東北大学機関リポジトリTOUR

(1)

Do governmental policies mimic natural

disasters? : Assessment of the 2012 Drug

Policy and its Effect on the Southern

Netherlands

著者

MOONEN KIRIN PEPIJN

学位授与機関

Tohoku University

学位授与番号

11301甲第18370号

(2)

Do governmental policies mimic natural

disasters?

Assessment of the 2012 Drug Policy and its Effect on the

Southern Netherlands

Moonen Kirin Pepijn

Supervised by

Professor N. Terui

Professor A. Hibiki

Professor K. Nakajima

A thesis submitted in fulfillment of the requirements for the degree of

Doctor of Philosophy

(3)

Do Governmental Policies Mimic Natural Disasters? Assessment of the 2012 Drug Policy on Coffee Shop Municipalities

Moonen K.P.

Abstract

Numerous studies have been conducted within the field of disaster and development economics on how natural disasters and governmental policies impact regional economics and their economic recovery or growth. While natural disasters and governmental policies share few characteristics, the theme raised within this dissertation is whether the outcome of government policies can result into negative consequences that affect a specific region as commonly seen with natural disasters. Though a direct comparison between natural disasters and government policies that behave as artificial disasters would provide no particular insight as the commonalities are too few, an possible issue is that the potential negative impact of governmental policies are understated. Therefore, this dissertation seeks to assess the impact of a natural disasters and governmental policy through two cases in particular, the Great East Japan Earthquake in Japan and the Dutch drug policy environment.

In Chapter 2, I focus on the impact and economic recovery of the Japanese manufacturing industry located within Eastern coastal municipalities struck by the March 2011 Tsunami. Using propensity score matching in combination with difference-in-differences on manufacturing census data from the Ministry of Trade, Economy, and Industry, I found that the impact of Tsunami on Japanese manufacturing was statistically significant. The most affected measurement was hu-man capital. While there is evidence of an economic recovery within the hu-manufacturing industry, discrepancies still remain after 3 years, mostly in the labor force when compared to similar munic-ipalities in other prefectures. In addition, I found that the rate of economic recovery is different dependent on municipality size, the amount of capital assets, and the type of industry. As is con-sistent with the literature, municipalities with larger density and capital have a quicker recovery. Industries classified as light manufacturing were less affected than industries classified as heavy manufacturing with quicker rates of recoveries. In addition, I checked for the possibility of spill over by limiting the census data to non-adjacent prefectures. There is evidence of a production spill-over when compared to prior results, as the magnitude of coefficient is significantly larger than prior results.

In Chapter 3, I analyze the impact of the 2012 marijuana sales restriction to foreign tourists on the municipal crime rate. Using propensity score matching in combination with difference-in-difference on municipal crime data from the Center Bureau of Statistics in The Netherlands, I found that the initial introduction of the revised drug policy, after controlling for municipal characteristics, led to an increase in drug crimes. The drug policy eventually led to a strong significant reduction in soft drug crimes over a period of 3 years.

In Chapter 4, I analyze the revision of the drug policy and its impact on firms dependent on tourism. In Chapter 4, the analysis shows that the policy had no initial impact on selected firm performance indicators such as net profit, operating result, and the net result. While there was a short reduction in firm revenues and costs in 2012, this result was not reported as statistically significant leading to two assumptions. First, the reduction in drug tourists did not significantly affect firm profitability indicating that firms are not reliant enough on drug tourists for the income. Second is the possibility that the aggregation from firm- to municipal level masks the impact of the policy.

(4)

List of Figures

2.1 Reconstruction Issues Faced By Firms in the Tohoku Region (2014) . . . 37 2.2 Status of Sales Recovery in Manufacturing Firms Receiving Group Subsidies . . . . 38 4.1 Volatility of the net profit per treatment and control group . . . 84 4.2 Volatility of net profit of Southern coffee shop municipalities . . . 85 4.3 Volatility of net profit of Northern coffee shop municipalities . . . 86

(6)

List of Tables

2.1 Descriptive statistics of Japanese coastal municipalities in 2010 . . . 17

2.2 Propensity score balancing test . . . 19

2.3 Effects of the Tsunami on Manufacturing Industries in Affected Coastal Municipal-ities (PSM-DID) . . . 20

2.4 State of Manufacturing Industries Before and After the Impact of the Tsunami in Affected Coastal Municipalities (PSM-DID) . . . 21

2.5 Effects of the Tsunami on Capital-Intensive Manufacturing Industries in Affected Coastal Municipalities (PSM-DID) . . . 23

2.6 State of Capital-Intensive Manufacturing Industries Before and After the Impact of the Tsunami in Affected Coastal Municipalities (PSM-DID) . . . 24

2.7 Effects of the Tsunami on labor-Intensive Industries in Affected Coastal municipal-ities (PSM-DID) . . . 25

2.8 State of labor-Intensive Manufacturing Industries Before and After the Impact of the Tsunami in Affected Coastal Municipalities (PSM-DID) . . . 27

2.9 Effect of the Tsunami on Affected Low Capital Municipalities (PSM-DID) . . . 29

2.10 State of Low Capital Municipalities Before and After the Impact of the Tsunami (PSM-DID) . . . 30

2.11 Effects of the Tsunami on Affected High Capital Municipalities (PSM-DID . . . . 31

2.12 State of Affected High Capital Municipalities Before and After the Impact of the Tsunami (PSM-DID) . . . 33

2.13 Reallocation of Production to Neighboring Coastal Municipalities . . . 36

3.1 Descriptive Statistics of Coffee Shop Municipalities in 2011 . . . 51

3.2 Descriptive Statistics of Coffee Shop Municipalities in 2012 . . . 52

3.3 Effect of the Soft Drug Sales Restriction on the Drug Crime Rate of Southern coffee shop Municipalities (PSM-DID) . . . 53

3.4 Drug Crime Rate of coffee shop Municipalities Before and After the Implementation of the Sales Restriction (DID) . . . 54

3.5 Propensity score balancing test of coffee shop municipalities . . . 55

3.6 Rosenbaum Bounds Test - Nearest-Neighbor Matching (4) . . . 56

3.7 Treatment effect of the 2012 Drug Policy on the Municipal Drug Crime Rate within Southern coffee shop Municipalities (PSM-DID) . . . 57

3.8 Drug Crime Rate of coffee shop Municipalities Before and After the Implementation of the Sales Restriction (PSM-DID) . . . 58

4.1 Descriptive statistics of hospitality and cultural firms within coffee shop municipal-ities before the implementation of the 2012 drug policy . . . 71

4.2 Descriptive statistics of hospitality and cultural firms within coffee shop municipal-ities after the implementation of the 2012 drug policy . . . 72

4.3 Propensity score balancing test - hospitality industry . . . 73

4.4 Propensity score balancing test - cultural industry . . . 74

4.5 Rosenbaum Bound Test on propensity score specification sensitivity - hospitality industry . . . 74

4.6 Rosenbaum Bound Test on propensity score specification sensitivity - culture industry 75 4.7 Effect of the Soft Drug Sales Restriction in coffee shop Municipalities on the Hospi-tality Firm Profitability (PSM-DID) . . . 76

4.8 Hospitality Firm Profitability Before and After the Implementation of the Soft Drug Sales Restriction in coffee shop Municipalities (PSM-DID) . . . 77

(7)

4.9 Effect of the Soft Drug Sales Restriction in coffee shop Municipalities on the Cultural

Firm Profitability (PSM-DID) . . . 78

4.10 Cultural Firm Profitability Before and After the Implementation of the Soft Drug Sales Restriction in coffee shop Municipalities (PSM-DID) . . . 79

4.11 Effect of the Soft Drug Sales Restriction in coffee shop Municipalities on the Hospi-tality Firm Assets (PSM-DID) . . . 80

4.12 Hospitality Firm Assets Before and After the Implementation of the Soft Drug Sales Restriction in coffee shop Municipalities (PSM-DID) . . . 81

4.13 Effect of the Soft Drug Sales Restriction in coffee shop Municipalities on the Cultural Firm Profitability (PSM-DID) . . . 82

4.14 Cultural Firm Assets Before and After the Implementation of the Soft Drug Sales Restriction in coffee shop Municipalities (PSM-DID) . . . 83

A.1 Census of Manufacturing - Variable Definition . . . 101

A.2 Census of Manufacturing - Variable Equation . . . 101

A.3 Statistics Bureau of Japan - Variable Definition . . . 102

B.1 StatLine - Definition of variables . . . 103

B.2 Micro Data - Income statement - Variable definitions . . . 104

(8)

Chapter 1

Introduction

In this introduction, I will briefly explain the relationship on how economies are affected by natural disasters, governmental policies, and what affects the recovery of these regional economies. Within the literature of regional, disaster, and development economics, not only the type of disaster and its magnitude but also the resilience of the affected region often determine the extent of the impact of exogenous shocks. Resilience within the field of disaster management is often divided in the ability of social-environment to absorb the initial impact of the disaster and recover from the lingering treatment effect such as organizing a recovery effect or displaced residents returning to the affected regions.12 _{Exogenous shocks, whether these shocks are natural disasters or governmental policies,}

impact regional areas on the social, economic, and ecological aspects.34

Skidmore and Toya (2002) argued that economies that frequently experienced natural disasters also had higher capital accumulation, factor productivity, and economic growth in the long run. They reason that this was due to the consistent cycle of creative destruction. The concept of creative destruction implies that when older physical technology is destroyed, there is an impetus on gaining and incorporating new and more efficient production methods. However, Noy and Nulasri (2007) argue that natural disasters do not provide economic growth in the long-run in their paper on the Great Hanshin Earthquake and Kobe’s economic recovery since 1990s. Further evidence that disproved the notion of Skidmore and Toya was Jaramillo (2009), Hallegatte and Dumas (2009), and Coffman and Noy (2011).

While macro-level governmental policies may be intended with the best of intentions, there is still the possibilities for policies to have unintended consequences that bring forth negative costs to the social, economic, and ecological aspects of society. Understanding the mechanics of how natural disasters affect the economy, is it possible that governmental policies can have similar patterns, where good intentions result in bad outcomes? Several studies such as Sweet (2014), Lawrence (2010), Thompson (2014), and MaCurdy (2015) show that government policies may be legislated with the intent to alleviate specific problems in terms of the social, economic, or ecological aspect, but policy makers may not understand the possible negative externalities because of policy implementation. An example of such a governmental policy is the Renewable Fuel Standard Act (RFS) in the United States. The intention of this federal policy program was to reduce greenhouse emissions and the wean the dependency of the United States economy of traditional fossil fuels in exchange for bio fuel by requiring fuel to be partially composed of ethanol. According to Stock (2015), the RFS is regarded as an economically inefficient program that has been a strong contributor to higher fuel and food costs.5_.

Sweet(2013) analyzed the federal transport policy objectives of alleviating traffic congestion in

1_{The strength of a social-ecological system partially dependent on the quality of organizational institutions within}

a region e.g. governmental agencies, cooperation between agencies, previous policies and re-assessment of these after the impact of a disaster.

2_{Economies near the coast tend to more severely impacted as ports tend to be large facilitators of the goods}

and services market and related processes. In addition, the economy of coastal regions are both economically and ecologically affected reducing production output in the agricultural and manufacturing industry

3_{The impact on these aspects depend on various characteristics such as type of disaster, its magnitude,}

environ-mental characteristics, and resilience policies.

4_{According to a literature review by Cutter et al. (2003), the recovery rate of individuals and communities is}

impacted by socio-economic metrics, gender, age, urbanization, quality of infrastructure, and occupation.

5_{This is further collaborated by Gorter and Dabrik (2016) who support that the link between bio-fuel and food}

prices caused a boom in food prices. In addition, countries that did no react through policy changes faced world high food prices

(9)

US metropolitan cities. Sweet found that while high traffic congestion had an initial positive impact on economic performance, the negative long-term effect in job growth and subsequent problem of economic inefficiency of constructing new highways annulled the initial positive effect. MaCurdy (2015) argued that raising the minimum wage fails as an anti-poverty initiative as the end-consumer would eventually be forced to pay for increased good prices, where it would be more beneficial to institute a typical sales tax. Other studies beside Sweet (2014) and MaCurdy (2015) show that governmental policies might mimic natural disasters by having unintended consequences regardless of the initial intention of the policy maker. Though the theme of this dissertation is primarily on the impact of the drug policy within the Netherlands implemented in 2012, the contribution of Chapter 2 to this dissertation is to show the adverse effects of a natural disaster on a regional economy and the subsequent recovery process.

The research focus of this dissertation is on the impact of exogenous events on regional economies. In Chapter 2, the focus is on how significant the manufacturing industry within coastal munici-palities along the Northeastern coastline were affected by the Great East Japan Earthquake and Tsunami that occurred in March 2011 and whether there is a significant difference in the recovery process between manufacturing industries. Chapter 3 focuses on the issue of drug tourism and the relationship between a drug prohibition policy that restricted sales to local municipal residents and registered drug crimes within municipalities covered by the policy in question. Chapter 4 continues with the theme of Chapter 3 by focusing on whether the reduction of drug tourists to Southern coffee shop municipalities affected the financial performance of firms within Southern coffee shop municipalities that implemented a drug policy that limited marijuana sales to municipal residents only.

The methodology used to analyze the prior mentioned research questioned within this disser-tation is of quasi-natural design. The combination of the Difference-in-Differences (DID) model to examine the effect of an exogenous shock on the outcome of interest and the Propensity Score Matching (PSM) model to create a balanced treatment and control group gives researchers with observational data the opportunity to mimic the natural sciences.

The DID model uses a simple setup where the change in the outcome of interest is observed for two groups. The first group is the treatment group, which consists out of observations who were exposed to the exogenous shock and the second group referred to as the control group, which consists out of observations not exposed to the exogenous shock. By deducting the average gain between both group before and after the occurrence of the exogenous shock, the difference between these two groups could be attributed to an exogenous shock.

Existing literature on the DID model states that it is imperative to verify whether the change between the treatment- and control group prior to the start of the exogenous shock are similar in order to fulfill the assumption that the control observations can be used as a valid comparison. Should there instead be a significant difference between these two groups, the estimation of the DID model is considered to be biased and thus be considered unreliable.

Therefore, in order to control for the issue of significant differences, I combine the DID model with propensity score matching (PSM) by Rosenbaum and Rubin (1983) to create a group of control observations that more closely resemble the observations in the treatment group and thus eliminate the possibility of significant differences. The purpose of the PSM method is to improve the estimation of the treatment effect by matching observations in the treatment and control group based on of a set of observable characteristics. While the DID model accounts for the possibility of unobservable fixed effects and time invariant factors, the addition of PSM would account for the observable differences in the set of chosen characteristics prior to the start of the exogenous shock. Therefore, the PSM-DID method is thus more likely to achieve an improved estimation of the treatment effect by additionally controlling for observable time-invariant and time-varying factors.

The combination of census data on the manufacturing industry from the Japanese Ministry of Economy, Trade, and Industry (METI) and other governmental statistics on Japanese coastal municipalities through the Japanese Statistics Bureau (E-STAT), the findings from the PSM-DID model revealed that the Tsunami significantly affected the manufacturing industry within coastal municipalities in 2011. Comparing the impact from 2011 to 2014, the manufacturing industry within affected coastal municipalities recovered except on human capital, which remained significantly affected. Difference in the speed of the recovery process, estimation results from the PSM-DID model indicates that capital intensive industries recovered more quickly in comparison to labor intensive industries. Similar pattern emerged when comparing the role of capital in recovering

(10)

from the Great East Japan Earthquake and Tsunami from 2011 for coastal municipalities where results show that municipalities with a capital value above the industry median recovered had a quicker recovery.

Chapter 3 focuses on the municipal crime rate and the revised drug policy which was introduced in 2012 to reduce the negative externalities because of increasing drug tourism by restricting the sale of marijuana to foreign tourists within Dutch municipalities located within the Southern Netherlands that had establishments, referred to as coffee shops, with a license to sell marijuana. Prior qualitative studies on analyzed the initial introduction of the revised drug policy in Southern municipalities and found that the policy resulted in a sharp increase in drug crimes ending in a significant reduction in 2013 and 2014. Due to the nature of the drug policy, the combination of PSM-DID is used to capture the treatment effect of the policy on Southern municipalities within the Netherland using data from the public Statline database maintained by the Dutch Center Bureau of Statistics (CBS). The empirical results from Chapter 3 thus indicates that the policy implemented in 2012 has had a beneficial impact on affected municipalities and verifies prior qualitative studies that highlighted the reduction in the drug crime rate. The contribution of Chapter 3 highlights that a drug prohibition policy such as a local sales restriction can effectively target drug tourism and reduce drug crimes.

The impact of the drug prohibition policy on the profitability of firms located within munici-palities covered by the policy in question is covered in Chapter 4. Chapter 4 continues with the original StatLine data set from CBS but adds an additional data set from the Microdata database that contains data on the financial performance of firms located within the Netherlands. With a similar methodical approach of combining the PSM and DID model, the findings of Chapter 4 indicates that the local sales restriction implemented in 2012 had no significant effect on firm profitability. The contribution of the research conducted in Chapter 4 to existing literature is empirical evidence that the significant reduction in drug tourism did not lead to a reduction in profitability for firms located within affected municipalities.

The rest of this dissertation is structured as follows. Chapter 2 presents the literature review, methodology, empirical results, and conclusion on the impact of the Great East Japan Earthquake on the manufacturing industry. Chapter 3 presents prior research on the effect of drug policies and the drug environment within the Netherlands, methodology, empirical results from the PSM-DID model, and the conclusion on the relationship between the drug prohibition policy, drug tourism, and the impact on the municipal drug crime rate. Chapter 4 presents existing literature on the profitability of drug tourism and prior studies examining the effect of drug tourism and firm prof-itability, methodology used for the empirical results, the data used and descriptive statistics, and conclusion. Chapter 5 presents the overall conclusion and contribution of the research presented in Chapters 2, 3, and 4, the limitations, and possible direction for future research.

(11)

Chapter 2

The recovery process of the

manufacturing industry in

(12)

2.1 Introduction

On March 11 2011 at 14:46, an Earthquake of 9.0 on the Richter scale occurred off the coast of Sanriku unleashing devastation that severely damaged the Eastern Coast of Japan. Approximately 561 km2 _{was flooded across 6 prefectures and 62 local municipalities with Miyagi Prefecture}

ac-counting for 60% (327 km2_{) of the total flooded area. As a result of the Earthquake, a Tsunami}

was unleashed on the coastal municipalities resulting in 10,549 human casualties, destruction of 82,999 and heavily damaging 155,129 residences, resulting in costs running up to 9.23 trillion yen. Existing literature on the impact of the Great East Japan Earthquake and Tsunami has shown that the two natural disasters had a tremendous impact on the social health of the residents in affected coastal municipalities (Olcott and Oliver, 2014; Goodwin, 2015; Kanehara et al., 2016; Hikichi, 2016; Yokomichi, 2016; Hikichi et al. 2017), the regional economy of affected regions along the Northeastern coast of Japan (MacKenzie et al., 2012; Kajitani and Tatano, 2013; Hamaguchi, 2015; Bachev and Ito, 2017), and the ecology (Urabe et al., 2013; Miura et al., 2017; Kijima et al., 2018).

Much of the literature on the Great East Japan Earthquake is concentrated on the factors that determine the recovery process of the affected regions since the impact in 2011. Factors often cited in recent literature are the importance of the supply chain (Todo et al., 2015; Wakasugi and Tanaka, 2015; Carvalho et al., 2016; Tokui et al., 2017), risk management (Zare, 2012; Cole et al., 2017; Yamori and Asai, 2017), the labor market within the affected region (Umezawa, 2014; Yamasaki et al, 2016, Kondo, 2017), and the assistance of third parties such as banks or government institutions (Hosono et al., 2012; Uchida et al., 2015).

Previous contributions to the literature on the impact of the natural disaster on firm supply chains such as Todo et al. (2015), Wakasugi and Tanaka (2015), and Tokui et al. (2017) indicated that the Great East Japan Earthquake resulted in a substantial reduction in manufacturing output. Todo et al. (2015) and Tokui et al. (2017) found that the diversification of supply chain networks to be an important factor for the recovery speed of firms within affected regions. Carvalho et al. (2016) examined the relationship between disrupted supply chains and sales performance of firms and found that if firm customers were located within affected regions led to a strong reduction of sales growth. However, a similar effect was not observed for firm suppliers located in affected regions and concluded that downstream directly and indirectly linked with the supply chain also exhibited significant negative sales growth.

According to research by Dekle et al. (2016), their analysis showed how the shock of the 2011 Great East Japan Earthquake affected Japanese manufacturing. They found that even though the Tohoku region experienced the direct impact of the natural disaster, the Chubu region within Japan, which is often cited as the manufacturing heartland of Japan was significantly affected by production disruptions. Dekle et al. argue that this is primarily due to the focus of the manufacturing industry within the Tohoku region on component production. Though firms may restart their manufacturing procedures, the impact of available human capital on firm performance should not be understated.

Previous studies such as Umezawa (2014) and Yamasaki et al. (2016) investigated the impact of the Great East Japan Earthquake and natural disaster on the labor market in affected regions and found that there was a mismatch between job vacancies and applicants within the labor market but the discrepancy remained even after changes in government policy.1 _{Research cites that one}

of the issues was the focus of local agricultural and manufacturing is concentrated on fisheries and marine products which offered lower wages than other forms of employment delaying the industry from recovering quickly (Umezawa, 2014; Yamasaki et al., 2016).

This chapter focuses primarily on the recovery rate of the manufacturing industry located within these affected coastal municipalities after the initial impact in 2011 up to 2014. The data sources for this chapter uses municipality data from the Census of Manufacturing published by the Japanese Ministry of Economy, Trade, and Industry to investigate the impact of the Great East Japan Earthquake and Tsunami on coastal municipalities in the prefectures of Iwate, Miyagi, and Ibaraki. Municipality characteristics were drawn from the e-stat portal maintained by the Japanese Statistics Bureau. I used propensity score matching with difference-in-differences to control for municipality characteristics before the initial impact of the Tsunami to match coastal municipalities

1_{Government policies include but were not limited to: lowering of the minimum requirements for unemployment}

benefits and an increase in the number of job vacancies related to the reconstruction activities, fund programs such as the temporary job creation subsidy (Cash-For-Work) to increase employment in affected regions.

(13)

not affected by the 2011 natural disaster with coastal municipalities that were affected in 2011. The findings of this chapter showed that the initial impact was indeed significant on the manu-facturing industry located within affected coastal municipalities in accordance with existing liter-ature. Results showed that the manufacturing industry has since almost recovered to pre-disaster levels since 2011 except for the amount of human capital and the generated gross value of man-ufacturing firms. To understand the difference in the recovery rate between industries I divided the sample into two groups through a capital-labor ratio. Industries with a capital-labor ratio above the industry median were designated as capital-intensive industries and industries with a capital-labor ratio below the industry median were designated as labor-intensive industries. Results indicated that capital-intensive industries had not yet recovered in the number of manufacturing firms in comparison to labor-intensive industries.

To test whether the amount of capital assets owned by manufacturing had an effect on the recovery rate I divided the sample in two by using the median of capital assets. Municipalities with the value of capital assets above the median were designated as high capital municipalities and those below the median were designated as low-capital municipalities. Results indicated that both groups were still significantly affected in terms of human capital, total cash wages and earnings paid to employees, and gross value added derived from manufacturing activities. However, while high capital had a quicker recovery rate in terms of gross value when comparing the magnitude with low-capital municipalities, the situation concerning human capital did not recover for high capital municipalities in comparison to low-capital municipalities. In addition, while not reported as significant, the degree of capital accumulation was greater to quite an extent when compared to low-capital municipalities indicating that capital access may be correlated with the amount of capital assets.

The contribution of this chapter is two-fold. First, while the research on the Great East Japan Earthquake has become quite extended since its initial occurrence, prior research has primarily focused on the damage and recovery of firms. Few studies have been conducted about the specific recovery rate of manufacturing industries and whether they have fully recovered to the level prior the occurrence of the natural disaster. The findings of this chapter therefore add to the gap in the literature by indicating that specific industries in affected coastal municipalities were still affected in 2014. Second, the combination of propensity score matching with difference-in-differences as the empirical approach in this chapter allows for the control of observed and unobserved municipality characteristics that further provides robustness to the findings of this chapter.

The structure of this chapter is as follows. Section 2.2 gives a literature review on the initial impact and recovery rate of the affected regions, Section 2.3 presents the methodology and Section 2.4 describes the data set constructed for the empirical analysis and descriptive statistics. Section 2.5 presents and discusses the results from the empirical model. Section 2.6 finalizes with the conclusion of the chapter and briefly highlights the findings and limitations of the research in combination with recommendations for future research.

2.2 Literature review

Earlier literature on the relationship between natural disasters and their impact on regional economies reported that the occurrence of natural disasters had positive correlations with long-run economic growth. Skidmore and Toya (2002) examined the impact of natural disasters on regional economies and found that the occurrence of natural disasters led to long-run growth in human capital accumulation, total factor productivity, and GDP per capita growth. Cuersma et al. (2008) suggested that the increase in GDP growth was likely a result of capital upgrading the infrastructure within affected regions to modern standards. However, later research pointed out that was mostly due to the creative destruction hypothesis where the increase in economic growth is often spend on recovery activities which will decrease after recovery processes are finished as pointed out by Dupont and Noy (2015) in their re-assessment on the city of Kobe which suffered the Great Awaji-Hanshin Earthquake in 1995. Using synthetic control to create a counterfactual Kobe, their findings revealed that Kobe did not return to pre-disaster income or price levels.

Anderson (1995) and Kreimer and Arnold (2000) claim that an important factor that determines the costs and subsequent recovery rate is due to the resilience of the region in terms of their institutions, the technological infrastructure, the socio-economic status of the region’s inhabitants, and the intensity and frequency of natural disasters in the region. Additional factors that are possible contributors to the recovery rate of affected regions are the resilience of supply chains

(14)

(Henriet et al., 2012), the capital access of affected regions, replacement of capital, (Rose, 2007; Noy and Vu, 2016), and risk management of firms (Abe and Ye, 2013).

Umezawa (2014) examined the impact on the local labor economy and government policies after the impact of the natural disaster. Umezawa investigated the local labor economy by using data from job-exchange activities, public job exchange offices, ”Hello-Work Offices” to analyze the number of job openings, job seekers, and new employment.2

Umezawa (2014) revealed that the number of job openings increased greatly after the natural disaster due to demand for restoration-related activities. This led to a mismatch between the supply and demand in the labor market after the disaster due to the closure of firms and loss of employment within affected regions. However, this trend reversed in 2012 where the ratio between job openings to applicants became high than the national average of Japan.3 _{Umezawa notes that}

the recovery of employment was difficult or slow within specific industries that were significantly devastated in the natural disaster. Interviews revealed that the reason for the slow employment recovery in specific industries is cited to due to higher wages offered for reconstruction activities while existing industries offered much lower wages.

Results presented by Umezawa (2014) were corroborated by Yamasaki et al. (2016). Their study analyzed the population movements and employment situation in Iwate, Miyagi, and Fukushima prefecture through using the Population Census data and the Basic Resident Registers. Umezawa (2014) used a similar approach by examining the differences between job offers and applicants, Yamasaki et al. (2016) found that the number of job offers increased after the natural disaster concerning reconstruction activities and in the manufacturing industry. Yamasaki et al. stresses the dire employment situation due to differences in the supply and demand within the local labor market.

Kondo (2017) estimated the impact of the Great East Japan Earthquake and the relationship between supply chain disruptions and employment. Using data from the Employment Status Survey conducted by the Statistics Bureau of Japan, Kondo found that shocks from the Great East Japan earthquake resulted in job relocation to other prefectures in Japan, which is corroborated by earlier research such as Umezawa (2014) and Yamasaki et al. (2016).

Olcott and Oliver (2014) examined the impact of the Great East Japan Earthquake and Tsunami on the manufacturing industry and the roles of social capital and sense in the rapid recovery after the impact in 2011.45 _{By interviewing 16 key individuals across five manufacturing}

firms within the Tohoku region between May and August 2011, their findings revealed that these manufacturing firms engaged in sharing of proprietary knowledge, alternative manufacturing re-sources, and firm manpower between rival firms. Therefore, while the social capital on individual level was severely impacted as indicated by Hikichi et al. (2017), the social capital for firms proved to be a contributing factor for the recovery rate for manufacturing firms within the Tohoku region. The impact on the public infrastructure was investigated by Sato and Suzuku (2013) by ex-amining the impact of the Great East Japan Earthquake on the transportation network within the prefecture and Iwate and the consequences to goods distribution and economic performance. Focusing on the initial three weeks after the impact of the natural disaster and using time-series data from the Annual Report on Prefectural Economic Accounting from the Cabinet Office of Japan, their findings indicated that disruptions from the impact had a significant effect on food distribution in the short-term and long-term consequences for the gross regional product. An example of transportation disruption was seen with the Nissan factory in Fukushima that could not be supplied with essential components for production, leading to a complete shutdown of all production until May 17.6

Wakasugi and Tanaka (2015) investigated the factors that prolonged the recovery period of manufacturing plants using data from the Research Institute for Economy, Trade, and Industry survey for firms located within the Tohoku region. Their findings revealed that the natural disaster led to a collapse of the supply chain, which caused a chain reaction affected even firms not located

2_{Hello Work is a Japanese government Employment Service center that maintains an extensive database of recent}

job offers for job seekers and provides job matching programs for the unemployed as part of its services.

3_{1.10 for Miyagi prefecture, 1.00 for Iwate- and Fukushima prefecture, and 0.80 for the national average.} 4_{Olcott and Oliver (2014) defines social capital as the reservoir of goodwill within a community of individuals or}

firms that is characterized by a sense of obligation to assist other members of the community who are in difficulty; by trust that those giving or receiving assistance will not unreasonably exploit the situation to their advantage; and by a high degree of shared knowledge and understanding, accumulated over repeated interactions.

5_{Olcott and Oliver (2014) defines sense making as the process by which an appropriate mental model of a situation}

is developed that allows the process of information and make intelligent choices.

6_{http : //www.meti.go.jp/english/earthquake/recovery/pdf /20110811}

(15)

in the disaster area resulting in a nationwide reduction in production. The disruption to the electricity network within the Tohoku region created a difference between the supply and demand of electricity with a shortage of 7.3% (5.85 million kW) to satisfy market demand.7 The cumulative effect of disruptions in the infrastructure, public utilities, and supply chain increased the recovery period for affected firms.

The impact of the Great East Japan Earthquake on the public infrastructure and utilities caused increasing disruptions within the supply chains of affected firms. Studies such as Todo et al. (2015), Cole et al. (2015), and Dekle et al. (2016) highlight the importance of the supply chain network for the recovery process of affected firms. Todo et al. (2015) examined the influence of supply chain network characteristics and its relationship between the impact of the Great East Japan Earthquake and firm resilience. With data from the Research Institute of Economy, Trade, and Industry survey and Teikoku Databank, Todo et al. found that broad supply chain networks were subject more to supply chain disruptions. However, diversification in the supply chain network also increased the probability of resuming manufacturing activities due to social capital between firms in the supply chain. If supply chain networks were located outside the affected region, this would contribute to a faster recovery of sales.

Cole et al. (2015) also used plant level data from the Research Institute of Economy, Trade, and Industry survey to estimate the effect of policies before and after the occurrence of disas-ter. They found that pre-disaster planning or post-disaster aid had no significant effect on the short-term impact on plant operations. However, pre- and post-disaster policies did influence the post-disaster sales growth and that cooperation between plants on production allocation / sub-stitutions also affected sales after the occurrence of the natural disaster. Cole et al. also looked at the influence on sales from assistance from third parties such as banks, trading partners, and government institutions. While banks and trading partners were reported as significant influences on post-disaster sales, the same could not be proven for governmental aid.

While Cole et al. (2015) revealed that governmental aid may not have had a significant effect on the recovery process for manufacturing firms, Uchida et al. (2015) investigated the indirect damage of the Great East Japan Earthquake on non-affected firms with trading partners within the affected area and whether such a shock had an effect on the probability of bankruptcy. Using firm-level data from the Teikoku Databank Ltd, their findings revealed that damaged trading partners reduced the probability of bankruptcy. They also investigated the influence of public capital injection into damaged firms and found that there was some evidence that capital injections reduced the probability of bankruptcy. The fact that there is some evidence that supports that capital injections are significant to prevent firm bankruptcy indicates that governmental aid or policies are indirectly a factor for the recovery process.

Dekle et al. (2016) followed by examining whether the Great East Japan Earthquake signifi-cantly affected production in other regions than Tohoku, Dekle et al. used monthly regional-level industrial production to estimate the shock to the industrial production of other regions within Japan and found that production in the Chubu region had the largest response to disruptions in the Tohoku region. Their reasoning for this is that a large percentage of the industrial firms within the Chubu region rely on components made by factories within the Tohoku region.

The literature review on prior studies indicates that the Great East Japan Earthquake had a significant effect on the state of the manufacturing industry within the Tohoku region. Though these studies are extensive, the contribution of this chapter seeks to broaden existing literature by examining the recovery process of the manufacturing industry of Japanese coastal municipalities that were affected by the Tsunami in March 2011.

2.3 Methodology

The purpose of this chapter is to estimate the impact of the Great East Japan Earthquake and Tsunami on the manufacturing industry within affected regions along the Northeastern coastline of Japan. Unlike natural experiments, the issue with observational data is that only one outcome for an affected coastal municipality can be observed. To overcome this issue, I use the difference-in-difference (henceforth referred to as DID) model on yearly manufacturing data from 2008 to 2014. The DID model compares the change of the outcome variable of interest between the treated- and

7_{The shortage of electricity, especially at peak demand, led to several blackouts during the early months after}

the Great East Japan Earthquake. However, in response to this, the Ministry of Trade, Economy, and Industry devised several policies to reduce electricity usage and increase capacity.

(16)

control coastal municipalities before and after the occurrence of the Great East Japan Earthquake and Tsunami in 2011. Treated coastal municipalities are defined as coastal municipalities located within the prefectures of Iwate, Miyagi, and Ibaraki that were that were directly harmed by the occurrence of the Tsunami in March 2011 and are situated along the Northeastern coastline with direct access to the sea. The control coastal municipalities are those located in other Japanese prefectures and did not experience the Tsunami in March 2011. The DID model specification is as follows:

yit= α + δDi x dt+ µi+ λt+ ωi+ it, (2.1)

where yit is the outcome of interest of municipality i in year t, D is the treatment variable that

is equal to 1 if a coastal municipality experienced the Tsunami in 2011 (where the Tsunami hit the municipal domain and caused damage) and 0 otherwise, d is the binary variable that denotes when treatment starts which is equal to 1 after the occurrence of the Tsunami in 2011 and 0 otherwise, δ is the difference-in-differences estimator that captures the average treatment effect on the treated (ATT) which in this case is the impact of the Tsunami on coastal municipalities, µi is the municipality fixed effect, λt is the time fixed effect, ωi is the linear time trend, and it

is the error term. The municipality fixed effects term (µi) captures the time invariant differences

between coastal municipalities. The time fixed effect (λt) captures the change in municipality

characteristics over a period of time that effects all coastal municipalities, and the linear time trend (ωi) de-trends any municipal specific changes.

The variable of interest is the interaction term, δ, captures the ATT of coastal municipalities affected by the Tsunami in 2011. The DID estimator captures the changes between treated-and control coastal municipalities in the pre-treatment treated-and post-treatment periods. In order for the estimation results from the DID estimator to be considered robust, the trend of the outcome variables between treated coastal municipalities (those affected by the Tsunami in 2011) and control coastal municipalities has to follow a similar path before the occurrence of the Great East Japan Earthquake in 2011 which is known as the parallel trend assumption. If treated- and control coastal municipalities do not follow a similar trend in the outcome variable of interest then the DID estimator is assumed to be biased due to significant differences continuing after the occurrence of the Tsunami. To check for bias in the pre-treatment DID estimator, I follow Autor (2003) which estimates the pre-treatment trend of treated- and control coastal municipalities through the following specification: yit= 2 X k=1 βkDi x dt+k+ 3 X k=0 β−kDi x dt−k+ µi+ λt+ ωi+ it, (2.2)

where yit reports the outcome of interest of municipality i in year t, βkDi x dt+k represents the

two lead dummies which is equal to 1 for treated coastal municipalities in the pre-treatment period and 0 otherwise, β−kDi x dt−k represents the three lag dummies which is equal to 1 for

treated coastal municipalities in the post-treatment period and 0 otherwise. Equation 2.2 estimates two pre-treatment effects, the treatment effect at the time of impact, and three post-treatment effects. If the trend of the outcome variable in the pre-treatment is reported as being statistically different between treated- and control coastal municipalities then the DID model does not satisfies the parallel trend assumption. Additionally, the post-treatment interaction terms as detailed by Equation 2.2 also captures the yearly change on the outcome variables of interest after the impact of the Tsunami in 2011.

The DID model captures the unobservable characteristics between treated- and control coastal municipalities through municipality- and time fixed effects as shown in Equation 2.1 and 2.2. In addition, by adding a linear time trend to the specification, the DID model also controls for a specific trend in the outcome variable of interest for each municipality. Though the sample is restricted to coastal municipalities and the DID model captures for unobservable characteristics, there might be a significant difference between the observable characteristics between treated- and control coastal municipalities and thus lead to a biased DID model.

Rosenbaum and Rubin (1983) propose a statistical matching technique named propensity score matching (henceforth abbreviated as PSM) that estimates the effect of an intervention on the outcome by comparing the differences across observations in the treatment- and control group. The PSM method uses a variable referred to as the propensity score (PS) which Rosenbaum and Rubin (1983) define as the probability of treatment that an observation can receive based on a set

(17)

of chosen covariates. However, the PSM method has to satisfy two key assumptions in order to have a successful match between the treated- and control group.

First is the conditional dependence assumption and states that the probability of being as-signed to the treatment group at the start of the intervention dependent on a set of observable characteristics where (yi0, yi1` Di|p(Xi)). If the first assumption is satisfied then there would

no significant bias if treatment selection is fully captured by the set of observable characteristics and the difference between treated- and control coastal municipalities is reported as statistically insignificant.

The second assumption is the common support between treated- and control coastal municipal-ities which states that dependent on the treatment status in the pre-treatment period that there are control coastal municipalities with a similar or equivalent PS value derived from their set of X variables.

I used a logit model with a set of control variables in the pre-treatment period to create a counterfactual group of coastal municipalities that more closely resembled the group of treated coastal municipalities affected by the occurrence of the Tsunami in 2011. I used the gross value added of firms which is calculated as their sales minus cost of sales and taxation as the outcome variable and following explanatory variables for the matching procedure: number of manufacturing firms, number of employees working in manufacturing firms, total cash wages and earnings paid to manufacturing employees (billions of yen), the value of goods manufactured and shipped (billions of yen), population density (number of people per hectare), the percentage share of the agriculture industry of the total municipal economy, and the percentage share of the manufacturing of the total municipal economy.8

To check for robustness in terms of matching and results reported by the DID estimator, I used pair matching with and without replacement, kernel matching, and nearest neighbor matching with replacement. To improve matching results, I imposed a maximum distance between the PS of treated- and control coastal municipalities by using a caliper. As for the caliper size, I followed Austin (2011) and used a caliper size that was 25% of the pooled standard deviation of the logit model. Comparing the matching result from different algorithms in addition to the estimation results from the DID model, I found results to be consistent. In the empirical analysis, I used the results from the nearest-neighbor matching up to 8 neighbors with replacement as it scored the best on the balancing test between treated- and control coastal municipalities on the chosen set of control variables.

The combination of propensity score matching with difference-in-differences (abbreviated as PSM-DID) shows that selecting an appropriate group of control coastal municipalities to be matched to treated coastal municipalities eliminates significant observable differences. The DID model controls for unobservable and observable time invariant characteristics that may influence both groups. Therefore, the PSM-DID model improves the overlap between the treated- and control coastal municipalities and control for the influence of unobservable and observable time invariant characteristics.

2.4 Data and descriptive statistics

In March 2011, the Great East Japan Earthquake unleashed a Tsunami that hit the Northeastern coast of Japan in the Tohoku region. Coastal municipalities located within prefectures such as Aomori, Iwate, Miyagi, Fukushima, Ibaraki, and Chiba were significantly damaged by the Tsunami when it hit. The Great East Japan Earthquake and Tsunami present an opportunity to examine on whether the initial impact of the Tsunami was significant and whether the affected coastal municipalities have recovered.

To examine whether the manufacturing industry has recovered since 2011, I therefore chose to include coastal municipalities from the Northeastern coastline that were damaged in 2011 by the Tsunami and that have direct access to the sea. The majority of coastal municipalities were located within the prefectures of Iwate, Miyagi, and Ibaraki. Affected coastal municipalities from Aomori and Chiba are also included into the treatment group but reports state that the damage inflicted by the Tsunami was minor in comparison to coastal municipalities within the prefecture of Iwate or Miyagi.

8_{Tables A.1, A.2, and A.3 in Appendix A provides the definition of the variables used and defined by METI and}

(18)

Fukushima prefecture is not included in the analysis of the Tsunami impact due to several endogenous factors such as the Daiichi Nuclear Power Plant incident, the evacuation zone of most coastal municipalities within Fukushima, the reputation damage of agricultural, and the lack of agricultural and industrial production due to damage to infrastructure and the labor market.

The reason for why these coastal municipalities were chosen is due to the fact that the Tsunami as a result of the Great East Japan Earthquake can be categorized as an exogenous shock due to its characterization as an act of god. While other prefectures within the Tohoku region or close to might have experienced the Earthquake, only coastal municipalities along the Northeastern coastline experienced the Tsunami in 2011.

The sample dropped the prefectural capital of Tohoku, Sendai, from the sample due to as-sumption that as the largest coastal municipality within the affected regions, it would most likely experience a quicker recovery of the manufacturing industry than in comparison to other coastal municipalities. Factors such as an increase in incoming migration from the surrounding affected regions such as Fukushima would possibly bias the results in matching procedure and thus present a skewed treatment effect. Additionally, according to a theory proposed by Cuersma et al. (2008), the investment-hypothesis might have a factor in the recovery.

The coastal municipalities from Tokyo and Kyoto prefecture were also not included in the sample. The assumption behind dropping these observations is that though coastal municipalities share commonalities with one another based on their observable characters, the coastal municipal-ities or wards of Tokyo and Kyoto prefecture are located within the financial and cultural capital of Japan. One issue is the likelihood that even though the municipal characteristics are similar, the output and state of manufacturing might have access to capital markets due to their favorable location.

I used two data sources for the empirical analysis of this chapter. The first data source is the Census of Manufacturing constructed by the Japanese Ministry of Economy, Trade, and Industry (henceforth referred to as METI). The Census of Manufacturing houses data on the manufacturing industry within Japan and is created through an annual questionnaire aimed at all manufactur-ing firms (with four or more employees) located within Japan through enumerators.910_{. METI}

distributes two forms of the questionnaire: Form A is designated to manufacturing firms with 30 or more employees while Form B is for establishments with 29 or fewer employees. The survey is based on entries in these forms filled by firm managers or administrators. The response rate by manufacturing firms to the annual questionnaire is 95.6%.

The questionnaire collects data on the following aspects of manufacturing firms: the number of employees working for the manufacturing firm, the total cash wages and earnings paid to man-ufacturing employees (bil. yen), the cost of raw materials and utilities (bil. yen), the value of manufactured goods shipped (bil. yen), the gross value added of manufacturing firms (mil. yen), and the end-of-year value of capital assets owned by manufacturing firms (bil. yen).11 _{I used}

these as outcome variables to measure the status of the manufacturing industry within coastal municipalities.

It should be noted that I used nominal value for wage, unit sales, gross value and capital assets instead of real value. I understand that I should use the real value for them theoretically. In order to do this, I need data of price index for each city, which is not available. Thus, it might be one idea to use preternatural level price index, which is available to calculate city level real value. However, the problem remains, since the price level and its change are not identical for the cities in the same prefecture. The Japanese overall inflation rate in the sample periods (2008-2014) was -1.35 2.76%. Thus, the difference in price level and inflation rate among cites are likely to be small. For this reason, I expect that real value calculated using prefecture level price index does not necessarily have advantage in estimation over the nominal values. Thus, I decided to use the nominal value for total cash wages and earnings, the value of manufactured goods shipments, the gross value added, and the value of capital assets.

The second data source is the Japanese Statistics Bureau which contains data from government institutions such as the Ministry of Health, Labor and Welfare, the Ministry of Economy, Trade,

9_{Census of Manufacturing (JP) can be accessed at http://www.meti.go.jp/english/statistics/tyo/kougyo/index.html}

(English)

10_{However, manufacturing firms located within areas that affected the survey process and whose headquarters}

had two or more establishments received the survey directly from METI

11_{Figures for 2011 were obtained by aggregating the results of the 2012 Economic Census for Business Activity}

which contains manufacturing firms with four or more employees, are not only engaged in administrative and ancillary economic activities, and gave figures on the value of manufactured goods shipment by commodity.

(19)

and Industry, and Ministry of Agriculture, Forestry and Fisheries of Japan. The Japanese Statistics Bureau’s data set contains a number of variables that I used to control for the observable charac-teristics of coastal municipalities. This set of variables includes the population density (people per hectare) of coastal municipalities from the Population Census conducted by the Japanese Ministry of Health, Labor, and Welfare, the percentage share of the agricultural industry of the economy in terms of the number of firms in that industry from the Japanese Ministry of Agriculture, Forestry, and Fisheries, and the percentage share of the manufacturing industry of the economy in terms of the number of firms within the industry from the Census of Manufacturing conducted by the Japanese Ministry of Economy, Trade, and Industry.

I constructed a balanced municipal-year panel set which contained data for the period between 2008 and 2014. I collected data on 508 coastal municipalities that had direct access to the sea within Japan under the assumption that coastal communities share similarities regardless of their geographical location. Of these 508 coastal municipalities, 474 (93%) coastal municipalities were located in outside of the three prefectures that were severely damaged by the March 2011 Tsunami. The remaining 34 (7%) coastal municipalities were affected by the Tsunami that occurred on March 2011. The final data set contains 508 observations over a period of 7 years from 2009 to 2014 resulting in a total number of 3,556 observations.

Table 2.1: Descriptive statistics of Japanese coastal municipalities in 2010

Note: ∗p < 0.1, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.01. Wages and earnings, value of manufactured shipments, gross value added, and value of capital assets are measured in billions of YEN. Treated coastal municipalities are located within the prefectures of Iwate, Miyagi, and Ibaraki.

Table 2.1 presents the descriptive statistics (mean, mean differences, and standard errors) of all-, treated, and control coastal municipalities across Japan prior to the impact of the Great East Japan Earthquake and Tsunami in 2011. The results presented in Table 2.1 indicate that the state of the manufacturing industry in 2010 does not differ greatly in terms of statistical significance. Though all of the manufacturing characteristics are reported as insignificant, when examining the absolute difference between treated- and control coastal municipalities, the control group has a workforce when examining the difference in the number of manufacturing firms and manufacturing employees. In terms of the performance of the manufacturing industry between both groups, the

(20)

value of manufactured shipments, the gross value added, and value of capital assets differs by a mean value of 8,832.90, 169.39, and 922.83. Therefore, the financial state of manufacturing firms within the control group has as a whole a better position than those in the treated group.

As for the municipality characteristics of the coastal municipalities, the share of the manufactur-ing industry is reported as bemanufactur-ing statistically significant indicatmanufactur-ing that the coastal municipalities in the control group have an manufacturing industry that is a larger share of the economy in comparison to the treated group. This discrepancy could explain the differences in the before mentioned manufacturing industry characteristics.

2.5 Results

In this section, I will estimate the impact and fade-out of the treatment effect on coastal munic-ipalities struck by the Great East Japan Earthquake and Tsunami in Iwate, Miyagi, and Ibraki prefecture using difference-in-differences with propensity score matching. The initial impact of the Great East Japan Earthquake had a significant impact on the manufacturing industry on coastal municipalities along the Northeastern coastline. However, using Autor (2003), the yearly interaction terms revealed that the basic difference-in-differences model was biased as it showed significance in the pre-treatment period, which violates the parallel trend assumption. Therefore, I use propensity score matching to match the affected coastal municipalities with non-affected coastal municipalities that are similar based on their observable municipal characteristics before the impact of the natural disaster.

The first step of the balancing procedure is to create the propensity score specification and choosing the matching algorithm. The outcome variable used is the generated gross value of manufacturing firms with the number of employees working within the manufacturing industry, percentage share of the municipal industry in agriculture, the percentage share of the municipal industry in manufacturing, municipality’s population density, the total cash wages and earnings paid to employees in manufacturing firms, the value of manufactured goods shipped, and the value of the capital assets owned by manufacturing firms as independent variables.

I used several matching algorithms such as pair matching with and without replacement, kernel, and nearest-neighbor matching up to 4, 6 and 8 neighbors with a caliper size of 0.01 and bandwidth size of 0.06 for kernel matching. The results presented in this study were based on the matching result using the nearest-neighbor matching algorithm up to 8 neighbors with a caliper size of 0.01.12 12_{Matching results were consistent across other matching algorithms.} _{Best matching results came from the}

(21)

Table 2.2: Propensity score balancing test

Note: ∗p < 0.1, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.01. U = Unmatched sample, M = Matched sample. Gross value is the dependent variable. Nearest-Neighbor matching algorithm with 8 neighbors and a caliper size of 0.01

As outlined by Rosenbaum and Rubin (1983) it is important that the bias difference between the treatment and control observations should ideally be around 5% to ensure strong ignorability and 10% for weak ignorability. Table 2.2 shows the matching results of using nearest-neighbor matching up to 8 neighbors with a caliper size of 0.01. Examining Table 2.2, it is evident that the matching procedure has resulted in a better fit between the treatment and control group in terms of the bias difference. In addition, all except one variable (agricultural percentage share of the municipal industry composition) is reported as being under or around the threshold of 5%. This is further confirmed when examining the result of the t-test on whether the treatment and control group is significantly different. None of the covariates are reported as being significantly difference indicating that the balancing test was successful. This left a sample of 34 treatment-and 174 control observations across 7 years from 2008 to 2014 with a total of 1,456 observations.

(22)

Table 2.3: Effects of the Tsunami on Manufacturing Industries in Affected Coastal Municipalities (PSM-DID)

Note: ∗p < 0.1, ∗ ∗ p < 0.05, ∗ ∗ ∗p < 0.01.. Clustered standard errors were grouped on municipal level and given in parenthesis. Treatment effect of the natural disaster is captured by Equation 3.1. Wages and earnings, value of manufactured shipments, gross value added, and value of capital assets are measured in billions of YEN

Table 2.3 presents the treatment effect of the natural disaster using the basic difference-in-difference estimator of the difference-in-differences with propensity score matching model. Column 1 presents the number of manufacturing firms as the outcome variable and is reported as insignificant indicat-ing that the Tsunami had no long-lastindicat-ing effects on the number of firms. Even though the Tsunami was reported to have significantly disrupted the activities of firms within affected municipalities. However, the coefficient reported in Column 1 is reported as insignificant possibly indicating that firms within affected municipalities were able to overcome the treatment effect of the Tsunami through the aid of government assistance by replacing damaged tangible assets.

Column 2 presents the number of manufacturing employees within affected coastal municipal-ities as the outcome variable and is reported as strongly significant indicating that as of 2014, the manufacturing firms within affected municipalities had a smaller labor force in comparison to the control group. Factors that may have influenced the recovery rate of human capital within affected municipalities is the reconstruction of infrastructure and the labor market situation within affected regions. Considering the intensity and lethality of the Great East Japan Earthquake and Tsunami on the regional economy and human capital, it might take more time for affected coastal municipalities to recover to pre-disaster levels due to lack of human capital to expand current firm activities.

Column 3 presents the total cash wages and earnings paid to employees as the outcome variable and is reported as insignificant. The definition of the total cash wages and earnings paid to employees includes payment towards temporary workers that are working for manufacturing firms. Therefore, the coefficient in Column 2 on the number of employees and the coefficient in Column 3 are closely correlated. However, because the coefficient is reported as insignificant this could potentially indicate that manufacturing firms relied on a temporary work force to continue firm operations.

Column 4 presents the value of goods shipped (sales) as the outcome variable and reports an insignificant positive coefficient. Even though the coefficient is not reported as significant, the fact that it is positive would indicate that the sales of manufacturing firms have recovered from the treatment effect of the natural disaster. Dependent on the type of manufacturing industry, the damage may have had a different impact on the manufacturing activities and inventory in 2011. If the inventory or manufacturing process itself was not heavily affected by the Great East Japan Earthquake and Tsunami, then the manufacturing firms could still sell their remaining inventory to continue manufacturing operations even in a limited capacity. In addition, the sales performance of manufacturing firms does partially rely on the firm’s supply chain and relative position within that

(23)

chain. Should the larger clients of the manufacturing firm be heavily damaged, it would require the firm to find alternative sales channels, which might be a difficult task to accomplish after the initial impact in 2011.

Column 5 presents the generated gross value of manufacturing firms as the outcome variable and is reported as significantly affected by the natural disaster. Comparing the results presented in Column 4 which indicates that the sales of manufacturing firms has recovered, this is not represented in the coefficient of generated gross value seen in Column 5. Possible indication is that the manufacturing firms in affected firms might have additional expenses that have reduced the overall profitability. However, without further information on the profit-and-loss statement of manufacturing firms, this is only an assertion.

Column 6 presents the value of capital assets owned by firms as the outcome variable as is re-ported as being statistically insignificant indicating that the Tsunami did not have a lasting impact on the amount of assets accumulated since 2011. Prior studies on this subject have highlighted that the Tsunami did have an significant impact on the capital assets of manufacturing firms. Uchida et al. (2015) argues that one of the reasons for the quick recovery of the capital assets is due to the intervention of governmental institutions by creating funds such as SME Group Fund and Tohoku Small Business Recovery Program that aimed at easing the financial requirements and access to capital for affected firms. The availability of these funds for affected firms would have likely proven to be a strong factor in the recovery process in terms of capital assets.

Table 2.4: State of Manufacturing Industries Before and After the Impact of the Tsunami in Affected Coastal Municipalities (PSM-DID)

Note: ∗p < 0.1, ∗∗p < 0.05, ∗∗∗p < 0.01. Clustered standard errors are given in parenthesis. Treatment effect of the natural disaster is captured by Equation 2.1. Wages and earnings, value of manufactured shipments, gross value added, and value of capital assets are measured in billions of YEN