investing to move production to Southeast Asia but keep the high added value production part in their countries.
Countries in Asia exhibit diverse urban population percentage. East Asia is rapidly urbanizing with more than half population resides in urban areas. According to the World Bank, Japan and South Korea are the most urbanized countries with 90.5 percent and 81.9 percent population living in urban areas in 2010, while China’s urban population is at 49.2 percent. Southeast Asia shows a wider range of urban population percentage. Malaysia, which lies in upper-middle-income group, shows the highest percentage at 70.9 percent, followed by Indonesia, the Philippines, and Thailand with 49.9 percent, 45.3 percent, and 44.1 percent respectively. Vietnam and Cambodia, which have the lowest GNI per capita observed in this study, feature relatively low urban population percentage at 30.4 percent and 19.8 percent. This urban population percentage data shows that the advanced countries are more urbanized and indicate that less developed countries will keep urbanizing as their economies develop.
24
smaller locations in these groups exhibit dispersion. Large denser cities also show dispersion unlike the medium dense cities that grow faster. It indicates that agglomeration economies in the large cities are outweighed by diseconomies of scale as suggested by Duranton (2015). As the urbanization process is increasing, large cities may have inadequate abilities to reap agglomeration benefit.
Many mega cities have grown following the concentration of manufacturing industries and over time the services sector has come to concentrate in these cities as well High density can be considered efficient if it reduces daily commuting distance. As high-density cities like Tokyo and Seoul have good public transportations, many other Asian cities, particularly in less developed countries, lack well-functioning public transport and thus the cities are crowded out by personal vehicles (UN HABITAT, 2010). High density also promotes mixed uses of land and highly competitive land market. As a consequence, this leads to congestion.
Small and medium cities or towns typically play an important role as indirect links between the rural areas and the large cities which play part in global economy. They serve as local ‘growth centers’ or markets for rural products and urban services. Moreover, rural migrants often move to these cities temporarily before they move to the larger cities. The importance of small-medium sized cities may explain the faster growth of these locations.
Country’s spatial distribution is changing following the development and importance of sector during structural transformation as stated by Desmet and Henderson (2014) in previous section. Less developed countries in Southeast Asia are moving towards manufacturing sector while more developed countries in East Asia have moved towards service sector based on the share of the total employment. This may explain why municipalities in different country development levels show different patterns.
Cities become increasingly essential in rapid economic growth prevailing in Asian countries because they become more attractive to manufacturing and services, the concentration of which enhances productivity and growth or so-called agglomeration economies. As cities have become more integrated in the global economy, urban employment patterns undergo a structural shift. In the early phase of structural transformation, economic growth is led by the manufacturing sector, which absorbs large portions of the labor force and induces urbanization. As economy develops, service sector becomes more dominant economic activity in many Asian cities (UN HABITAT, 2010).
25
and increasing employment share in service sector from 2000 to 2010. Moreover, Japan, South Korea, Taiwan, Malaysia, and Philippines show decreasing employment share in manufacturing sector, unlike other countries3. These countries have positive correlation between initial population density and density growth, while other countries, except Thailand, have a negative correlation. Positive correlation reflects large cities grow faster than small cities, while negative correlation reflects the opposite. Hence, the city size distribution within a country may be affected by the dominance of manufacturing sector.
As the manufacturing sector becomes less dominant and service sector increases in dominance, large cities display higher population growth. Glaeser (2010) argues that service industries involve a lot more face-to-face contact than manufacturing, and thus, require to locate closer to each other in denser areas. There is a strong tendency of service industries to locate near their suppliers and customers, in contrast to weaker links between customers and suppliers in manufacturing. Transport cost is still an important determinant to firms cluster together as the costs of delivering services are much higher than the costs of delivering goods. This explains why large cities grow faster as the share of manufacturing sector decreases.
Chinese municipalities are the only one that doed not violate Gibrat’s law, or in other words, there is proportionate growth of all locations. Despite the existence of many large dense urban areas in China, there is a reason that can explain the proportionate growth of China. The Chinese government has been maintaining a strict separation of rural and urban areas, making labor mobility more difficult than other countries (Henderson et al., 2009). However, as there is proportionate growth of all locations, the benefit of agglomeration of large denser cities might be questioned. Henderson also argues that based on more developed countries experiences, there may be significant potential increase in China’s national income achievable by further increase in urbanization.
4.5 Policy Recommendations
The governments’ policy has been controlling urbanization as large cities become too large and dense. Larger cities are attractive because they offer better employment choices, access to better social services such as health and education, and higher social status. A large number of migrants remaining in the urban informal sector creates another problem that may countervail agglomeration economies occurred in large cities. People who work in informal
3
26
jobs and live in informal housings are excluded from any governments regulation and hamper the benefit of large cities through congestion and environmental degradation.
Although most of Asian countries do not impose any barriers to rural-urban migration, some have adopted mechanism to regulate this process. The most common policy implemented by the governments is reducing or even reversing the flow of rural-urban migration through a combination of rural employment creation, anti-slum drives, and restricted entry to urban areas. In China, for example, hukou system (residency permits) has been a major barrier preventing more urbanization until it was dismantled in 1978. The Vietnamese governments are granting residency permits for migrants to work in the cities, even though temporary permits are also available to ensure a steady supply of labor.
In fact, urbanization is generally associated with higher income and productivity. The urbanization process that accompanies the structural transformation process in East Asia Pacific is a key to provide economic opportunity. The agglomeration effects of cities in the form of reducing the cost of service provision and the transport of goods, allowing specialization, and enabling the flow of ideas and spillovers of knowledge between firms, mean that urbanization prompts a boost to productivity and economic growth. Hence, the World Bank (2015) advises that developing countries should increase concentration in cities.
Large dense cities should implement appropriate urban policies in order to benefit from agglomeration economies. Diseconomies arise from congestion and high land price hamper the benefit of large cities. Therefore, Hamaguchi (2008) and the World Bank (2015) recommend cities in middle-income countries in Asia to, first, mitigate diseconomies by providing infrastructure and land use regulation, and second, encourage specialization of cities in knowledge-intensive activities and shed uncompetitive activities to the periphery.
27
Large cities have an important role to perform knowledge-intensive activities such as innovation by absorbing foreign knowledge and making sure it diffuses to the rest of the country. Thus, while large cities specialize in knowledge-intensive activities, the less productive activities should be relocated to smaller cities to prevent large cities from getting too large and congested. This mostly applies to countries that have a high share of manufacturing activities. Relocating production to specialized cities can also exploit the economy in those cities by allowing technology and knowledge diffusion previously developed in denser areas and providing employment.
Regarding urban areas, the regression result of individual countries show while some countries do not reject Gibrat’s law, some other countries display negative correlation of urban size and urban growth. This indicates the danger of increased congestion and environmental degradation of large urban areas (Soo, 2007). The World Bank (2015), furthermore, points out specific policy recommendation for urban areas. It is important that regional government authorities coordinate urban service provision across municipal boundaries. Concerning land use pattern, less developed countries in Asia should facilitate access to land for future urban growth as predicted that urban expansion is still continuing in coming decades. For more developed countries, urban areas should make efficient use of the available land. As high density can actually bring more efficient use of vehicular trips, thus the development of public transportation is essential to reduce congestion and benefit from dense area.
5. Conclusion
This study’s focus is to observe Gibrat’s law on ten countries in East Asia Pacific during the structural transformation occurred at the moment, and whether particular locations grow faster than others. This study also contributes to the literature of Gibrat’s law by utilizing two different spatial units analyses: entire distribution of municipalities (untruncated data) and urban areas. The findings are as follows.
First, we find that aggregate municipalities reject Gibrat’s law and show positive correlation between population growth and initial density. Divided into three country development levels, low & lower-middle-income countries group shows negative correlation while the other two groups show positive results. Individual country regressions display that China is the only country that does not violate Gibrat’s law.
28
growth than smaller cities, the other two groups show distinct patterns between small, medium, and large locations. The smaller locations in these groups exhibit dispersion. Large denser cities also show dispersion unlike the medium dense cities that grow faster. This reflects the urbanization process occurred during current structural transformation in low & lower-middle-income countries and upper-middle-income countries groups.
Third, the share of manufacturing activities within a country may affect the city size distribution of the country. Countries that exhibit decreasing share of manufacturing show positive correlation between population growth and initial population density. This is due to the need of service sector to locate closer to each other compared to manufacturing sector. Thus, as the manufacturing sector shares decrease and are replaced by service sector, large cities grow faster.
Based on the United Nations (2010) and the World Bank (2015) reports, the governments policies have been restricting large urban areas to grow more although they suggest that urbanization in developing countries is encouraged to reap the benefit of agglomeration economies. However, public policy is needed so that the high density of large cities does not offset the agglomeration force. Mitigating the diseconomies in large cities is recommended to be done via better infrastructure and land use regulation. Moreover, large cities should focus on knowledge-intensive activities while relocating less productive activities such as manufacturing assembly to more specialized smaller locations. Furthermore, the important role of small-medium cities proposes that public policy should not focus on large cities only.
Finally, we also highlight the importance of choosing spatial unit data in testing Gibrat’s law as the different results yielded by municipalities and urban areas data. Urban areas show negative correlation between population growth and initial population density. When investigated by individual countries, only China and Taiwan violate Gibrat’s law at 1% percent significance. Nonetheless, urban areas only consist of high density areas unlike the municipalities. Thus, choosing either one of these should depend on the purpose of the study.
29 References
Amiti, M., & Cameron, L. (2007). Economic geography and wages. The Review of Economics and Statistics, 89(1), 15-29.
Black, D., & Henderson, V. (2003). Urban evolution in the USA. Journal of Economic Geography, 3(4), 343-372.
Brakman, S., Garretsen, H., & Van Marrewijk, C. (2009). The new introduction to geographical economics. Cambridge University Press.
Ciccone, A., & Hall, R. E. (1996). Productivity and the Density of Economic Activity. The American Economic Review, 86(1), 54-70.
Desmet, K., & Henderson, J. V. (2014). The geography of development within countries. Desmet, K., & Rappaport, J. (2015). The settlement of the United States, 1800–2000: The long transition towards Gibrat’s law. Journal of Urban Economics.
Desmet, K., & Rossi-Hansberg, E. (2009). Spatial growth and industry age. Journal of Economic Theory, 144(6), 2477-2502.
Desmet, K., & Rossi-Hansberg, E. (2014). Spatial development. The American Economic Review, 104(4), 1211-1243.
Duranton, G. (2015). Growing through cities in developing countries. The World Bank Duranton, G., & Puga, D. (2000). Diversity and specialisation in cities: why, where and when does it matter?. Urban studies, 37(3), 533-555.
Eeckhout, J. (2004). Gibrat's law for (all) cities. American Economic Review, 1429-1451. Fujita, M., & Ishii, R. (1998). Global location behavior and organizational dynamics of Japanese electronics firms and their impact on regional economies. The dynamic firm: The role of technology, strategy, organization and regions, 343-383.
Gabaix, X. (1999). Zipf's Law and the Growth of Cities. The American Economic Review, 89(2), 129-132.
Glaeser, E. L. (Ed.). (2010). Agglomeration economics. University of Chicago Press. Gollin, D., Jedwab, R., & Vollrath, D. (2013). Urbanization with and without structural transformation. mimeograph. George Washington University, 1(1.1).
González-Val, R., Lanaspa, L., & Sanz-Gracia, F. (2013). Gibrat’s law for cities, growth regressions and sample size. Economics Letters, 118(2), 367-369.
Hamaguchi, N. (2008). Regional Income Inequalities in East Asia. In Koji, N. Economic in East Asia Economic Integration.
Henderson, J. V. (1974). The sizes and types of cities. The American Economic Review, 64(4), 640-656.
Henderson, J. V., Quigley, J., & Lim, E. (2009). Urbanization in China: Policy issues and options. Unpublished manuscript, Brown University.
Henderson, V., Lee, T., & Lee, Y. J. (2001). Scale externalities in Korea. Journal of Urban Economics, 49(3), 479-504.
30
Kooij, P. (1988). Peripheral cities and their regions in the Dutch urban system until 1900. The
Journal of Economic History, 48(2).
Michaels, G., Rauch, F., & Redding, S. J. (2012). Urbanization and Structural Transformation. The Quarterly Journal of Economics, 127(2), 535-586.
Ottaviano, G., & Thisse, J. F. (2004). Agglomeration and economic geography. Handbook of regional and urban economics, 4, 2563-2608.
Portnov, B. A., Reiser, B., & Schwartz, M. (2012). Does Gibrat’s law for cities hold when location counts?. The Annals of Regional Science, 48(1), 151-178.
Resende, M. (2004). Gibrat's law and the growth of cities in Brazil: A panel data investigation. Urban Studies, 41(8), 1537-1549.
Salway, R., & Wakefield, J. (2005). Sources of bias in ecological studiesof non-rare events.
Environmental and Ecological Statistics, 12(3), 321-347.
Schmutzler, A. (1999). The new economic geography. Journal of Economic Surveys, 13(4), 355-379.
Soo, K. T. (2007). Zipf's Law and urban growth in Malaysia. Urban Studies, 44(1), 1-14. United Nations Human Settlements Programme (UN HABITAT). (2010). The State of Asian Cities 2010/11. Fukuoka: UN HABITAT.
Vaturi, A., Portnov, B. A., & Schwartz, M. (2004). Internal migration and fiscal capacity of local authorities: a case study of the greater Tel Aviv metropolitan area, 1985–97. Space and
Polity, 8(1), 61-85.
World Bank. (2009). World Development Report: Reshaping Economic Geography. Washington DC: World Bank
World Bank. (2015). East Asia’s Changing Urban Landscape: Measuring a Decade of Spatial Growth. Urban Development Series. Washington DC: World Bank
Appendix 1
Economies are currently divided into four income groupings: low, lower-middle, upper-middle, and high. Income is measured using gross national income (GNI-formerly referred to as GNP) per capita, in US dollars, converted form local currency using World Bank Atlas method. The purpose of using the Atlas conversion factor instead of simple exchange rates conversion is to reduce the impact of exchange rate fluctuations in the cross-country comparison of national income.
31 Appendix 2
Table 4. Regression result of municipalities data (N=5782) with population growth (ln) as the dependent
variable and population 2000 (ln) as the independent variable
Model Population 2000 (ln) Constant F R2 (1) All Municipalities 0.0191*** -0.1524*** 153.38*** 0.0259 (0.002) (0.017)
(2) Low& lower-middle income 0.0012154 0.1587*** 0.18 0.0001
(0.003) (0.031) (3) Upper-middle income 0.0124*** -0.1093*** 17.75*** 0.0127 (0.03) (0.035) (4) High-income 0.0363*** -0.4246*** 491.15*** 0.1811 (0.002) (0.017) (5) Cambodia -0.1225*** 1.4991*** 51.33*** 0.2298 (0.017) (0.184) (6) China -0.0145*** 0.2804*** 4.97* 0.0152 (0.006) (0.097) (7) Indonesia -0.0857*** 1.2653*** 136.28*** 0.2929 (0.007) (0.091) (8) Japan 0.0376*** -0.4438*** 672.89*** 0.2848 (0.001) (0.015) (9) Malaysia 0.0261** -0.1483326 5.59** 0.0412 (0.011) (0.128) (10) Philippines 0.0214*** -0.0684* 34.89*** 0.0215 (0.004) (0.037) (11) South Korea 0.0919*** -1.1028*** 40.88*** 0.2025 (0.014) (0.168) (12) Taiwan 0.0136** -0.1484*** 6.58** 0.0176 (0.005) (0.056) (13) Thailand 0.0297*** -0.3106*** 8.08*** 0.0087 (0.010) (0.114) (14) Vietnam -0.0562** 1.0073*** 3.96** 0.0466 (0.028) (0.314)
Low & lower-middle-income countries: Cambodia, Indonesia, Philippines, Vietnam. Upper-middle-income countries: China, Malaysia, Thailand. High-income countries: Japan, South Korea, Taiwan.
32
Table 5. Regression result of urban areas data (N=5782) with population growth (ln) as the dependent variable
and population 2000 (ln) as the independent variable
Model Density 2000 (ln) Constant F R2
(1) All Urban Areas -0.0158*** 0.4691*** 7.83*** 0.0078
(0.006) (0.070)
(2) Low& lower-middle income -0.0214*** 0.6388*** 7.44*** 0.0424
(0.008) (0.096) (3) Upper-middle income -0.0091 0.3826*** 1.29*** 0.0018 (0.008) (0.010) (4) High-income -0.0097*** 0.2596*** 7.74*** 0.0647 (0.003) (0.043) (5) Cambodia -0.1226 2.080 1.98 0.6643 (0.087) (0.983) (6) China -0.0087 0.3733*** 1.11 0.0016 (0.008) (0.104) (7) Indonesia -0.0497*** 1.0048*** 9.69*** 0.1068 (0.016) (0.206) (8) Japan 0.0013 0.1221*** 0.25 0.0034 (0.003) (0.031) (9) Malaysia -0.0014 0.4078*** 0.03 0.0017 (0.008) (0.100) (10) Philippines -0.0026 0.4065*** 0.04 0.0008 (0.013) (0.141) (11) South Korea -0.0209** 0.4295*** 6.18** 0.2049 (0.008) (0.103) (12) Taiwan 0.0033 0.0266 0.09 0.0104 (0.011) (0.144) (13) Thailand -0.0389 0.8238 1.03 0.0931 (0.038) (0.465) (14) Vietnam 0.0065 0.3056** 0.41 0.0136 (0.010) (0.123)
Low & lower-middle-income countries: Cambodia, Indonesia, Philippines, Vietnam. Upper-middle-income countries: China, Malaysia, Thailand. High-income countries: Japan, South Korea, Taiwan.
