Impact of Natural Disaster on Industrial Agglomeration: A Case of the Great Kanto Earthquake in 1923
Tetsuji Okazaki（The University of Tokyo） Asuka Imaizumi（JSPS）
Kaori Ito（Tokyo University of Science）
1 Background of the paper • Development of spatial economics （Fujita 1988; Krugman 1991; Venables 1996; Fujita, Krugman and Venables 1999） and its impact on economic history research (Kim 1995, 1998; Crafts and Mulatu 2005, 2006; Roses 2003)
• Historical economic geography project – Testuji Okazaki, Asuka Imaizumi and Kaori Ito, “Impact of Natural Disaster on Industrial Agglomeration: A Case of the Great Kanto Earthquake in 1923” – Yutaka Arimoto, Asuka Imaizumi, Kaori Ito and Kentaro Nakajima,”An information theoretic approach to identify spatial patterns of industrial agglomerations” – Asuka Imaizumi, Kaori Ito, Tomohiro Machikita and Tetsuji Okazaki, “The determinants of size distribution of plants in the early stage of agglomeration and industrial development: Japan, 1902-1919” – Yutaka Arimoto, Kentaro Nakajima and Tetsuji Okazaki ”Agglomeration or selection ?: The case of the Japanese silk reeling industry, 1909-1916”
2 Historical sources of spatial data on Japan
• Location data of each factory – Kojo Tsuran (Directory of Factories) • 1902 (issued in 1904) • 1919 (issued in 1921) • 1935 (issued in 1937) – Zenkoku Seishi Kojo Chosa (Census of Silk Reeling Factories) various issues from 1895-1932
• Historical administrative boundary – Historical Administrative Boundary web GIS database (by Murayama Lab. Tsukuba Univ.) • 1902 • 1919 • 1935 →city-(ward)-town-village level boundary
3 Distribution of factories in 1902: Colored by density on a city-town-village basis
4 Distribution of factories in 1935: Colored by density on a city-town-village basis
5 Motivation of the paper and related literature
• Mechanisms determining geographical distribution of economic activities – New Economic Geography (NEG) • First nature of a region is not the necessary and sufficient condition • Possibility of multiple equilibria – Empirical studies on multiple equilibria • Natural experiment using a temporary shock – Davis and Weinstein (2002), Brakman, Garresten and Schramm (2004), Miguel and Roland (2006) • Natural experiment using a persistent shock – Redding and Strum(2005), Redding, Strum and Wolf (2007), Nakajima(2007)
6 Strategy of the research
• Focusing on a temporary shock by the Great Kanto Earthquake which hit Tokyo in 1923
• Expanding the scope of empirical analysis – Geographical distribution of industrial activities in a relatively small area (Tokyo prefecture) • Similarity of first nature →Identifying the role of externality and scale economy (NEG factors) – Industry-level analysis • Revealing the difference in relative importance of the first nature and the second nature by industry – Temporary shock by a natural disaster • Pure exogeneity
7 Outline of the paper
• Shock by the Great Kanto Earthquake
• Overtime change in industrial agglomeration
• Testing the long-run impact of the Great Kanto Earthquake
8 9 10 11 Human damage
person, % Population just before Prefecture Death Missing Total Ratio to the population the Earthquake Total 11,743,100 91,344 13,275 104,619 0.89 Tokyo 4,035,700 59,593 10,904 70,497 1.75 Tokyo City 2,265,300 58,104 10,556 68,660 3.03 The other area 1,770,400 1,489 348 1,837 0.10 Kanagawa 1,379,000 29,614 2,245 31,859 2.31 Yokohama City 446,600 21,384 1,951 23,335 5.23 The other area 932,400 8,230 294 8,524 0.91 Chiba 1,347,200 1,373 47 1,420 0.11 Saitama 1,353,800 280 36 316 0.02 Shizuoka 1,626,300 450 42 492 0.03 Yamanashi 602,000 20 0 20 0.00 Ibaraki 1,399,100 14 1 15 0.00
12 13 Damage to buildings
Completely burnt Number of buildings just before the Prefecture and Percentage Earthquake destroyed Total 2,284,200 464,909 20.4 Tokyo 826,600 328,646 39.8 Tokyo City 483,000 305,146 63.2 The other area 343,600 23,500 6.8 Kanagawa 274,300 115,353 42.1 Yokohama City 98,900 72,408 73.2 The other area 175,400 42,945 24.5 Chiba 262,600 13,372 5.1 Saitama 244,900 4,562 1.9 Shizuoka 289,100 2,257 0.8 Yamanashi 117,000 562 0.5 Ibaraki 269,700 157 0.1
14 Variation of damage within Tokyo City by ward
Completely Total buildings destroyed Percentage and burntl Tokyo City Total 483,000 305,146 63.2 Kojimachi Ward 11,590 6,821 58.9 Kanda Ward 30,910 27,620 89.4 Nihonbashi Ward 23,190 21,616 93.2 Kyobashi Ward 31,880 29,290 91.9 Shiba Ward 38,640 17,167 44.4 Azabu Ward 19,320 906 4.7 Akasaka Ward 11,590 2,186 18.9 Yotsuya Ward 15,940 766 4.8 Ushigome Ward 26,080 515 2.0 Koishikawa Ward 32,360 1,663 5.1 Hongo Ward 28,500 7,463 26.2 Shitaya Ward 45,400 33,791 74.4 Asakusa Ward 63,760 59,325 93.0 Honjo Ward 59,890 55,274 92.3 Fukagawa Ward 43,950 40,743 92.715 16 Map of Tokyo prefecture (before 1932)
Minaniadachi Nishitama county county Kitatoshima county Minami- Kitatama county Toyotama katsushika Tokyo city county county
Minamitama county Ebara county
参考1 東京府(1932年まで）
17 Map of Tokyo city (before 1932)
Mimamiadachi county
Kitatoshima county
Shitaya Koishikawa Mimami- Hongo Asakusa katsushika county Ushigome Kanda Honjo
Yotsuya Kojimachi Nihonbashi Toyotama county Akasaka Kyobashi Fukagawa
Azabu
Shiba
Ebara county
参考2 東京市周辺部（1932年まで） 18 Estimation of the scale of damage: Comparison with Kobe Earthquake in 1995
• Estimated amount of damage by the Great Kanto Earthquake – 5.5 billion yen at 1923 price →6,168 billion yen at 1995 price =62.1% of that by Kobe Earthquake
• Ratio to GNP – Great Kanto Earthquake 35.4% – Kobe Earthquake 2.1%
19 Proportion of the population of Tokyo city
Figure 1 Proportion of the population of Tokyo Prefecture and Tokyo City
0.120
0.100
0.080
Tokyo Prefecture 0.060 Tokyo City
0.040
0.020
0.000 1893 1898 1903 1908 1913 1918 1923 1928 1933 1938
Source：Statistics Bureau of the Ministry of Internal Affairs and Communications ed. (2006); Statistical Yearbook of Tokyo Prefecture, various issues．
Note：From 1893 to 1918, Otsu-type de facto population．From 1923 to 1938, de facto population. Due to the adustment of the civil register, the population of Tokyo City incontinuously declined in 1909. Hence, we adjusted the population of Tokyo City before 1909, using the ratio of the de facto l i f T k Ci i 1908 h l i f T k Ci i 1908 i h T k Ci
20 Changes In geographic distribution in industrial workers
Number of workers Percentage 1922 1923 1936 1922 1923 1936 Total 183,521 119,012 376,718 100.00 100.00 100.00 Tokyo City Kojimachi Ward 2,335 1,671 4,008 1.27 1.40 1.06 Kanda Ward 5,984 1,435 7,676 3.26 1.21 2.04 Nihonbashi Ward 2,075 552 3,105 1.13 0.46 0.82 Kyobashi Ward 13,914 2,154 12,810 7.58 1.81 3.40 Shiba Ward 15,684 6,456 23,955 8.55 5.42 6.36 Azabu Ward 2,567 2,486 4,019 1.40 2.09 1.07 Akasaka Ward 421 568 844 0.23 0.48 0.22 Yotsuya Ward 675 834 934 0.37 0.70 0.25 Ushigome Ward 2,838 3,216 4,668 1.55 2.70 1.24 Koishikawa Ward 6,300 6,835 6,641 3.43 5.74 1.76 Hongo Ward 2,388 1,611 3,397 1.30 1.35 0.90 Shitaya Ward 3,227 1,827 6,564 1.76 1.54 1.74 Asakusa Ward 3,471 866 9,486 1.89 0.73 2.52 Honjo Ward 23,206 7,613 31,582 12.64 6.40 8.38 Fukagawa Ward 13,525 2,176 12,670 7.37 1.83 3.36 21 Correlation between the earthquake damage and change in the worker share from 1922 to 1923
Figure 4-A Correlation between the damage by the Earthquake and the change in the worker share from 1922 to 1923 Growth rate of worker share from 1922 to 1923 0.100
0.080
ρ=-0.706 0.060
0.040
0.020
Proportion of completely burnt and 0.000 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 destroyed buildings
-0.020
-0.040
-0.060
-0.080
22 Testing the persistence of the shock by the Earthquake (Davis and Weinstein 2002)
s itiit
it 1 itit 1
1 ss itit itit 1 it 1])1([)1(
1936 ss ii 1923 )(1( 1923 ii 1922) ess i
ρ=1：The influence of the temporary shock was persistent ρ<1：The influence of the temporary shock faded out at least in the long-run
23 Instrumental variable for sij1923-s ij1922
• Burn – The ratio of the buildings totally burnt of destroyed to the buildings just before the Earthquake
24 Power of IV (1)
Dependent variable：si1923-si1922
BURN -1.758 (-0.162) ***
Si1922-1915 0.057 (0.040)
R2 0.728
Obs. 97
Note: White heteroschedasticity robust standard errors are in parentheses.
*** statistically significant at 1% level.
25 Control variables
• sij1922-s ij1915 : Trend of deurbanization of industry i
• Density1923 : Level of congestion just after the Earthquke
• Indareai: Ratio of the area that was designated as “industrial zone” by the City Area Architecture Law in ward i
• Comareai : Ratio of the area that was designated as “commercial zone” by the City Area Architecture Law in ward i
• Readjusti : Ratio of the area where the division of land was readjusted by the government policy
26 27 Estimation results for the manufacturing industry total
Dependent variable: sij1936-sij1923 (1) (2) (3) sij1923-sij1922 -0.580 (0.109) *** -0.609 (0.098) *** -0.013 (0.354) sij1922-sij1915 -0.105 (0.125) -0.121 (0.108) -0.146 (0.116)
Density1923 -0.851 (0.232) *** -0.703 (0.284) ** Indarea 1.768 (0.487) *** Comarea 1.190 (0.597) ** Readjust 1.449 (4.520) Const. -0.003 (0.092) 0.316 (0.134) -0.211 (0.218) Number of obs. 97 97 97 R2 0.454 0.528 0.445
28 Summary of estimation results
• ρ=0.420 for the manufacturing industry as a whole, in case we do not control for policy variables →More than half of the influence of the Earthquake had disappeared by 1936 • ρis close to 1, in case we control for policy variables
→Influence of the temporary shock of the Earthquake was not persistent, but the recovery of the spatial distribution of industries can be basically attributed to policy intervention
29 Testing the persistence of the shock by the Earthquake by industry ij1936 ss ij1923 )(1( iji 1923 ss ij1922 *) i eIndustry ij i
30 Estimation results by industry
Dependent variable: sij1923-sij1936 (1) (2) (3)
(sij1923-sij1922)*Textile -0.752 (0.147) *** -0.772 (0.159) *** -0.468 (0.252) *
(sij1923-sij1922)*Machinery and Metal -0.057 (0.273) -0.085 (0.256) 0.578 (0.545)
(sij1923-sij1922)*Chemical -0.487 (0.201) ** -0.521 (0.201) ** -0.113 (0.408)
(sij1923-sij1922)*Foods -0.676 (0.180) *** -0.709 (0.167) *** -0.187 (0.330
(sij1923-sij1922)*Miscellaneous -0.807 (0.311) ** -0.852 (0.323) ** -0.032 (0.621) sij1922-sij1915 -0.104 (0.133) -0.120 (0.116) -0.164 (0.119)
Density1923 -0.895 (0.241) *** -0.686 (0.276) ** Indarea 1.772 (0.557) *** Comarea 1.287 (0.570) ** Readjust -0.638 (4.289) Const. -0.059 (0.094) 0.333 (0.138) ** -0.232 (0.234) Number of obs. 97 97 97 R2 0.409 0.488 0.441
31 Summary of estimation results by industry
• Persistence of influence differed across industries
– ρ is close to 1 for machinery and metal industry, even if we do not control for policy variables – ρ is smaller for textile, food and miscellaneous industries
32 33 Concluding remarks
• The Great Kanto Earthquake of 1923 gave a serious shock to the Japanese economy, 16.9 times larger than Kobe Earthquake of 1995, in terms of the ratio of damage to GNP
• Investigating the mechanisms determining the geographic distribution of industries, using that event as a natural experiment
• Testing the persistence of the influence of the shock on the distribution of workers – For the manufacturing industry as a whole, the influence of the shock was not persistent, but it was basically due to the policy intervention →Power of spontaneous recovery would not be large →Suggesting multiple equilibria – Persistence differed across industries • Influence was particularly persistent for machinery and metal industry • That industry was composed of numerous small and medium-sized firms • Inter-linkage of many firms might be a source of the multiple equilibria
34