the impacts of natural disasters on plants' growth: evidence from the great hanshin-awaji...

��

The Impacts of Natural Disasters on Plants’ Growth: Evidence from the GreatHanshin-Awaji (Kobe) Earthquake

Ayumu Tanaka

PII: S0166-0462(14)00111-2DOI: doi: 10.1016/j.regsciurbeco.2014.11.002Reference: REGEC 3097

To appear in: Regional Science and Urban Economics

Received date: 7 February 2014Revised date: 31 October 2014Accepted date: 5 November 2014

Please cite this article as: Tanaka, Ayumu, The Impacts of Natural Disasters on Plants’Growth: Evidence from the Great Hanshin-Awaji (Kobe) Earthquake, Regional Scienceand Urban Economics (2014), doi: 10.1016/j.regsciurbeco.2014.11.002

This is a PDF file of an unedited manuscript that has been accepted for publication.As a service to our customers we are providing this early version of the manuscript.The manuscript will undergo copyediting, typesetting, and review of the resulting proofbefore it is published in its final form. Please note that during the production processerrors may be discovered which could affect the content, and all legal disclaimers thatapply to the journal pertain.

http://dx.doi.org/10.1016/j.regsciurbeco.2014.11.002

http://dx.doi.org/10.1016/j.regsciurbeco.2014.11.002

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

The Impacts of Natural Disasters on Plants’ Growth:

Evidence from the Great Hanshin-Awaji (Kobe) Earthquake

Ayumu Tanaka†

November 13, 2014

Abstract

This study is the first attempt to examine the impacts of a natu-ral disaster at the plant level, focusing on the Great Hanshin-Awaji(Kobe) earthquake, which occurred in 1995 and affected numerousplants in Kobe city. In this study, I use plant-level data to re-examinethe creative disaster hypothesis, which states that natural disasters en-hance growth of firms or plants in stricken areas. I employ the match-ing method together with the difference-in-difference (DID) approachto reveal the quake’s effects. While most country and firm-level evi-dence support the creative disaster hypothesis, this study shows thatthe plants that survived in Kobe’s most devastated districts faced se-vere negative effects from the quake in terms of employment and valueadded growth during the subsequent three years.

Keywords : Natural disasters, Difference-in-difference, Plant growthJEL Classification: Q54, R11, C21

†Research Institute of Economy, Trade and Industry (RIETI).E-mail: [email protected], Tel: +81-3-3501-8577Adress: RIETI, 1-3-1, Kasumigaseki Chiyoda-ku, Tokyo, 100-8901 JAPAN

1

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

1 Introduction

The Great Hanshin-Awaji (Kobe) earthquake, which occurred on January17, 1995, is one of the largest natural disasters in Japanese history. Thedeath toll reached 6,437,*1 and the number of injured was 43,792 (Hayashi,2011). Furthermore, 182,751 buildings were completely destroyed. Suchdevastation severely affected the economy of the Japanese port city of Kobeand its surrounding area. The estimated economic damage was 9,926.8billion yen, approximately 2.1% of Japan’s GDP. *2

This study aims to investigate the impacts of natural disasters on plantgrowth, focusing on the Kobe quake. Previous studies on natural disas-ters such as Leite et al. (2009) confirm the creative disaster hypothesis,which posits that natural disasters enhance the growth of firms or plants inthe affected areas. The Kobe quake presents an ideal case for testing thishypothesis because it was unexpected and affected a large industrial area.

In this study, I employ plant-level data as well as the difference-in-difference (DID) and matching techniques. The empirical results of thisstudy show that the surviving plants in Kobe experienced lower employ-ment and value added growth during the three years following the quakethan plants in unaffected areas, although some did experience higher capitalgrowth.

The remainder of this paper is organized as follows. In Section 2, I reviewthe literature and explain the creative disaster hypothesis. In Section 3, Iexplain the methodology. Section 4 provides a description of the data usedin this study together with an overview of the quake’s economic impacts onplants in Kobe. In Sections 5 and 6, I present the results. Finally, Section7 presents the conclusion.

2 The creative destruction hypothesis

Natural disasters usually have negative impacts on the economy: they arehighly prone to destroying physical and human capital as well as publicinfrastructure. In the case of the Kobe quake, many studies have reportedsubstantial economic loss (Hayashi, 2011; Hondai and Uchida, 1998).

However, several empirical studies have reported a positive correlationbetween the frequency of natural disasters and long-run economic growth.

*1The death tolls of the great Kanto quake (1923) and Great East Japan quake (2011)were 105,000 and 15,845, respectively.

*2See, for example, Hayashi (2011), Horwich (2000), and Sawada and Shimizutani (2007;2008) for more details on the economic aspects of the Kobe quake.

2

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Skidmore and Toya (2002) investigated the long-run impact of natural disas-ters upon growth for the period 1960–1990. They found that the frequencyof climatic disasters is positively correlated with human capital accumula-tion, total factor productivity (TFP) growth, and GDP per capita growth.

The positive economic effect of natural disasters is termed creative de-struction (Cuaresma et al., 2008) or creative disasters (Leiter et al., 2009).The term reflects the positive correlation that can be interpreted as evidencefor the hypothesis that natural disasters provide opportunities to updateexisting capital stock and adopt new technologies, thereby functioning as atype of Schumpeterian creative destruction. Previous empirical studies suchas Skidmore and Toya (2002) and Leiter et al. (2009) are consistent withthe creative destruction hypothesis.*3 However, no study in the literature sofar has found a positive relationship between a geological disaster such as aquake and economic growth. In other words, the creative disaster hypothesishas not yet been supported by an empirical study in the case of geologicaldisasters.

The contribution of this study to the existing literature is twofold. First,it employs plant-level data. Skidmore and Toya (2002) employed cross-country macroeconomic data, and Leiter et al. (2009) employed Europeanfirm-level data. Neither of these types of data can capture the pure effectsof natural disasters since they include unaffected plants and regions. Evenif positive effects from natural disasters are observed at the country or firmlevel, the direct impacts on plants in the affected area can be negative. Fromthe viewpoint of policymakers in the affected area’s local government, theeffect on plants in their area is the primary concern. To the best of myknowledge, no study has thus far used plant-level data to investigate theeffects of natural disasters.

Second, this study employs the matching technique to control for pre-quake plant characteristics. After matching plants in the affected area withthose in unaffected areas, this study compares their growth paths. Thismethod will yield more precise impacts of natural disasters. Previous studieson the Kobe quake and other natural disasters employ neither plant-leveldata nor the matching technique. Therefore, this study is the first empiricalattempt to provide evidence on the precise impacts of natural disasters atthe plant level.

*3Siodla (2012) also studied the positive impacts of disasters but focused on urbanredevelopment.

3

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

3 Empirical strategy

This study employed two empirical methods to examine the impact of theKobe quake on plant growth: first, a simple DID estimation, following Leiteret al. (2009) and second, the matching method.

Treated (affected) plants were distinguished from nontreated (nonaf-fected) plants on the basis of their location since plant-level damage infor-mation is not available. Plants in Kobe were classified as treated plants andthose in other areas as nontreated plants. Plants in Osaka and other desig-nated areas around Kobe that obtained special support from the governmentafter the Kobe quake as part of the Disaster Relief Act were excluded fromthe analysis.

3.1 Difference-in-difference (DID) estimation

Using DID estimation, plants in the most devastated area of Kobe were com-pared with those in other major cities designated by government ordinance(seirei shitei toshi in Japanese). These include Kyoto, Nagoya, Yokohama,Kitakyusyu, Sapporo, Kawasaki, Fukuoka, Hiroshima, Sendai, and Chiba.*4

Following Leiter et al. (2009), the following DID equation was estimated:

ln yisr,t = η0 + β1 ∗ ln yisr,1993 + β2 ∗ kobeir + β3 ∗ aftert (1)

+β4 ∗ (kobe ∗ after)ir,t + industryis + ǫisr,t,

where i, s, r, and t index plant, industry, region, and year, respectively. Theyear t represents the period from 1995 to 1998. The dependent variables,ln yisr,t, are the log of employment (number of workers) and log of capital.I included their initial 1993 values, ln yisr,1993, as one explanatory variable,following Leiter et al. (2009).*5 Further, kobeir is a treatment dummyvariable that takes the value of one if a plant is located in Kobe, while

*4The choice of these cities as the control group appeared arbitrary. Therefore, plantsfrom the entire area of Japan were selected except the Kobe and surrounding areas whenthe matching method was employed. This study employed the matching method to addressthe selection bias, while DuPont and Noy (2014) adopted the synthetic control method toavoid potential bias arising from the selection of the control group.

*5In the literature on firm growth such as Angelini and Generale (2008), using the laggeddependent variable at the period t−1 as an explanatory variable is common practice. Thistype of partial adjustment model identifies the quake’s incremental impact. This study,however, followed Leiter et al. (2009) and used the initial value prior to the quake toestimate its cumulative impact to make it easier to compare the results from the DIDestimation with those from the matching method.

4

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

aftert is a dummy variable that takes the value of one after the year 1995—the year in which the Kobe quake occurred. Therefore, (kobe ∗ after)ir,tis the DID dummy that captures the effects of the Kobe quake. Finally,industryis is an industry fixed effect and ǫisr,t is an error term.

Based on the Cobb–Douglas production function, the log of value addedwas regressed as follows:

ln visr,t = δ0 + δ1 ∗ ln kisr,1993 + δ2 ∗ ln lisr,1993 + δ3 ∗ kobeir + δ4 ∗ aftert (2)

+δ5 ∗ (kobe ∗ after)ir,t + industryis + ηisr,t,

where ln visr,t is the log of value added while ln kisr,1993 and ln lisr,1993 arethe initial values of the log of capital and log of employment, respectively.The last term, ηisr,t, is an error term. As indicated by Leiter et al. (2009),the coefficient of DID dummy, δ5, captures the productivity effects of thedisaster.

3.2 Matching method

The matching method was employed to compare plants in the affected areawith those in unaffected areas to evaluate the quake’s effects on employment,capital, and value added growth, using plant-level panel data.

The effects of the Kobe quake on plant i’s outcome variables, ∆z, canbe written as follows:

∆z1ir,p+g −∆z0ir,p+g, (3)

where z represents the log of employment, capital, and value added. Su-perscript 0 refers to the nontreated (nonaffected) case, and 1 refers to thetreated (affected) case. The outcome variable has subscript r, which refersto a plant’s location to clearly express the nonavailability of the plant-leveldamage information. The subscript p represents the year in which the quakeoccurred. The fundamental problem of the causal inference is that ∆z0ir,p+g

is unobservable. The matching method was adopted to construct an appro-priate counterfactual that can be used in place of ∆z0ir,p+g.

Further, the average effect of treatment on the treated (ATT) was ex-amined as follows:

δ = E(∆z1ir,p+g −∆z0ir,p+g|Dir,p = 1) (4)

= E(∆z1ir,p+g|Dir,p = 1)− E(∆z0ir,p+g|Dir,p = 1),

where Dir,p indicates whether a plant i is located in an area affected by thequake in year t. Using the matching method, the counterfactual for the lastterm was constructed, E(∆z0ir,p+g|Dir,p = 1).

5

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Plants were matched using one-to-one and one-to-three nearest neighbormatching methods. Matched plants were selected from the entire area ofJapan except Kobe, Osaka, and other designated areas around Kobe thatreceived special support from the government. In the case of the one-to-onenearest neighbor matching method with replacement, a plant that has theclosest value of employment, capital or value added before the quake wasselected for each plant i in affected area r as follows:

c(i) = minj∈{Djr,p=0}

|| ˆzi,1993 − ˆzj,1993||. (5)

Plants were matched separately for each two-digit industry. After con-structing the control group through this matching method, the ATT wasestimated.

4 Data and overview

This section describes the data used in my empirical analysis. The datawere taken from the Census of Manufacturers, an annual compulsory surveyconducted by Japan’s Ministry of Economy, Trade, and Industry (METI).The survey covers all manufacturing plants in Japan that have more thanfour employees*6. The response rate is very high, over 90%. The rangeof variables collected by the survey depends on the year and plant size.For example, data on capital are collected for plants with more than 10employees. In the empirical analysis, all nominal values are deflated by anindustry-level deflator taken from the System of National Account Statistics.

Table 1 reports the number of plants, number of workers, sum of capital(fixed tangible assets), and sum of value added in Kobe before and afterthe quake. Table 1 also reports the recovery rate. The results for 1994 areunreliable because the number of plants is very small. Most plants in Kobecould not respond to the Census of Manufactures because the Kobe quakeoccurred on January 17, 1995—shortly after the 1994 survey was initiatedon December, 1994. Therefore, the 1994 survey is not used in my analysis.

Table 1 shows that the sum of capital and the sum of value added reachedover 90% of pre-quake levels in the three years after the quake, while thenumber of workers did not increase as rapidly and substantially. The number

*6The survey covered plants with less than three employees for some years before 2008,but this study analyzes plants with more than four employees due to computational limi-tations.

6

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 1: Aggregated variables of manufacturing plants in Kobe city (1993–2000)

Year No. of plants No. of workers Capital Value addedLevel Recovery Level Recovery Level Recovery Level Recovery

rate rate rate rate(%) (%) (billion) (%) (billion) (%)

1993 4,197 (100.0) 105,227 (100.0) 1,010.6 (100.0) 1,466.8 (100.0)1994 525 (12.5) 41,874 (39.8) 646.2 (63.9) 919.7 (62.7)1995 3,308 (78.8) 88,207 (83.8) 885.5 (87.6) 1,299.6 (88.6)1996 3,215 (76.6) 83,274 (79.1) 869.1 (86.0) 1,267.6 (86.4)1997 3,111 (74.1) 81,862 (77.8) 931.4 (92.2) 1,288.2 (87.8)1998 3,137 (74.7) 80,456 (76.5) 980.2 (97.0) 1,352.2 (92.2)

Notes: The data were taken from the annual report of the Census of Manufactures. Thedata on capital are the sum of tangible fixed assets for plants with more than 10 employees.The recovery rate indicates the ratio of each year’s values to 1993 values.

of workers remained at less than 80% of the pre-quake level in the three yearsafter the quake*7.

Table 2 reports the number of plants by plant type in Kobe. All plantswere divided into four types: exit, new, survived, and other. Exit plants areplants that existed in 1993 but exited before 1998. New plants are plantsthat did not exist in 1993 but entered during 1995–1997 and continued toexist until 1998. Survived plants are plants that existed during 1993–1998,except 1994. Plants excluded from the above three types are included inthe “other” plant category. This study focuses only on survived plants sinceplant growth after the Kobe quake is analyzed.

The terms “exit,” “new,” and “survived” do not indicate actual exit,entry, and survival for two reasons. First, the data does not cover plantswith fewer than three employees. Second, the survey cannot follow a plantif the plant relocates across a city or town. Thus, exit plants include plantsthat relocated to another city or town from Kobe, and new plants includeplants that relocated to Kobe from another city or town.

Table 2 shows that 1,842 plants, 43.9% of all plants in 1993, disappearedfrom the sample between 1994–1998. This suggests a substantial impact ofthe quake on manufacturing plants in Kobe. The number of new plants,780, is much lower than the number of exit plants. Therefore, the total

*7The number of workers in 1998 is even smaller than that in 1995. This tendency issimilar to other variables, but the reason for it is not clear.

7

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 2: Number of plants and fraction of plants by plant type

Exit plants New plants Survived plants Other Total

1993 1,842 0 2,091 264 4,197(0.439) (0.000) (0.498) (0.063) (1.00)

1994 130 0 359 36 525(0.248) (0.000) (0.684) (0.069) (1.00)

1995 585 249 2,091 383 3,308(0.177) (0.075) (0.632) (0.116) (1.00)

1996 372 384 2,091 368 3,215(0.116) (0.119) (0.650) (0.114) (1.00)

1997 222 475 2,091 323 3,111(0.071) (0.153) (0.672) (0.104) (1.00)

1998 0 780 2,091 266 3,137(0.000) (0.249) (0.667) (0.085) (1.00)

Notes: Exit plants are plants that existed in 1993 but exited between 1994–1998. New

plants are plants that did not exist in 1993 but entered during 1995–1997 and continued

to exist until 1998. Survived plants are plants that existed during 1993–1998, except 1994.

The figures in parentheses indicate the fraction of each plant type per year.

number of plants in Kobe decreased by over 25% (Table 1). Furthermore,the number of survived plants is 2,091. The proportion of survived plantswas approximately 50% before the quake but reached 66.7% three yearsafter the quake. Thus, it is evident that the relative importance of survivedplants increased after the quake.

The following sections focus only on the survived plants because theDID estimation and matching method examine only survived plants. Thismay result in an attrition bias because imaginably, the plants that did notsurvive were those that did not experience growth and could not reboundafter the quake. Descriptive statistics that compare survived plants with theexit plants based on observables in 1993 are presented in Table 3 to observepotential differences. Table 3 shows that survived plants are larger than exitplants in terms of employment, capital, and valued added. The differencesin the mean level of these variables between the survived and exit plants arestatistically significant. This suggests that smaller plants were more likelyto exit due to the quake. Therefore, it is possible that my estimates aredownward-biased because of a restricted sample and also a relatively higherseriousness as a whole of the quake’s impact on plants in Kobe city thanshown in the following sections.

Table 4 reports the number of survived plants and death toll due to

8

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 3: Comparison of pre-quake values of main variables by plant type inKobe city (1993)

N. of workers Capital Value added(billion) (billion)

Survived plants N 2,091 1,130 2,085Mean 35.95 29,878.59 30,885.70

[4.73] [3097.03] [3129.83]

Exit plants N 1,842 608 1,841Mean 14.63 10,808.21 11,935.40

[1.27] [1944.87] [1372.05]

Difference Mean 21.32 19,070.39 18,950.30t-statistic 4.36 5.21 5.55

Notes: Standard errors are given in brackets. T-statistics are derived from t-tests on the

equality of means. The data on capital are the sum of tangible fixed assets of plants

having more than 10 employees.

the quake in Kobe city by district. In addition, Figures 1 and 2 displaythe geographic distribution of the number of survived plants and death toll,respectively. Nine districts were classified into most and least devastatedareas, and plants in the most devastated six districts were identified asthe affected plants; data on plants in the three least devastated districts,Tarumi, Kita, and Nishi, were not used.*8 The numbers of plants in themost devastated and least devastated areas are 1,501 and 590, respectively.Furthermore, the death tolls in these two areas are 4,523 and 48, respectively.

Before undertaking detailed analysis in the subsequent sections, plantsin Kobe are briefly compared with those in other major cities. Table 5presents summary statistics for the full sample, which are then broken downon the basis of Kobe versus non-Kobe plants. The plants in Kobe city’smost devastated area are, on average, larger in terms of employment, cap-ital, and value added in 1993 and 1998 than those in other major cities inJapan. Their employment, capital, and value added growth are, however,much lower than those in other major cities between 1993–1998. This pro-

*8The estimated negative impact in the following sections might have been exaggeratedsince these three districts were not included. However, these three districts are clearlydifferent from the most devastated seven districts since the earthquake tremors were mea-sured at the highest level 7 on the seismic intensity scale (shindo in Japanese) by theJapan Meteorological Agency in the most devastated six districts only.

9

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 4: Number of survived plants and death toll in Kobe city (1995)

District No. of survived plants Death toll

Higashi-Nada 207 1,470Nada 90 934Hyogo 311 556Nagata 565 921Suma 119 399Tarumi* 79 26Kita* 79 13Chuo 209 243Nishi* 432 9

Most devastated area 1,501 4,523Least devastated area 590 48Total 2,091 4,571

Notes: The least devastated districts are marked with *. Data on the number of survived

plants are taken from the Census of Manufactures, and that on the death toll are from

Kobe city.

vides evidence against the creative disaster hypothesis, although it does notgive a conclusive explanation. Using the DID framework and the match-ing method, the difference between growth in the plants in Kobe and thosein other areas is formally examined to estimate the quake’s impacts in thefollowing sections.

10

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 5: Descriptive statistics: Kobe versus other major cities

Level (1993)Full sample Kobe Other major cities

N. of plants 23322 1501 21821Mean [SD] Mean [SD] Mean [SD]

N. of workers 33.45 [193.89] 37.10 [251.43] 33.20 [189.29]Capital 25242.50 [88692.68] 33200.55 [106712.10] 24725.69 [87375.51]Value added 25621.48 [125446.50] 28721.06 [157896.50] 25407.17 [122888.20]

Level (1998)Full sample Kobe Other major cities

Mean [SD] Mean [SD] Mean [SD]N. of workers 29.64 [160.24] 31.72 [203.85] 29.50 [156.80]Capital 27346.26 [95464.68] 38349.69 [112277.50] 26666.19 [94292.36]Value added 25193.22 [130700.10] 29133.22 [148371.60] 24920.60 [129387.70]

Growth (1993-1998)Full sample Kobe Other major cities

Mean [SD] Mean [SD] Mean [SD]N. of workers -2.60 [35.58] -4.62 [52.47] -2.46 [34.10]Capital 55.93 [1288.60] 52.46 [398.26] 56.15 [1325.92]Value added -0.88 [1975.18] -155.63 [6881.87] 9.83 [945.75]

Notes: Standard errors are given in brackets. The data on capital are sum of tangible fixed assetsfor plants with more than 10 employees. “Kobe” includes the most devastated area only, excludingTarumi, Kita, and Nishi districts. “Other major cities” are cities designated by governmentordinance (seirei shitei toshi in Japanese): Kyoto, Nagoya, Yokohama, Kitakyusyu, Sapporo,Kawasaki, Fukuoka, Hiroshima, Sendai, and Chiba.

11

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

[0,150)

[150,300)

[300,450)

[450,600]

HigashinadaNada

Hyogo

NagataSuma

Tarumi

Kita

Chuo

Nishi

Figure 1: Number of survived plants in Kobe city (1995)Note: Data on the number of survived plants are from Table 4.

12

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

[0,100)

[100,500)

[500,1000)

[1000,1500]

HigashinadaNada

Hyogo

NagataSuma

Tarumi

Kita

Chuo

Nishi

Figure 2: Death toll due to the Kobe quake (1995)Note: Data on the death toll are from Table 4.

13

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

5 Results

5.1 DID estimation

First, this section discusses the OLS results from the DID estimation ofequations (1) and (2). Table 6 reports the estimation results of equation (1)using the log of capital as the dependent variable. The first column reportsthe estimation results using data for the years 1993 and 1995. The base yearfor analysis is 1993 as it represents the plants’ pre-quake variable levels. Thesecond, third, and fourth columns report the results using data for the years1996, 1997, and 1998, respectively, as well as the base year 1993.

The DID dummy, kobe ∗ after, is insignificant in the first year after thequake and positively significant after 1997, two years after the quake. Thequake’s estimated impact on capital is approximately 13.0% (= exp(0.122)−1) in 1998, three years after the quake. This implies that the capital ofsurvived plants in the most devastated area of Kobe increased by 13.0%,on average, compared with plants in the other major cities. This result isin line with that of previous empirical studies such as Leiter et al. (2009)and the creative destruction hypothesis. Another dummy, after, is alsopositively significant, thereby suggesting that in both Kobe and other majorJapanese cities, the capital of manufacturing plants increased by 1.7%, onaverage, three years after the quake. In the fourth column, the dummy,kobe, is negative and significant. Therefore, on average, plants in Kobe hada comparatively low level of capital within each industry, but succeeded inincreasing their capital to a much greater extent than plants in other majorJapanese cities.

The results in Table 7 show that employment growth in manufacturingplants in the most devastated area of Kobe is significantly lower than inother major Japanese cities. The DID coefficient, kobe∗after, is significantlynegative. The estimated impacts of the Kobe quake on employment threeyears after the quake is −3.7%(= exp(−0.038)−1). As the coefficient aftersuggests, the average number of workers decreased in survived plants inmajor Japanese cities after 1995. Comparatively, a greater decrease occurredin the number of workers in plants in Kobe than in those in other majorcities. In the same period, Kobe city experienced a large population outflow.Both facts suggest that the labor market shrunk after the Kobe quake, asshown in Ohtake et al. (2012).

Finally, Table 8 reports the estimation results of equation (2). Again,the quake’s negative effects are evident. The DID coefficients, kobe ∗ after,are significantly negative in all the columns. The quake’s negative effect

14

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 6: Impact of the quake on capital growth: Kobe versus other majorcities

1995 1996 1997 1998

Initial capital 0.975*** 0.971*** 0.962*** 0.954***[0.002] [0.002] [0.003] [0.003]

kobe -0.039 -0.056** -0.063** -0.062**[0.025] [0.026] [0.028] [0.030]

after 0.045*** 0.031*** 0.029*** 0.017**[0.007] [0.008] [0.008] [0.009]

kobe*after -0.001 -0.019 0.067** 0.122***[0.031] [0.032] [0.034] [0.036]

Observations 20159 20038 19884 19622R-squared 0.906 0.896 0.884 0.87

Notes: Standard errors are given in square brackets. Constants and industry fixed effectsare suppressed. ***, **, * indicate significance at the 1%, 5%, and 10% levels, respectively.The base year is 1993.

Table 7: Impact of the quake on employment growth: Kobe versus othermajor cities

1995 1996 1997 1998

Initial employment 0.987*** 0.981*** 0.978*** 0.971***[0.001] [0.001] [0.001] [0.001]

kobe -0.014** -0.018*** -0.017** -0.012[0.006] [0.007] [0.007] [0.008]

after -0.018*** -0.025*** -0.035*** -0.071***[0.002] [0.002] [0.002] [0.002]

kobe*after -0.028*** -0.036*** -0.037*** -0.038***[0.007] [0.008] [0.008] [0.009]



15

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

on value added in survived plants in the most devastated area of Kobe isapproximately 11.1% (= exp(−0.118)−1), three years after the quake. Thisnegative effect can be interpreted as the negative effect on productivity sincethe Cobb–Douglas type production function was estimated.

In summary, the significantly positive effects of the quake are found onlyfor the growth of capital two years after the quake. With regard to em-ployment and value added, the study found that the quake had significantlynegative effects. The positive effects on capital are consistent with the cre-ative disaster hypothesis that natural disasters provide an impetus to capitalstock. However, the negative effects on value added are not consistent withthe hypothesis that natural disasters result in improvement in total produc-tivity, given that destroyed capital stock is replaced by new capital. Thelarge outflow of Kobe’s population can be one reason for these negativeeffects.

Furthermore, it should be emphasized that the positive effect on capitalgrowth appeared three years after the quake. A possible explanation for thisfinding is that it might take a long time to replace old capital with new aftera disaster. In contrast, the negative effects on the growth of employment andvalue added emerged shortly after the quake. These different time trendsare not taken into account by the creative disaster hypothesis.

5.2 Matching results

This section compares plants in the most devastated area of Kobe and thosein other areas of Japan, using the matching results. Unlike the DID es-timation, this section selects plants in the unaffected area of Japan withpre-quake characteristics similar to those of the plants in Kobe’s most dev-astated area. This enables us to estimate more rigorous impacts of thequake, controlling for pre-quake characteristics. The control plants in thissection are not restricted to plants in other major cities but are selectedfrom plants across all of Japan, excluding the quake-affected area. Figures3–5 depict the main results and Tables 15–17 in the Appendix present thedetailed results.

Figure 3 compares the average capital growth between survived plantsin the most devastated area of Kobe and those in other areas. The solidline represents the survived plants in Kobe while the broken line representsthose in other areas. The short and long broken lines represent results usingmatched and unmatched plants in the unaffected area, respectively. Figure 3shows a sharp difference between the average capital growth paths in plantsin Kobe and those in other areas.

16

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 8: Impact of the quake on the growth of value added: Kobe versusother major cities

1995 1996 1997 1998

Initial capital 0.187*** 0.192*** 0.194*** 0.196***[0.004] [0.004] [0.004] [0.004]

Inital employment 0.873*** 0.862*** 0.862*** 0.858***[0.007] [0.007] [0.007] [0.007]

kobe 0.137*** 0.134*** 0.136*** 0.144***[0.031] [0.031] [0.032] [0.033]

after 0.011 0.019** 0.006 -0.067***[0.009] [0.009] [0.009] [0.009]

kobe*after -0.123*** -0.105*** -0.107*** -0.118***[0.036] [0.036] [0.037] [0.038]


Notes: Standard errors are given in square brackets. Constants and industry fixed effects

are suppressed. ***, **, * indicate significance at the 1%, 5%, and 10% levels, respectively.

The base year is 1993.

17

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

100.0

104.3

100.6

110.0

113.8

100.0

104.7104.1

103.7103.2

100.0

105.9

106.9

105.8

104.4

10

0.0

10

5.0

11

0.0

11

5.0

1993 1994 1995 1996 1997 1998year

Kobe Rest of Japan (Unmatched)

Rest of Japan (Matched: N1)

Figure 3: Impact of the quake on the log of capitalNotes: The dependent variable is the change from t − 2 (1993) in the log of capital for

the plants in Kobe and the matched and unmatched control groups in the rest of Japan.

“Kobe” includes only the most devastated districts of Kobe city. The matched control

group is selected by one-nearest-neighbor matching. The lines represent the average of

each group.

After a decrease in plants’ capital in Kobe in 1996, their capital increasedby 13.8% through 1998. The capital of matched plants in other major citiesof Japan increased by 4.4% while that in unmatched plants increased by3.2%. Therefore, the average impact of the quake, ATT, three years after itsoccurrence is 9.4%, as reported in Table 15 of the Appendix*9. The resultsconfirm the positive effects of the quake on capital, as found in previoussections, and are in line with the creative disaster hypothesis.

Next, Figure 4 displays the average growth path of employment for eachgroup. The number of workers in plants in Kobe decreased by a muchgreater extent in the three years after the quake than in either the matched

*9It is difficult to directly compare the magnitude of the quake’s impacts in this withthose in the previous sections, since this section employs a different specification andsample; however, we can compare the sign of the impacts. The next section conducts theDID estimation with the matched sample to directly compare the magnitude.

18

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

100.0

95.4

93.793.1

90.0

100.0

99.1 98.898.3

95.3

100.0

101.8101.4

99.6

97.5

90

.09

5.0

10

0.0

10

5.0

1993 1994 1995 1996 1997 1998year



Figure 4: Impact of the quake on the log of employmentNotes: The dependent variable is the change from t− 2 (1993) in the log of employment

for plants in Kobe and the matched and unmatched control groups in the rest of Japan.



each group.

or unmatched control groups in other areas of Japan. In the three yearsafter the quake, the number of workers decreased by 10.0%. This negativeemployment impact from the quake is similar to that found in the previoussection. The quake’s average employment impact is −7.4%.

Finally, Figure 5 shows the quake’s impacts on the log of value added.Unlike plants in other areas of Japan, plants in Kobe saw their value addedlargely reduced by 10.0% until the end of 1998. The quake’s estimatedimpact on value added is −10.6%. This result is, again, inconsistent withthe creative disaster hypothesis.

In sum, the results of the matching method discussed in this sectionsuggest that plants in Kobe saw an increase in their capital after the quake,but no increase in value added. In addition, the number of workers alsodecreased. The empirical results suggest that the large increase in capitaland decrease in labor resulted in a less productive factor composition.

19

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

100.0

88.3

97.4 97.7

90.0

100.0

104.8

106.4 106.8

99.3100.0

104.4

107.5 107.9

100.7

90

.09

5.0

10

0.0

10

5.0

11

0.0

1993 1994 1995 1996 1997 1998year



Figure 5: Impact of the quake on the log of value addedNotes: The dependent variable is the change from t−2 (1993) in the log of value added for

the plants in Kobe and the matched and unmatched control groups in the rest of Japan.



each group.

6 DID estimation using the matched sample

6.1 Regression for each year

This section provides the results of the DID estimation using the matchedsample, whereas the previous section presented the results from the DIDestimation and the matching method separately. This approach enablesus to obtain a more precise picture of the quake’s impacts and comparethe estimated impacts as calculated by the two methods. In particular, theestimates based on the selected control cities may involve upward bias in thequake’s absolute impact if “spillovers” or “externalities” exist from Kobe tothese cities. In fact, Yokohama and Nagoya ports as well as Osaka portsgained from Kobe’s loss in the months following the Kobe quake (Chang,2000). This section attempts to alleviate potential bias by using the matched

20

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 9: Impact of the quake on capital growth: matched sample

1995 1996 1997 1998

Initial capital 0.990*** 0.978*** 0.960*** 0.946***[0.008] [0.009] [0.009] [0.010]

kobe -0.180 -0.154 -0.082 -0.450**[0.145] [0.158] [0.170] [0.180]

after 0.034 0.059* 0.047 0.034[0.033] [0.036] [0.039] [0.041]

kobe*after 0.009 -0.053 0.055 0.105*[0.047] [0.051] [0.055] [0.058]



control plants from the entire area of Japan except Kobe and other affectedarea and plant fixed effects, in this and the next subsections, respectively.

Tables 9–11 show the DID estimation results using the matched samplefrom one-nearest-neighbor matching. First, Table 9 presents the quake’simpact on capital growth. The number of observations decreases to 1,464from 20,159 in Table 6 since I used the matched sample in Table 9. The DIDdummy, kobe ∗ after, remains positive in both 1997 and 1998 but is signif-icant in 1998 only. The quake’s impact on capital growth is approximately11.1% (= exp(0.105) − 1). This result lies between the ATT based on thematching results, 9.4%, and the estimated impacts of the DID estimationwith unmatched sample, 13.0%, in the previous section.

Second, Table 10 shows the quake’s impact on employment growth. Theresults are qualitatively similar to those in Table 7 where the unmatchedsample was used. However, the results have quantitatively changed. Withthe matched sample, the quake’s estimated negative impact on employmentgrowth is approximately 7.1% (= exp(−0.074) − 1), three years after thequake. This is much higher than the estimated impact when the unmatchedsample, 3.7%, is used and is very close to the ATT, 7.4%. This finding sug-gests that the quake’s negative impacts on capital growth are more seriousafter controlling for pre-quake plant characteristics. In other words, plants

21

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 10: Impact of the quake on the employment growth: matched sample

1995 1996 1997 1998

Initial employment 0.980*** 0.976*** 0.970*** 0.960***[0.003] [0.003] [0.004] [0.004]

kobe 0.002 0.004 0.005 0.007[0.008] [0.008] [0.009] [0.009]

after 0.018** 0.014* -0.003 -0.024**[0.007] [0.008] [0.009] [0.009]

kobe*after -0.064*** -0.076*** -0.065*** -0.074***[0.011] [0.011] [0.012] [0.013]



in Kobe’s most devastated area after the quake decreased their employmentconsiderably, compared with plants with similar pre-quake characteristics inother area of Japan.

Third, Table 11 presents the quake’s impacts on value added using thematched sample. The DID dummy, kobe∗after, is again significantly nega-tive in all years after the quake. Furthermore, the estimated coefficients arelarger than those in Table 8. This results in a more serious, negative impact,−13.2% (= exp(−0.142)− 1), than that obtained from Table 8, −11.1%.

This subsection shows that the results using the matched sample arequalitatively similar to those found using the unmatched sample. How-ever, the results have quantitatively changed. The quake’s positive impacton capital growth reduced, while its negative impacts on employment andvalue added growth increased. To summarize, the quake’s impacts are moreserious when we control for pre-quake plant characteristics.

6.2 Pooled regression with plant fixed effect

In this subsection, the pooled sample for the period 1991–1998 is used, andthe treatment effect is estimated using the DID framework with plant fixed

22

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 11: Impacts of the quake on value added: matched sample

1995 1996 1997 1998

Initial capital 0.213*** 0.219*** 0.218*** 0.216***[0.012] [0.012] [0.012] [0.012]

Inital employment 0.843*** 0.840*** 0.841*** 0.852***[0.024] [0.024] [0.024] [0.024]

kobe -0.069 -0.126 -0.101 -0.061[0.130] [0.130] [0.127] [0.130]

after 0.036 0.048 0.049 0.001[0.039] [0.039] [0.038] [0.039]

kobe*after -0.160*** -0.111** -0.123** -0.142**[0.056] [0.056] [0.055] [0.056]



23

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

effects. I can control for the plant-specific factors by including the plantfixed effects into the estimation model. The model is as follows:

ln υisr,t = θ0 + γ1 ∗ aftert +Σq∈[1995,1998](γ2,q ∗ didir,q) + planti (6)

+yeart + νisr,t,

where i, s, r, and t index plant, industry, region, and year, respectively.The subscript t represents all years from 1991 to 1998, while the subscript qrepresents the after-quake years from 1995 to 1998. The dependent variables,ln υisr,t, are the log of employment (number of workers), log of capital, andlog of value added. The DID dummy is now included as didir,q for the yearsafter the quake, instead of kobe ∗ after in (1). The coefficients of the DIDdummies, γ2,q, can vary across years: 1995–1998.*10 Unlike equation (1),the plant fixed effect, planti, is included to control for any plant specificfactors, instead of industry fixed effects. Time-invariant variables, kobei,r,initial capital, and initial employment are dropped from the above equationdue to plant fixed effects. Another new variable is the year fixed effect,yeart. This allows the intercept to vary over time. Finally, νisr,t is an errorterm.

The estimation results of the equation (6) are reported in Table 12.For comparison purposes, Table 12 presents results using both the matchedand unmatched samples. Columns (1)–(3) present the results using the un-matched sample, while columns (4)–(6) present results using the matchedsample. The matched sample is that from the one-nearest neighbor match-ing.

First, the impact on the capital growth is shown in columns (1) and (4)of Table 12. Both columns present a significantly positive coefficient of theDID dummy in 1998, three years after the quake. However, there are nega-tive coefficients in both columns shortly after the quake. In particular, thecoefficient in 1996 is significantly negative in column (4). This negative co-efficient may reflect the fact that plants in Kobe had their capital destroyedby the quake. The coefficient turned to be significantly positive after 1997,shown in column (1) and after 1998, shown in column (4). This suggeststhat plants in Kobe took at least a few years to recover and increase theircapital. The magnitude of the positive impact on capital growth in 1998 isapproximately 6.2%(= exp(0.060) − 1) when the matched sample is used.This is much smaller than the 13.4% magnitude when the unmatched sam-ple is used. This is also much smaller than the 11.1% magnitude when the

*10Belasen and Polachek (2009) also investigated the time-varying effects of a naturaldisaster.

24

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

matched sample is used but with industry fixed effect instead of plant fixedeffect. These results, therefore, suggest that the positive impact on capitalgrowth decreases when plants in Kobe are compared with those in the restof Japan with similar pre-quake plant characteristics, and that a large partof it can be attributed to the plant-specific factors.

Second, the significantly negative impacts on employment and valueadded growth are found again immediately after the quake in both matchedand unmatched cases, contrary to the creative disaster hypothesis. The mag-nitude of the negative impact for the matched sample is quantitatively largerthan that for the unmatched sample. The results from the matching, there-fore, make the negative impact appear more serious. On the other hand, itmakes little difference whether I use the plant fixed effect or industry fixedeffect.

25

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Tab

le12:Cumulative

effects

ofthequake:

pan

elregression

swithplantan

dyearfixed

effects

(1991-2000)

(1)

(2)

(3)

(4)

(5)

(6)

Unmatched

sample

Matched

sample

lncapital

lnem

ploymen

tln

valueadded

lncapital

lnem

ploymen

tln

valueadded

after

0.016***

-0.062***

-0.047***

0.120***

-0.046***

-0.054***

[0.006]

[0.002]

[0.005]

[0.016]

[0.004]

[0.013]

did1995

0.016

-0.030***

-0.126***

-0.033

-0.050***

-0.179***

[0.026]

[0.006]

[0.021]

[0.021]

[0.006]

[0.017]

did1996

-0.014

-0.038***

-0.103***

-0.093***

-0.065***

-0.120***

[0.026]

[0.006]

[0.021]

[0.021]

[0.006]

[0.017]

did1997

0.069***

-0.039***

-0.107***

-0.025

-0.061***

-0.131***

[0.026]

[0.006]

[0.021]

[0.021]

[0.006]

[0.017]

did1998

0.126***

-0.040***

-0.110***

0.060***

-0.066***

-0.137***

[0.026]

[0.006]

[0.021]

[0.021]

[0.006]

[0.017]

Observations

57743

138088

66934

21320

40989

23536

No.ofplants

10841

23322

11385

3169

5496

3166

R-squared

0.002

0.027

0.008

0.012

0.029

0.018

Notes:

Standard

errors

are

given

inbrackets.

Constants

andyearfixed

effects

are

suppressed

.***,**,*indicate

significance

atthe

1%,5%,and10%

levels,

resp

ectively.

Thedata

isseven

-yearpanel

data

coveringtheperiod1991–1998,excluding1994because

ofthe

low

resp

onse

rate

dueto

thequake.

Thebase

yearis

1993.

26

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 13: Comparison of the estimated impacts in 1998, three years afterthe Kobe quake (%)

Capital Employment Value added

Unmatched resultsDID with industry fixed effects 13.0 -3.7 -11.1DID with plants fixed effects 13.4 -3.9 -10.4

Matched resultsATT 9.4 -7.4 -10.6DID with industry fixed effects 11.1 -7.1 -13.2DID with plants fixed effects 6.2 -6.4 -12.8

In sum, this section confirms and reinforces the previous results, whichare not consistent with the creative disaster hypothesis as follows. First,in contrast to the creative disaster hypothesis, the quake had significantlynegative impacts on employment and value added growth even when plant-specific factors are controlled for by using the panel data and the matchedsample. Second, the quake had positive impacts on capital growth, but theydecrease when the panel data with plant fixed effects is used. The matchingmethod reduces them even further.

To facilitate comparison of the quake’s impacts across different estima-tion methods, Table 13 presents all the estimated impacts of the quake. Theestimates from the DID estimation with plant fixed effects and matched sam-ples are the most rigorous since they fully control for plant specific factors.They indicate that the Kobe earthquake’s negative impacts on the growthof employment and value added are −6.4% and −12.8%, respectively, threeyears after the quake, while the positive impact on capital growth is 6.2%.

7 Concluding remarks

This study investigated the effects of the 1995 Kobe quake on the growthof capital, employment, and value added in plants in Kobe city. Plant-leveldata and both DID estimation and matching approaches were employed toisolate the pure effects of the quake. By using plant-level data, this studyprovides the first evidence that the Kobe quake’s impact on employment andvalue added are significantly negative, while its impact on capital is positivebut can only be found two or three years after the quake. Most previousstudies used country or firm-level data and support the creative disasterhypothesis, which posits that natural disasters enhance firms’ productivity

27

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

by encouraging firms to replace their existing capital with new capital andadopt new technology. However, the results from the plant-level data suggestthat plants in Kobe suffered from over-investment in physical capital and afailure to enhance productivity.

These over-investments in physical capital might have been induced byvarious government programs after the Kobe quake. In particular, the gov-ernment and the Bank of Japan conducted a special program to financiallysupport affected firms immediately after the quake. In some cases, theyfinanced firms without any collateral. The resulting over-financing mighthave resulted in over-investment.

To conclude, I would like to point out the limitations of this study andsuggest potentially effective ways to design an improved empirical study.One limitation of this study is the lack of damage information on individualplants. The quake may have generated idiosyncratic damage even withineach affected area. This study attempts to eliminate the potential biascaused by such uneven damage by using the plant fixed effects but onecould estimate the impacts of the quake more precisely if plant-level damageinformation was available.

Another limitation of this study is the lack of the data after 1998. Morerecent data would allow the impacts of the Kobe quake to be estimated overa longer period of time to obtain long-term effects. For example, the effecton capital only became significant two or three years after the quake. Itwould be useful to investigate if this impact is sustained over time but suchan examination is beyond the scope of this study.*11

Acknowledgments

This study was conducted as part of the Project on International Tradeand Investment in Japan undertaken by the Research Institute of Econ-omy, Trade and Industry (RIETI). I would like to thank Andrew Bernard,Masahisa Fujita, Toshiyuki Matsuura, Kentaro Nakajima, Yasuyuki Todo,Ryuhei Wakasugi, and other participants of seminars at RIETI for theirhelpful comments. I am deeply grateful to the editor (Daniel McMillen) andtwo anonymous referees for their valuable comments. I am also grateful toAndrea Leiter for explaining the methodology.

*11DuPont and (2014) examine the long-term effects of the Kobe quake but use aggre-gated data at the prefecture level instead of firm- or plant-level data.

28

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

References

[1] Angelini, P., and A. Generale. (2008). “On the Evolution of Firm SizeDistributions,” American Economic Review, 98(1): 426–438.

[2] Belasen, A., and S. W. Polachek. (2009). “How Disasters Affect Lo-cal Labor Markets: The Effects of Hurricanes in Florida,” Journal ofHuman Resources, 44(1): 251–276.

[3] Cuaresma, J. C., J. Hlouskova, and M. Obersteiner. (2008). “NaturalDisasters as Creative Destruction? Evidence from Developing Coun-tries.” Economic Inquiry, 46(2): 214–226.

[4] duPont IV, W. and I. Noy. (2014). “What Happened to Kobe? AReassessment of the Impact of the 1995 Earthquake,” Economic Devel-opment and Cultural Change, forthcoming.

[5] Hayashi, T. (2011). Economics of Huge Disasters (Dai Saigai noKeizaigaku), PHP Publishing. (in Japanese)

[6] Hondai, S. and T. Uchida. (1998). “Estimation of Losses by the Earth-quake in Kobe Manufacturing Sector using the ‘with and without’ Con-cept (Kobe Shi Seizogyo no Shinsai Higaigaku: ‘with and without’Gainen niyoru Suikei),” Kobe University, Department of Economics,Kokumin Keizai Zasshi, 178(5): 29–43. (in Japanese)

[7] Horwich, G. (2000). “Economic Lessons of the Kobe Earthquake,” Eco-nomic Development and Cultural Change, 48(3): 521–542.

[8] Leiter, A. M., H. Oberhofer, and P. A. Raschky. (2009). “Creative Dis-asters? Flooding Effects on Capital, Labor and Productivity withinEuropean Firms,” Environmental and Resource Economics, 43: 333–350.

[9] Ohtake, F., N. Okuyama, M. Sasaki, and K. Yasui. (2012). “Impactof the Great Hanshin-awaji Earthquake on the Labor Market in theDisaster Areas,” Japan Labor Review, 9(4): 42–63.

[10] Sawada, Y. and S. Shimizutani. (2007). “Consumption Insuranceagainst Natural Disasters: Evidence from the Great Hanshin-Awaji(Kobe) Earthquake,” Applied Economics Letters, 14: 303–306.

29

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

[11] Sawada, Y. and S. Shimizutani. (2008). “How Do People Cope withNatural Disasters? Evidence from the Great Hanshin-Awaji (Kobe)Earthquake in 1995,” Journal of Money, Credit and Banking, 40(2-3):463–488.

[12] Siodla, J. (2012) “Razing San Francisco: The 1906 Disaster as a NaturalExperiment in Urban Redevelopment,” mimeo.

[13] Skidmore, M.and H. Toya. (2002). “Do Natural Disasters PromoteLong-run Growth?” Economic Inquiry, 40(4): 664–687.

Appendix 1: Descriptive statistics

30

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Table 14: Aggregated variables of manufacturing plants in Kobe city byplant type (1993–2000)

Value added (billion) Exit plant New plant Survived plant Other Total1993 219.7 0.0 644.0 27.4 891.1

(0.247) (0.000) (0.723) (0.031)1994 69.9 0.0 310.3 16.2 396.3

(0.176) (0.000) (0.783) (0.041)1995 86.6 16.0 623.8 61.1 787.4

(0.110) (0.020) (0.792) (0.078)1996 59.4 44.2 690.5 43.7 837.8

(0.071) (0.053) (0.824) (0.052)1997 33.3 83.4 644.5 48.2 809.4

(0.041) (0.103) (0.796) (0.060)1998 0.0 112.7 640.8 46.2 799.7

(0.000) (0.141) (0.801) (0.058)

No. of workers Exit plants New plants Survived plants Other Total1993 26,941 0 75,170 3,116 105,227

(0.256) (0.000) (0.714) (0.030)1994 4,261 0 37,189 424 41,874

(0.102) (0.000) (0.888) (0.010)1995 10,854 2,763 70,327 4,263 88,207

(0.123) (0.031) (0.797) (0.048)1996 6,048 4,522 69,197 3,507 83,274

(0.073) (0.054) (0.831) (0.042)1997 3,520 5,775 68,816 3,751 81,862

(0.043) (0.071) (0.841) (0.046)1998 0 10,186 66,692 3,578 80,456

(0.000) (0.127) (0.829) (0.044)

Capital (billion) Exit plants New plants Survived plants Other Total1993 65.7 0.0 337.6 9.8 413.1

(0.159) (0.000) (0.817) (0.024)1994 9.4 0.0 169.1 0.4 178.9

(0.053) (0.000) (0.945) (0.002)1995 32.0 2.3 330.7 18.3 383.3

(0.083) (0.006) (0.863) (0.048)1996 18.7 6.9 316.3 13.8 355.7

(0.053) (0.019) (0.889) (0.039)1997 12.9 12.9 317.0 14.4 357.2

(0.036) (0.036) (0.888) (0.040)1998 0.0 39.7 297.0 17.9 354.7

(0.000) (0.112) (0.837) (0.051)

Notes: Exit plants are plants that existed in 1993, but exited before 1998. New plants are plantsthat did not exist in 1993, but entered during the period 1995–1997 and continued to exist until1998. Survived plants are plants that existed during the period 1993–1998, except 1994. Thefigures in parentheses indicate the share of each plant type per year.

31

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Appendix 2: Matching results

32

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Tab

le15:Theim

pactof

thequakeon

thecapital

ofsurvived

man

ufacturingfirm

sin

themost

devastatedareaof

Kob

e

(1)

(2)

(3)

(4)

(5)

No.of

Treated

Control

ATT

t-value

Balancing

Outcome:

lnK

treatedfirm

sproperty

One-nearest-neighbormatching

366

t0.043

0.059

-0.016

-0.36

Yes

t+

10.006

0.069

-0.062

-1.23

Yes

t+

20.100

0.058

0.042

0.76

Yes

t+

30.138

0.044

0.094

1.57

Yes

Three-nearest-neighbormatching

366

t0.043

0.030

0.013

0.36

Yes

t+

10.006

0.034

-0.028

-0.69

Yes

t+

20.100

0.012

0.088

1.92

*Yes

t+

30.138

-0.009

0.147

3.00

**

Yes

Unmatched

results

366

t0.043

0.047

-0.004

-0.12

t+

10.006

0.041

-0.035

-1.01

t+

20.100

0.037

0.063

1.67

*t+

30.138

0.032

0.106

2.61

**

Notes:

Thefiguresin

columns(1)and(2)are

thech

angefrom

t−

2(1993)in

thelogofvariables.

Thecommonsupport

conditionis

imposed.ATT

istheav

eragetreatm

enteff

ectontreatedplants.**and*indicate

significance

atthe5%

and10%

levels,

resp

ectively.

33

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Tab

le16:Theim

pactof

thequakeon

number

ofworkers

insurvived

man

ufacturingfirm

sin

themost

devastated

area

ofKob

e

(1)

(2)

(3)

(4)

(5)

No.of

Treated

Control

ATT

t-value

Balancing

Outcome:

lnEmploymen

ttreatedfirm

sproperty


1,234

t-0.046

0.018

-0.065

-2.64

**

Yes

t+

1-0.063

0.014

-0.077

-2.83

**

Yes

t+

2-0.069

-0.004

-0.065

-2.01

**

Yes

t+

3-0.100

-0.025

-0.074

-2.18

**

Yes


1,234

t-0.046

0.022

-0.068

-4.31

**

Yes

t+

1-0.063

0.023

-0.086

-4.61

**

Yes

t+

2-0.069

0.013

-0.082

-3.74

**

Yes

t+

3-0.100

-0.013

-0.087

-3.72

**

Yes

Unmatched

results

1,234

t-0.046

-0.009

-0.037

-6.15

**

t+

1-0.063

-0.012

-0.051

-7.21

**

t+

2-0.069

-0.017

-0.052

-6.51

**

t+

3-0.100

-0.047

-0.052

-5.94

**

Notes:

Thefiguresin

columns(1)and(2)are

thech

angefrom

t−

2(1993)in

thelogofvariables.

Thecommonsupport

conditionis

imposed.ATT

istheav

eragetreatm

enteff


significance

atthe5%

and10%

levels,

resp

ectively.

34

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Tab

le17:Theim

pactof

thequakeon

valuead

ded

ofsurvived

man

ufacturingfirm

sin

themost

devastatedarea

ofKob

e

(1)

(2)

(3)

(4)

(5)

No.of

Treated

Control

ATT

t-value

Balancing

Outcome:

lnValueadded

treatedfirm

sproperty


1,148

t-0.117

0.044

-0.161

-6.62

**

Yes

t+

1-0.026

0.075

-0.100

-3.97

**

Yes

t+

2-0.023

0.079

-0.102

-4.12

**

Yes

t+

3-0.100

0.007

-0.106

-4.02

**

Yes


1,148

t-0.117

0.047

-0.164

-7.81

**

Yes

t+

1-0.026

0.067

-0.093

-4.43

**

Yes

t+

2-0.023

0.071

-0.095

-4.63

**

Yes

t+

3-0.100

-0.013

-0.086

-3.93

**

Yes

Unmatched

results

1,148

t-0.117

0.048

-0.165

-11.17

**

t+

1-0.026

0.064

-0.090

-5.66

**

t+

2-0.023

0.068

-0.091

-5.37

**

t+

3-0.100

-0.007

-0.092

-5.05

**

Notes:

Thefiguresin

columns(1)and(2)are

thech

angefrom

t−

2(1993)in

thelogofvariables.

Thecommonsupport

conditionis

imposed.ATT

istheav

eragetreatm

enteff


significance

atthe5%

and10%

levels,

resp

ectively.

35

ACC

EPTE

D M

ANU

SCR

IPT

ACCEPTED MANUSCRIPT

Highlights

• There are positive effects of the quake on capital growth.

• There are negative effects of the quake on employment growth.

• There are negative effects of the quake on value added.

the impacts of natural disasters on plants' growth: evidence from the great hanshin-awaji...

Documents