Professional Documents
Culture Documents
com
ScienceDirect
Explorations in Economic History 60 (2016) 52 – 68
www.elsevier.com/locate/eeh
Abstract
The Great Kantō Earthquake in 1923 did enormous damage to industries in Tokyo Prefecture. Around 40% of the buildings in
Tokyo Prefecture were completely burnt or destroyed. In this paper, we investigate whether this temporary shock had a persistent
impact on the spatial distribution of industries in Tokyo, using ward- and county-level panel data for industrial workers. It was
found that while the earthquake caused mean shifts in the shares and numbers of workers, these mean shifts disappeared by the early
1930s. On the other hand, the earthquake caused shifts in the trends in the share and number of workers. The combined effects of
these mean shifts and trend shifts were persistent for both the shares and the numbers of workers. The earthquake caused especially
serious damage to the old industrial clusters in the southeast of Tokyo, and provided an opportunity for newly developing industrial
clusters in non-damaged areas to take over the market. Further, the people and the local governments in non-damaged areas made an
effort to take advantage of this opportunity to attract factories. Arguably, these forces made the impact of the earthquake on the
spatial distribution of industry persistent.
© 2015 Elsevier Inc. All rights reserved.
Keywords: Industrial agglomeration; Economic geography; Natural disaster; Economic history; Japan
JEL classification: R11; R12; N95
http://dx.doi.org/10.1016/j.eeh.2015.11.005
0014-4983/© 2015 Elsevier Inc. All rights reserved.
A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68 53
basis for the possibility of path dependence (Nunn In the context of Japanese economic history, the period
2014). These models include those in spatial economics that includes the Great Kantō Earthquake was the period
(Krugman 1991; Fujita et al. 1999). It is remarkable that when the spatial distribution of industries in the Tokyo
formal empirical tests on multiple equilibria and path metropolitan area changed substantially, as shown below.
dependence have been conducted on spatial issues. One In this period, new industrial clusters developed rapidly in
of the reasons for this is that there are appropriate some counties, including Ebara County, which came to be
opportunities for natural experiments. Those studies the largest cluster of the machinery industry in Japan.
include Davis and Weinstein (2002, 2008), Brakman Investigating the impact of the Great Kantō Earthquake
et al. (2004), Miguel and Roland (2006), Bosker et al. on the spatial distribution of industries contributes to
(2007), Redding and Strum (2008), and Redding et al. understanding the spatial aspects of the development
(2011). The results of these studies on multiple of the Japanese economy (Whittaker 1997, pp. 63–6;
equilibria and path dependence are mixed. That is, Kantō Bureau of International Trade and Industry 1996,
whereas Davis and Weinstein (2002, 2008), Brakman pp. 137–8).
et al. (2004), and Miguel and Roland (2006) did not The remainder of the paper is organized as follows.
find evidence supporting the persistence of temporary Section 2 describes the damage caused by the Great
shocks, Bosker et al. (2007), Redding and Strum Kantō Earthquake, and the subsequent reconstruction
(2008), and Redding, et al. (2011) found supportive plan. Section 3 describes the changes seen over time in
evidence. the spatial of industries in Tokyo Prefecture and
This paper aims to contribute to this strand of literature econometrically analyzes the long-run impact of the
in two ways. First, the focus is on the effects of a earthquake. Section 4 discusses the descriptive and
temporary shock caused by a natural disaster rather than anecdotal evidence on the impact of the earthquake.
by a war or a political event. The damage caused by the Section 5 concludes the paper.
earthquake was exogenous, and, as shown below, the
shock provided by the Great Kantō Earthquake was
extremely large. Second, ward- and county-level data 2. The Great Kantō Earthquake and reconstruction
from Tokyo Prefecture are used. Previous studies that
focused on temporary shocks rejected the possibility of a At 11:58 am on September 1, 1923, a huge
long-run impact of a temporary shock, based on nation- earthquake with a magnitude of M 7.9 hit the southern
wide, city-level, or district-level data. However, even if area of the Kantō district in Japan. This earthquake, the
the nationwide spatial distribution of the population and Great Kantō Earthquake, was the worst natural disaster
economic activities is determined by the fundamental in the history of Japan. The damage extended over
conditions and, hence, a temporary shock has no long-run seven prefectures, Tokyo, Kanagawa, Saitama, Chiba,
impact, it is possible that a temporary shock has a Ibaraki, Shizuoka, and Yamanashi. The total number of
long-run impact within a more limited area. In such a deaths and missing persons was more than 100,000,2
case, the variations in fundamental conditions, such as which was 16.3 times larger than the losses from the
proximity to the coast and river, are supposed to be 1995 Kobe Earthquake and 5.4 times larger than those
smaller; hence, the existing equilibrium of spatial dis- from the 2011 Great East Japan Earthquake.3 In
tribution of economic activities within the area is less addition, more than 460,000 buildings were completely
stable. In addition, the case of the Great Kantō burnt or destroyed.4 The human and physical damage
Earthquake is useful in identifying the shift in equilibri- was concentrated in the prefectures of Tokyo and
um, because not only did those fundamental natural Kanagawa (see Table 1 and Figs. 1 and 2). The main
conditions not change due to the earthquake, but the reason why the damage was so serious was the
physical infrastructure, including public transportation, 2
also recovered swiftly after the earthquake.1 The number of deaths and missing persons include those who died
from the earthquake itself and those who died from the fire caused by
the earthquake.
3
The losses from the 1995 Kobe Earthquake and the 2011 Great
1
Railways were seriously damaged by the earthquake, but by East Japan Earthquake were 6,437 and 19,386, respectively (see the
October 1923, most of the major railway lines had been reconstructed Web page of the Fire and Disaster Management Agency, http://www.
by the activities of the engineer brigade (Social Affairs Bureau, fdma.go.jp/data/010604191452374961.pdf, and the Web page of the
Ministry of Home Affairs, 1926, pp. 439–441). Yokohama Port, the National Police Agency, http://www.npa.go.jp/archive/keibi/biki/
largest international port in Japan at the time, had also been higaijokyo.pdf).
reconstructed by 1925 (Society for Research on the Recovery, 1930, 4
“Destroyed” means “directly destroyed by the quake,” while
p. 2123). “burnt” means “burnt by the fire caused by the earthquake.”
54 A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68
the Bureau of Reconstruction, which was newly The city plan included zoning regulations.11 The
established in the Ministry of Home Affairs. The City Area Architecture Law prescribed three categories
Bureau of Reconstruction implemented the reconstruc- of zones: the residential zone, the commercial zone, and
tion plan, including land readjustment, construction of the industrial zone. With respect to factories, the
streets, and layout of parks (Tokyo City Government, following regulations were imposed in these zones.
1932, p. 169). The regulations were revised in 1926, 1929 and 1935.
A distinctive feature of the reconstruction plan was The regulations after 1929 revision were as follows. In a
that it was drawn up as part of the city plan. The City residential zone, factories using motors over 3 horse-
Planning Law and the City Area Architecture Law power (hp) were prohibited, whereas in a commercial
had already been legislated in 1919 (Ishida, 1987, zone, those using motors over 15 hp were prohibited. In
pp. 114–43; Koshizawa, 1991, pp. 13–23), and, in addition, a factory was prohibited in both residential
accordance with these two laws, the government had and commercial zones if it was dangerous or caused
prepared a city plan for Tokyo before the earthquake, inconvenience, such as where there was a risk of an
but it was not implemented. This city plan was then
incorporated in the reconstruction plan (Takebe, 1925, 11
This was the first time that zoning was implemented in Tokyo
pp. 44–45; Tokyo City Government, 1925, p. 94). City.
56 A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68
Fig. 2. Title: Damage to buildings in 7 prefectures. Source: Tokyo City Government (1925), p. 161.
explosion, discharge of harmful or pungent gases or zones, there was also a “zone not yet specified.” In this
liquids, or where significant amounts of dust, noise or zone, a factory using motors of 50 hp or less was
vibrations were generated. Factories using motors over allowed (Yano, 1932, pp. 40–172; Horiuchi 1978,
50 hp or that presented one of the above hazards were pp. 56–70; Numajiri 2002, p. 64). The zoning as of
only allowed in the industrial zone. Besides these three 1936 is shown in Fig. 4.
Fig. 3. Title: Damage to buildings by ward and county in Tokyo Prefecture. Source: Social Affairs Bureau of the Ministry of Home Affairs (1924);
Statistical Yearbook of Tokyo Prefecture, 1923 issue.
A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68 57
Table 2
Magnitude of physical damage: Comparison with recent earthquakes.
Event year Total damage GDP (GNP) in previous years Percentage to
(2010 price, million dollars) (2010 price, million dollars) GDP (GNP)
Great Kantō Earthquake 1923 78,800 221,894 35.5
Kobe Earthquake 1995 93,661 4,765,497 2.0
Great East Japan Earthquake 2011 192,549 5,459,451 3.5
Source: Tokyo City Government (1925) p. 161; the webpage of Kobe News Paper (http://www.kobe-np.co.jp/sinsai/kiroku/higai0012.html); the
webpage of the Cabinet Office (http://www.bousai.go.jp/oshirase/h23/110624-1kisya.pdf).
Note: To convert the 1923 values to 1995 values, we used a series of deflators, linking the GNP (GDP) deflators of Ohkawa et al.(1974), the
Economic Planning Agency (1965, 2001).
It is notable that the regulations were exempted to changes over time in the distribution of industries and
existing factories unless they were extended or recon- indicates the shares of industrial workers in the 15
structed. In addition, enforcement of the regulations on wards of Tokyo City compared with the total numbers
extension and reconstruction of existing factories was of industrial workers in Tokyo Prefecture, and for the
suspended until 1936 (Imaizumi 2010, pp. 35–7). Hence, five neighboring counties, which were merged into
until 1936, zoning regulations were imposed only on Tokyo City in 1932.
construction of new factories. In this sense, zoning First, it can be seen just how large the temporary
regulations were not so strict. This implies that there was shock of the Great Kantō Earthquake was, in that the
room for economic mechanisms to work in determining total number of workers declined by 35% from 1922 to
spatial distribution of industries, but in evaluating the 1923. Second, the shock varied substantially across
long-run impact of an earthquake on industrial growth, the wards and counties. Table 3 arranges the wards and
effects of zoning should be taken into account. counties (areas, hereafter) by the ratio of completely
burnt or destroyed buildings, and divides the areas into
3. Evaluating the impact of the earthquake on the two groups, “damaged” and “non-damaged.” The data
spatial distribution of industries on the damage to buildings are taken from Tokyo City
Government (1925). The ratios of completely burnt or
We now focus on the impact of the earthquake on destroyed buildings were higher in the southeast
industries within Tokyo Prefecture. Table 3 details the downtown areas Nihonbashi, Asakusa, Fukagawa,
Fig. 4. Title: Zoning in Tokyo Prefecture (1936). Source: City Planning Office, Bureau of Audit, Tokyo City Government Tokyo Toshi-keikaku
Chiiki-zu (City Planning Zoning Map of Tokyo), 1936.
58 A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68
Table 3
Earthquake damage and change in the spatial distribution of industry.
Ratio of completely burnt Number of workers Worker share (%)
or destroyed buildings
1922 1923 1936 1922 1923 1936
Total 0.203 183,521 119,012 376,718 100.00 100.00 100.00
Damaged Average 0.756 8,581 2,636 11,525 4.68 2.21 3.06
Nihonbashi Ward 0.932 2,075 552 3,105 1.13 0.46 0.82
Asakusa Ward 0.930 3,471 866 9,486 1.89 0.73 2.52
Fukagawa Ward 0.927 13,525 2,176 12,670 7.37 1.83 3.36
Honjo Ward 0.923 23,206 7,613 31,582 12.64 6.40 8.38
Kyōbashi Ward 0.919 13,914 2,154 12,810 7.58 1.81 3.40
Kanda Ward 0.894 5,984 1,435 7,676 3.26 1.21 2.04
Shitaya Ward 0.744 3,227 1,827 6,564 1.76 1.54 1.74
iKōjimachi Ward 0.589 2,335 1,671 4,008 1.27 1.40 1.06
Shiba Ward 0.444 15,684 6,456 23,955 8.55 5.42 6.36
Hongō Ward 0.262 2,388 1,611 3,397 1.30 1.35 0.90
Non-damaged Average 0.077 9,771 9,265 26,147 5.32 7.79 6.94
Akasaka Ward 0.189 421 568 844 0.23 0.48 0.22
Minami Katsushika County 0.146 36,608 32,827 72,418 19.95 27.58 19.22
Kitatoshima County 0.131 25,205 24,085 59,885 13.73 20.24 15.90
Minami Adachi County 0.096 1,693 1,610 10,778 0.92 1.35 2.86
Koishikawa Ward 0.051 6,300 6,835 6,641 3.43 5.74 1.76
Yotsuya Ward 0.048 675 834 934 0.37 0.70 0.25
Azabu Ward 0.047 2,567 2,486 4,019 1.40 2.09 1.07
Ebara County 0.027 15,284 15,205 83,059 8.33 12.78 22.05
Ushigome Ward 0.020 2,838 3,216 4,668 1.55 2.70 1.24
Toyotama County 0.011 6,121 4,985 18,219 3.34 4.19 4.84
Source: Tokyo City Government, Statistical Yearbook of Tokyo City, 1924 issue and 1925 issue; Tokyo City Government, Statistical Yearbook of
Industries in Tokyo City, 1938 issue.
Note: The areas in bold indicate that the shares of the areas declined from 1922 to 1923 and the areas did not recover half of the share decline by
1936.
Honjo, Kyōbashi, and Kanda. Among them, four areas, impact of the earthquake on industrial development in
Fukagawa, Honjo, Kyōbashi, and Kanda, were the areas Tokyo Prefecture. Classifications of the damaged and
where industries were most densely agglomerated non-damaged areas are the same as in Table 3. The
before the earthquake. However, they lost more than numbers of industrial workers in the damaged areas
50% of their shares of industrial workers. declined sharply in 1923, whereas even though the
On the other hand, some counties were not seriously number of workers gradually recovered, even by the
damaged, and, furthermore, increased their shares of middle of the 1930s, it was still around the same level
industrial workers. In this sense, the earthquake changed as it was in 1922. On the other hand, in the
the spatial distribution of industries in Tokyo Prefecture, non-damaged areas, the number of workers declined
at least temporarily. Finally, it is notable that at least some just slightly in 1923, and then the trend of steady
of this change seems to have been persistent. In the four growth continued.
wards that were seriously damaged, more than half of the These observations seem to suggest that the Great
decline in the shares of industrial workers from 1922 to Kantō Earthquake had a persistent impact on the spatial
1923 persisted until 1936, even though the Japanese distribution of industries within Tokyo Prefecture.
economy had passed through the Great Depression and However, we should exercise caution because the
the rapid growth of the 1930s.12 Fig. 5 illustrates the observations in Table 3 and Fig. 5 reflect the impact
of other factors in addition to the earthquake, including
the long-term trend in shares of workers in all areas. In
fact, the areas that were seriously damaged by the
12
Concerning the macroeconomy in Japan in this period, see earthquake had several attributes that were different
Nakamura (1983) and Cha (2003). from those of the areas that were not so seriously
A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68 59
Fig. 5. Title: Growth of industrial workers in damaged and non-damaged areas. Source: See Table 3.
damaged. This can be seen in Table 4, which compares Statistical Yearbook of Tokyo City (Tokyo-shi Tōkei
the attributes of damaged and non-damaged areas. Hyō). Because the 1915 issue of this yearbook does not
While the average plant density, i.e. the number of contain the data on industrial workers by county, we
plants per square kilometer, in the damaged areas was used data taken from the 1914 issue of Manufacturing
151.6 in 1922, that in the non-damaged areas was 42.4 Statistics of Tokyo Prefecture (Tokyo-fu Kōjō Tōkei)
in the same year. At the same time, while the average for 1915. The data for 1929–36 were obtained from
worker density, i.e. the number of industrial workers per Statistical Yearbook of Tokyo Prefecture (Tokyo-fu
square kilometer, in the damaged areas was 1443, it was Tōkei Sho). The data for 1925–28 were not available.
330 in the non-damaged areas. Finally, the growth rates These data cover all the factories employing five or
of the numbers of plants and the numbers of workers more workers, and the factories employing less than
were larger in the non-damaged areas in the period five workers but using motors.
before the earthquake (1914–1922). In sum, the The baseline equations to be estimated are:
damaged areas were relatively established industrial
areas, but the non-damaged areas were newly industri- WSHAREit ¼ α þ β t NBROKENi YEARt AFTERt
alizing areas. This is important to note when evaluating þ γNBROKENi TIMEAFTERt ð1Þ
the persistence of the impact of the earthquake, because
þ δ1 AREIi þ δ2 ARECi þ ζi þ εit
at least a part of the changes in the spatial distribution of
industries can be seen to reflect the difference in the
LNWORKERit ¼ α þ β t NBROKENi YEARt AFTERt
long-run industrialization trends between the damaged
areas and the non-damaged areas. þ γNBROKENi TIMEAFTERt
ð2Þ
We examine the persistence of the impact of the þ ηTIMEt þ δ1 AREIi þ δ2 ARECi
earthquake by regression analyses based on the ward- þ ζi þ εit
and county-level panel data on industrial workers in
Tokyo Prefecture. The data points are for the 13 years With respect to Eq. (1), the dependent variable,
1915, 1921, 1922, 1923, 1924, 1929, 1930, 1931, 1932, WSHAREit, is the share of industrial workers in area i
1933, 1934, 1935, and 1936, and the number of areas is in year t; NBROKENi is the ratio of completely
20 (15 wards and five counties). Thus, we have 260 destroyed or burnt buildings in area i, normalized to
(13 × 20) area–year observations. The data for the have a mean of zero and a standard deviation of one;
years 1921–24 were obtained from various issues of the YEARt is the year dummy; and AFTERt is the dummy
60 A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68
Table 4
Comparison of basic attributes of “damaged” and “non-damaged” areas.
Area Number Plant density, 1922 Worker density, 1922 Annual growth rate Annual growth rate
(km2) of plants, (number of plants (number of industrial of plant number, of worker number,
1922 per km2) workers per km2) 1914–22 (%) 1914–22 (%)
Damaged Average 5.8 808 151.6 1,443 16.4 6.7
Nihonbashi Ward 3.1 280 89.9 666 6.8 0.2
Asakusa Ward 5.3 948 179.7 658 16.5 4.0
Fukagawa Ward 8.2 1,226 148.9 1,643 20.7 5.6
Honjo Ward 6.5 1,250 192.5 3,575 7.7 5.8
Kyōbashi Ward 5.1 1,015 198.9 2,726 16.0 6.0
Kanda Ward 3.1 1,098 354.2 1,930 20.1 10.7
Shitaya Ward 5.0 511 101.3 640 12.7 4.1
Kojimachi Ward 8.4 174 20.8 279 28.8 12.9
Shiba Ward 8.6 1,066 123.9 1,823 18.7 9.2
Hongō Ward 4.9 514 105.5 490 16.5 8.2
Non-damaged Average 49.2 372 42.4 330 25.7 10.8
Akasaka Ward 4.3 168 39.1 98 34.2 13.9
Minami Katushika County 100.5 499 5.0 364 13.1 10.3
Kitatoshima County 125.5 534 4.3 201 13.8 8.7
Minami Adachi County 53.5 168 3.1 32 22.6 2.7
Koishikawa Ward 6.1 709 117.0 1,040 32.6 9.1
Yotsuya Ward 3.2 176 54.4 208 39.9 13.1
Azabu Ward 4.3 465 108.5 599 37.7 10.7
Ebara County 114.7 377 3.3 133 11.5 11.8
Ushigome Ward 5.2 454 87.1 545 38.3 9.0
Toyotama County 74.8 165 2.2 82 13.2 18.6
Source: Tokyo City Government, Statistical Yearbook of Tokyo City, 1922 issue; Tokyo Prefecture Government (1916).
variable, which equals 1 if the year is 1923 or later. well as the trend shift of the share of workers, caused by
Thus, the interaction term NBROKENi YEARt AFTE the earthquake.
Rt represents the mean shift of worker share of area i With respect to Eq. (2), the dependent variable,
in year t caused by the earthquake. By interacting LNWORKERit, is the log of the number of industrial
NBROKENi with YEARt AFTERt , we can evaluate workers in area i in year t. As an explanatory variable in
the change over time in the mean shift of the share of Eq. (2), we add the time trend, TIMEt, to control for the
workers. Another measure of persistence of the earth- overall upward trend in the number of industrial
quake impact is NBROKENi TIMEAFTERt , where workers in Tokyo Prefecture.
TIMEAFTERt is the time trend at the time of and after The summary statistics are reported in Table 5 and
the earthquake (0 for 1923 and 13 for 1936). This the results are reported in Table 6. Column (1) in
interaction term represents the shift in the trend of the Table 6 indicates that the earthquake had a substantial
share of workers caused by the earthquake. The variable negative impact on the shares of workers in various
AREIi is the ratio of the industrial zone to the total area areas. That is, a one standard deviation increase in
in area i, whereas the variable ARECi is the ratio of the the ratio of completely burnt or destroyed buildings
commercial area to the total area in area i.13 These reduced the share of workers by 2.79% points in 1923.
variables are introduced to control for the effects of the At the same time, it is remarkable that the mean shift
zoning regulation. We also introduce an area fixed captured by the variable NBROKENi YEARt AFTE
effect, ζi, to capture the area-specific worker shares. The Rt was temporal. For 1924, the coefficient is 1.35%,
last term, εit, is the random error. By this specification, and for 1929 and the following years the coefficient
we intend to evaluate the time-varying mean shift as was not significantly different from zero. On the
other hand, there was a significant negative trend shift
after the earthquake. That is, the coefficient of
NBROKENi TIMEAFTERt is negative and statistical-
13
For the years before 1925, when the zoning was implemented, ly significant. This result implies that a one standard
AREAI and AREAC equal to zero. deviation increase in the ratio of completely burnt or
A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68 61
Table 5
Basic statistics.
Obs. Mean Stdev. Min. Max.
WSHARE 260 0.050 0.006 0.001 0.277
LEORKER 260 8.455 1.254 4.860 11.228
NBROKEN 260 0.000 1.000 − 1.066 1.357
NBROKEN*TIMEAFTER 260 0.000 7.671 − 13.864 17.639
NPLANTDENSITY22 260 0.000 1.000 − 1.087 2.951
NPLANTDENSITY22*TIME 260 0.268 14.736 − 23.922 64.979
AREAI 260 0.010 0.226 0.000 0.826
AREAC 260 0.152 0.286 0.000 1.000
destroyed buildings would reduce the slope of the share explanatory variable NPLANTDENSITY1922i TIMEt ,
of workers by 0.18% points. Thus, the earthquake had a where NPLANTDENSITY1922i is the plant density in the
persistent negative impact on the share of workers in the area in 1922, normalized to have a mean of zero and a
sense that it caused a negative change in the trend. standard deviation of one. Thus, the interaction term of
A concern regarding this interpretation is that because plant density with the time variable represents the trend in
the seriously damaged areas were relatively industrialized the share of workers that was dependent on the plant
before the earthquake, the trend shift in the estimation density just before the earthquake. As shown in column (2)
result for Eq. (1) reflects the saturation of industrialization of Table 6, the coefficient of this variable is negative and
in those areas. To address this concern, we added the statistically significant, which implies that those areas that
Table 6
Persistence of the impact of the earthquake: Industry total.
(1) (2) (3) (4)
Dependent variable WSHARE WSHARE LNWORKER LNWORKER
NBROKEN*1923 − 0.0279 (− 7.18) ⁎⁎⁎ − 0.0260 (− 6.82) ⁎⁎⁎ − 0.6616 (− 10.09) ⁎⁎⁎ − 0.6175 (9.90) ⁎⁎⁎
NBROKEN*1924 − 0.0135 (− 3.54) ⁎⁎⁎ − 0.0118 (− 3.14) ⁎⁎⁎ − 0.2787 (− 4.31) ⁎⁎⁎ − 0.24043 (− 3.92) ⁎⁎⁎
NBROKEN*1929 − 0.0030 (− 0.76) − 0.0021 (− 0.54) − 0.1896 (− 2.62) ⁎⁎⁎ − 0.14867 (− 2.16) ⁎⁎
NBROKEN*1930 − 0.0004 (− 0.09) − 0.0004 (0.11) − 0.1453 (− 2.03) ⁎⁎ − 0.11019 (− 1.62)
NBROKEN*1931 0.0016 (0.39) − 0.0023 (0.57) − 0.0961 (− 1.34) − 0.06681 (− 0.99)
NBROKEN*1932 − 0.0008 (− 0.20) − 0.0003 (− 0.07) − 0.0858 (− 1.19) − 0.06259 (− 0.92)
NBROKEN*1933 − 0.0005 (− 0.11) − 0.0001 (− 0.02) − 0.0820 (− 1.12) − 0.06448 (− 0.93)
NBROKEN*1934 − 0.0002 (− 0.05) − 0.0001 (0.01) − 0.0587 (− 0.78) − 0.04696 (− 0.66)
NBROKEN*1935 0.0000 (− 0.00) 0.0001 (0.03) − 0.0231 (− 0.30) − 0.0173 (− 0.24)
NBROKEN*TIMEAFTER − 0.0018 (− 5.71) ⁎⁎⁎ − 0.0012 (− 3.29) ⁎⁎⁎ − 0.0345 (− 5.06) ⁎⁎⁎ − 0.0166 (− 2.29) ⁎⁎
TIME 0.0385 (6.96) ⁎⁎⁎
NPLANTDENSITY1922*TIME − 0.0008 (− 3.78) ⁎⁎⁎ − 0.0175 (− 5.27) ⁎⁎⁎
AREAI − 0.0065 (− 1.06) − 0.0110 (− 1.80) ⁎ 0.5519 (2.99) ⁎⁎⁎ 0.3456 (1.94) ⁎
AREAC − 0.0092 (1.60) 0.0108 (1.91) ⁎ 0.7503 (3.40) ⁎⁎⁎ 0.6519 (3.12) ⁎⁎⁎
Constant 0.0049 (34.30) ⁎⁎⁎ 0.0493 (35.29) ⁎⁎⁎ 7.81495 (136.27) ⁎⁎⁎ 7.81961 (146.30) ⁎⁎⁎
Ward fixed effect Yes Yes Yes Yes
Year fixed effect No No Yes Yes
R-sq within 0.2696 0.3128 0.6987 0.7332
Between 0.0056 0.0236 0.0071 0.0011
Overall 0.0200 0.0423 0.0782 0.0563
Obs. 260 260 260 260
Note: t-values are in parentheses.
⁎⁎⁎ Statistically significant at 1% level.
⁎⁎ Statistically significant at 5% level.
⁎ Statistically significant at 10% level.
62 A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68
had already been industrialized had a lower trend in terms earthquake. For 1923, the coefficient is − 0.6616,
of the share of workers. In addition, we confirm that the which implies that a one standard deviation increase in
coefficient of NBROKENi TIMEAFTERt is negative and the ratio of completely burnt or destroyed buildings
statistically significant, although the magnitude decreased. would cause a 48.4% reduction in the number of
Column (3) of Table 6 reports the estimation workers. Meanwhile, the magnitude of the negative
result for Eq. (2). The results are similar to those for mean shifts decreased over time, whereas they were
Eq. (1). We observe a large negative mean shift in relatively persistent in terms of share of workers. A
LNWORKER caused by the damage from the significant negative mean shift continued until 1930. On
Fig. 6. Title: Decomposition of the impact caused by one standard deviation of the earthquake. Source: See the text.
A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68 63
Fig. 6 (continued).
the other hand, the coefficient of NBROKENi is conducted in Panels A–C of Fig. 7 in relation to
TIMEAFTERt is negative and statistically significant. LNWORKER. In both Fig. 6 and Fig. 7, the downward
That is, as in the case of share of workers, the damage by trend shifts made the combined effects persistently
the earthquake caused a downward trend shift in the negative, whereas the mean shifts almost disappeared
number of workers. In column (4), we added by the early 1930s.
NPLANTDENSITY1922i TIMEt as an explanatory The worker data at the ward and county levels are
variable to control for the trend in the numbers of workers also available by industry from the sources we used
that were dependent on the plant density just before the for the analyses above. Industries are classified into
earthquake. In this case, the coefficient of NBROKENi five categories, namely textiles, metals and machinery,
TIMEAFTERt is also negative and statistically significant. chemicals and ceramics, foods, and miscellaneous. Using
Thus, we conclude that the impact of the earthquake on these data, we estimated the same regressions by industry.
the number of workers was persistent in the sense that it That is, we estimated Eqs. (1) and (2) adding the variable
caused a downward trend shift. Concerning the zoning NPLANTDENSITY1922i TIMEt . The results are re-
variables, AREAI and AREAC, in the worker share ported in Panels A and B of Table 7. With respect to the
regressions their effects are not so clear, but in the worker regressions with WSHARE as the dependent variable
number regressions they have significantly positive (Panel A), the magnitude and persistence of the mean
effects. This implies that those areas larger portion of shift are similar to those for the industry total. The
which was designated as industrial area or commercial immediate negative impact was 2–3% points in 1923, but
area tended to have more industrial workers, and in this it diminished by about half in 1924, and the significant
sense the zoning regulation in fact affected the spatial negative mean shift had disappeared by 1929. On the other
distribution of industries. hand, the results for the trend shift differ across industries.
Figs. 6 and 7 depict the results of columns (2) and (4) While the coefficients of NBROKENi TIMEAFTERt are
of Table 6. Panel A of Fig. 6 shows the mean shift in negative for all of the industries, they are statistically
share of workers caused by a one standard deviation significant in only three industries, namely metal and
increase in the ratio of completely burnt or destroyed machinery, chemical and ceramics, and miscellaneous.
buildings, and Panel B of Fig. 6 shows the trend shift in With respect to the regressions with LNWORKER as
the share of workers. Combining the graphs in Panels A the dependent variable (Panel B), it is found that the
and B produces the graph in Panel C. The same exercise persistence of the mean shift is basically shorter than the
64 A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68
result for the industry total, and that significant negative numbers of workers caused by the Great Kantō
trend shifts are observed for two industries, namely metal Earthquake were not persistent, it caused trend shifts in
and machinery, and miscellaneous. the shares and numbers of workers at least for some
industries, and that in this sense the earthquake had a
4. Discussion persistent impact on the spatial distribution of industries
in Tokyo Prefecture. Here, we examine descriptive and
The statistical analyses in the previous section indicate anecdotal evidence to check this result. With respect to
that although the mean shifts in the shares and the the transience of the mean shift, Imaizumi (2008)
Fig. 7. Title: Decomposition of the impact caused by one standard deviation of the earthquake. Source: See the text.
A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68 65
Fig. 7 (continued).
discusses the strength of the recovery in damaged areas. Ebara County now has 430 thousand people, because
In those areas, many of the factory owners whose 110 thousand evacuators settled in Ebara County,
factories were burnt or destroyed tried to reconstruct whose population had been 320 thousand. Three
their factories at the original sites, which they owned.14 In hundred factories are actively operating here. The
addition, they sometimes cooperated to achieve recovery Ebara Industrial Association has been discussing how
in clusters.15 For instance, just after the earthquake, they can make enthusiastic efforts to take over
factory owners in Honjo and Fukagawa Wards created industries in Greater Tokyo, because many factories
the Association for the Industrial Recovery in Honjo and in the southeastern part of Tokyo have burnt down and
Fukagawa Wards (Honjo Fukagawa Kogyo Fukkōkai), are not recovering easily. The industries in Ebara
which engaged in such activities as distribution of County have also been damaged, but they have been
materials to assist the recovery of industries in this area. recovering vigorously. There was only a short period
On the other hand, there is also anecdotal evidence that when there were unemployed workers here, and
is consistent with the trend shift caused by the earthquake. recently factories have been recruiting large numbers
In December 1923, one of the major newspapers of workers. Although the county office planned to
published an article entitled “Powerful development of mediate the transfer of vacant factories and factory
industries in Ebara County: taking over the position of sites, there are no vacant factories, and appropriate
Honjo Ward.” In this article, the mayor of Ebara County sites are being purchased immediately for the
responded to the interviewer as follows16: construction of new factories.
It is stated that industries in Ebara County were
developing rapidly after the earthquake, and that delays in
recovery in the old industrial areas in southeast Tokyo had
14
The role of property rights regarding assets in the stability of boosted this development. In other words, the earthquake
spatial equilibrium is pointed out by Nunn (2014, p. 377). At the same damage in the old industrial areas provided an opportunity
time, it is true that some factories moved out of the original areas after for newly developing areas to take over the market
the earthquake. For example, Hayakawa Electronics (Sharp Electron- previously occupied by the old industrial clusters. At the
ics Co.) moved from Honjo to Osaka.
15
For discussion regarding cooperation in clusters, see Shmitz and
same time, this opportunity motivated the people and local
Nadvi (1999). governments of non-damaged areas to attract factories
16
Tokyo Asahi Shinbun, 15 December 1923. from other areas, including the damaged areas, as
66 A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68
Table 7
Persistence of the impact of the earthquake: By industry.
A. Dependent variable: WSHARE
(1) Textile (2) Metal and (3) Chemical and (4) Foods (5) Miscellaneous
machinery ceramics
NBROKEN*1923 − 0.0178 (− 3.50) ⁎⁎⁎ − 0.029 (− 4.57) ⁎⁎⁎ − 0.0196 (− 4.17) ⁎⁎⁎ − 0.0309 (− 6.59) ⁎⁎⁎ − 0.0359 (− 6.60) ⁎⁎⁎
NBROKEN*1924 − 0.0161 (− 3.22) ⁎⁎⁎ − 0.012 (− 1.95) ⁎ − 0.0073 (− 1.59) − 0.0155 (− 3.38) ⁎⁎⁎ − 0.0155 (− 2.91) ⁎⁎⁎
NBROKEN*1929 − 0.0063 (− 1.24) − 0.001 (− 0.13) − 0.0094 (− 1.99) ⁎⁎ − 0.0021 (− 0.44) − 0.0035 (− 0.64)
NBROKEN*1930 − 0.0026 (− 0.50) 0.001 (0.22) − 0.0069 (− 1.44) 0.0002 (0.05) − 0.0032 (− 0.58)
NBROKEN*1931 − 0.0059 (− 0.11) 0.002 (0.34) − 0.0045 (− 0.91) 0.0039 (0.81) 0.0000 (− 0.00)
NBROKEN*1932 − 0.0013 (− 0.23) − 0.002 (− 0.24) − 0.0051 (− 1.03) 0.0027 (0.54) − 0.0033 (− 0.58)
NBROKEN*1933 − 0.0011 (− 0.20) − 0.001 (− 0.15) − 0.0032 (− 0.62) 0.0000 (− 0.01) − 0.0028 (− 0.47)
NBROKEN*1934 − 0.0010 (− 0.18) − 0.001 (− 0.09) − 0.0020 (− 0.38) − 0.0005 (− 0.09) − 0.0017 (− 0.28)
NBROKEN*1935 − 0.0004 (− 0.07) 0.000 (− 0.02) − 0.0010 (− 0.19) − 0.0012 (− 0.22) − 0.0006 (− 0.10)
NBROKEN*TIMEAFTER − 0.0006 (− 1.34) − 0.002 (− 3.25) ⁎⁎⁎ − 0.0015 (− 3.30) ⁎⁎⁎ − 0.0004 (− 0.82) − 0.0009 (− 1.84) ⁎
NPLANTDENSITY1922* 0.0000 (0.06) − 0.001 (− 4.11) ⁎⁎⁎ − 0.0009 (− 3.54) ⁎⁎⁎ − 0.0012 (− 5.03) ⁎⁎⁎ − 0.0012 (− 4.26) ⁎⁎⁎
TIME
AREAI − 0.0043 (− 0.53) 0.008 (0.79) − 0.0014 (− 0.18) 0.0019 (0.26) 0.0141 (1.64)
AREAC 0.0021 (0.28) 0.014 (1.53) 0.0264 (3.78) ⁎⁎⁎ 0.0038 (0.54) 0.0029 (0.36)
Constant 0.0501 (27.25) ⁎⁎⁎ 0.048 (20.64) ⁎⁎⁎ 0.0460 (27.04) ⁎⁎⁎ 0.0488 (28.78) ⁎⁎⁎ 0.0482 (24.50) ⁎⁎⁎
Ward fixed effect Yes Yes Yes Yes Yes
Year fixed effect No No No No No
R-sq within 0.0799 0.2897 0.2645 0.2931 0.0296
Between 0.0181 0.0307 0.0068 0.0051 0.1077
Overall 0.0185 0.0601 0.0796 0.0279 0.0067
Obs. 260 260 260 260 260
suggested in the comments by the mayor of Ebara County. of making Nagoya a great industrial city. They distributed
Similar movements were also observed in more distant leaflets in damaged areas advising that Nagoya City was
areas. For example, Nagoya City made moves to attract keen to attract factory owners whose plants had been
factories from the damaged areas in Tokyo, with the aim damaged by the earthquake and who were planning to
A. Imaizumi et al. / Explorations in Economic History 60 (2016) 52–68 67
Fujita, Masahisa, Krugman, Paul R., Venables, Anthony J., 1999. The Ohkawa, Kazushi, Takamatsu, Nobukiyo, Yammoto, Yuzo, 1974.
Spatial Economy: Cities, Regions, and International Trade. MIT Kokumin Shotoku (National Income). Toyo Keizai Shinposha,
Press, Cambridge, MA. Tokyo (in Japanese).
Horiuchi, Ryōichi, 1978. Toshi Keikaku to Yōtochiikisei (City Redding, Stephen J., Strum, Daniel M., 2008. The cost of remoteness:
Planning and Zoning). Nishida Shoten, Tokyo (in Japanese). evidence from German division and reunification. Am. Econ. Rev.
Imaizumi, Asuka, 2008. Tokyo-fu kikai kanren kōgyō shuseki ni 98 (5), 1766–1797.
okeru Kantō Daishinsai no eikyō: Sangyō shuseki to ichijiteki Redding, Stephen J., Strum, Daniel M., Wolf, Nikolaus, 2011. History
shokku (Effects of the Great Kantō Earthquake on the machinery and industry location: evidence from German airports. Rev. Econ.
industry in Tokyo: industrial agglomeration and a temporary Stat. 93 (3), 814–831.
shock). Shakai Keizaishigaku (Socio Econ. Hist.) 74 (4), 23–45 Shmitz, Hubert, Nadvi, Khalid, 1999. Clustering and industrialization:
(in Japanese). introduction. World Dev. 27 (9), 1503–1514.
Imaizumi, Asuka, 2010. Yōto chiikisei dōnyu ga Tokyo-fu kikai Social Affairs Bureau, Ministry of Home Affairs, 1924. Shinsai Higai
kanren kōgyō shuseki ni motarashita eikyō: Toshikeikaku no kōka Chōsa (Survey on the Damage by the Great Kantō Earthquake).
to sangyō shūseki (Effects of land-use zoning on agglomeration of Social Bureau of the Ministry of Home Affairs, Tokyo (in
machinery industry in Tokyo: industrial agglomeration under city Japanese).
planning). Keiei Shigaku (Jpn. Bus. Hist. Rev.) 45 (3), 31–56 (in Social Affairs Bureau, Ministry of Home Affairs, 1926. Taisho
Japanese). Shinsai Shi (Record on the Great Earthquake in Taisho Era). vol.
Ishida, Yorifusa, 1987. Nihon Kindai Toshi Keikaku no 100 Nen (100 1. Social Affairs Bureau, Ministry of Home Affairs, Tokyo (in
Years History of Modern City Planning in Japan). Jichitai Japanese).
Kenkyusha, Tokyo (in Japanese). Society for Research on the Recovery, 1930. Teito Fukkō Shi (History
Kantō Bureau of International Trade and Industry, Ministry of of the Recovery of the Imperial Capital) vol. 1–3. Society for
International Trade and Industry, 1996. Sangyo Shuseki. Shin- Research on the Recovery, Tokyo (in Japanese).
jidai: Kudoka Kokufuku eno Teigen (The New Age of ‘Industrial Takebe, Rokuzō, 1925. Tokyo Toshi Keikaku no Kihon Mondai:
Agglomeration’: a Proposal for Overcoming De-industrialization). Jyūkyo, Shōgyō oyobi Kōgyō Chiiki (Fundamental Issues of the
Nikkan Kogyo Shinbunsha, Tokyo (in Japanese). City Planning in Tokyo: Residential Commercial and Industrial
Koshizawa, Akira, 1991. Tokyo no Toshi Keikaku (City Planning in Zones). Bureau of Reconstruction, Tokyo (in Japanese).
Tokyo). Iwanami Shoten, Tokyo (in Japanese). Tokyo City Government, 1925. Shinsai niyoru Nihon no Sonshitsu
Krugman, Paul, 1991. Geography and Trade. MIT Press, Cambridge, (Damage to Japan by the Great Kantō Earthquake). Isseisha,
MA. Tokyo (in Japanese).
Miguel, Edward, Roland, Gerald, 2006. The long run impact of Tokyo City Government, 1932. Teito Fukkō Kukakuseiri Shi (Record
bombing Vietnam. NBER Working Papers 11954. of Reconstructing the Imperial Capital and Land Adjustment) vol.
Nakamura, Takafusa, 1983. Economic Growth in Prewar Japan. Yale 1. Tokyo City Government, Tokyo (in Japanese).
University Press, New Haven (translated by Robert A. Feldman). Tokyo Prefecture Government, 1916. Tokyo-fu Kojo Tokei-sho
Numajiri, Akimobu, 2002. Kōjō Ricchi to Toshi Keikaku: Nihon (Manufacturing Statistics of Tokyo Prefecture). Tokyo Prefecture
Toshi Keisei no Tokushitsu (Plant Location and City Planning: the Government, Tokyo (in Japanese).
Making of Japanese Cities, 1905–1954). The University of Tokyo Whittaker, D. Hugh, 1997. Small Firms in the Japanese Economy.
Press, Tokyo. Cambridge University Press, Cambridge.
Nunn, Nathan, 2014. Historical development. In: Aghion, Philippe, Yano, Kenzō, 1932. Kōjō to Chiikisei (Factories and Zoning).
Durlauf, Steven N. (Eds.), Handbook of Economic Growth vol. Moriyama Shoten, Tokyo (in Japanese).
2A. Amsterdam, North-Holland, pp. 347–402.