Embed Size (px)
Citation preview
1
Empirical analysis toward resilient and adaptive local
economy: Evidence from Japan
September 2021
Keisuke Takano
2
Empirical analysis toward resilient and adaptive local
economy: Evidence from Japan
Graduate School of Systems and Information Engineering
University of Tsukuba
September 2021
Keisuke Takano
3
Table of Contents
1. Introduction .......................................................................................................................................................... 7 1.1 Preface ........................................................................................................................................................... 7 1.2 Framework .................................................................................................................................................... 9
1.2.1 Basic Notions of Regional Economic Resilience .................................................................................. 9 1.2.2. Approaches for Analyzing Regional Economic Resilience and Adaptivity ...................................... 13
1.2.2 1. Equilibrium Approach ................................................................................................................ 14 1.2.2.2. Non-equilibrium Approach ........................................................................................................ 23
1.2.3. Policy Debate for Regional Economic Resilience ............................................................................. 31 1.3. Organization of This Thesis ....................................................................................................................... 36
1.3.1 Main Issues .......................................................................................................................................... 36 1.3.2 Outline of Essays ................................................................................................................................. 38
2. Firms’ Action: Supply Chain Diversification .................................................................................................... 42 2.1 Introduction ................................................................................................................................................. 42 2.2 Literature Review ........................................................................................................................................ 45
2.1.1 Natural Disasters and Firms’ Activities .............................................................................................. 45 2.2.2 Business Portfolio under Uncertainty ................................................................................................. 46
2.3 The Great East Japan Earthquake and the Nankai Trough Earthquake ...................................................... 49 2.4 Methodology ............................................................................................................................................... 51
2.4.1 Empirical Procedure ............................................................................................................................ 51 2.4.1.1 Conceptual Framework ............................................................................................................... 51 2.4.1.2 Estimation Methods ..................................................................................................................... 52
2.4.2 Interfirm Transaction Data .................................................................................................................. 55 2.5 Results ......................................................................................................................................................... 58
2.5.1 Propensity Score Matching ................................................................................................................. 58 2.5.2 Regression Results with DD ............................................................................................................... 62 2.5.3 Results with DDD ............................................................................................................................... 65 2.5.4 Discussion ........................................................................................................................................... 68
2.6 Conclusion .................................................................................................................................................. 70
3. Local Governments’ Action: Business Continuity Planning ............................................................................. 73 3.1 Introduction ................................................................................................................................................. 73 3.2 Literature Review ........................................................................................................................................ 76
3.2.1 Crisis management and emergency preparedness ............................................................................... 76 3.2.2 Policy diffusion as an implicit collaboration ....................................................................................... 78
3.3 Institutional Background ............................................................................................................................. 81 3.4 Methodology ............................................................................................................................................... 82
3.4.1 Data ..................................................................................................................................................... 82 3.4.2 Empirical model .................................................................................................................................. 83
3.4.2.1 Probit Model ................................................................................................................................ 83 3.4.2.2 Spatial Probit Model .................................................................................................................... 86
3.5 Result .......................................................................................................................................................... 88 3.5.1 Drivers and Deterrents of BCP Development ..................................................................................... 88 3.5.2 Spatial Diffusion of Municipal BCP Development ............................................................................. 91
3.5.2.1 Spatial Pattern of BCP Development Status ............................................................................... 91 3.5.2.2 Estimation Result of the Spatial Probit Model ............................................................................ 93
3.5.3 Discussion ........................................................................................................................................... 95
4
3.6 Conclusion .................................................................................................................................................. 98
4. Acquisition of Source of Resilience towards Industrial Structural Change: Place-based Innovation Policies 102 4.1 Introduction ............................................................................................................................................... 102 4.2 Literature Review ...................................................................................................................................... 104
4.2.1 Place-based Policies .......................................................................................................................... 104 4.2.2 Diversification of Economic Activities ............................................................................................. 106
4.3 Investigated Prefectures and Programs ..................................................................................................... 108 4.3.1 General Characteristics of Each Prefecture ....................................................................................... 108 4.3.2 Overview of R&D Support Programs ............................................................................................... 109
4.3.2.1 Decentralization of Japanese Innovation Policy ........................................................................ 109 4.3.2.2 Local R&D Subsidies in the Three Prefectures ......................................................................... 109
4.4 Methodology ............................................................................................................................................. 112 4.4.1 Firm-level Panel Data ........................................................................................................................ 112 4.4.2 Empirical Procedure .......................................................................................................................... 112
4.4.2.1 Conceptual Framework ............................................................................................................. 112 4.4.2.2 Definition of Outcomes ............................................................................................................. 113 4.4.2.3 Estimation Methods ................................................................................................................... 115
4.5 Results ....................................................................................................................................................... 119 4.5.1 Descriptive Statistics ......................................................................................................................... 119 4.5.2 Regression Results ............................................................................................................................ 120
4.5.2.1 Policy Effects on Industry Diversification of Customer Composition ...................................... 120 4.5.2.2 Policy Effects on the Regional Diversification of Customer Composition ............................... 120
4.5.3 Discussion ......................................................................................................................................... 126 4.6 Conclusion ................................................................................................................................................ 130
5. Consequences of Local Policy for Postwar Reconstruction: SME Financial Policy ....................................... 133 5.1 Introduction ............................................................................................................................................... 133 5.2 Literature Review ...................................................................................................................................... 135
5.2.1 SME Finance and Local Industrial Development ............................................................................. 135 5.2.2 SME Finance for Local Industrial Reconstruction ............................................................................ 136
5.3 Institutional Background ........................................................................................................................... 139 5.3.1 Industrial Characteristics of Postwar Osaka ...................................................................................... 139 5.3.2 Problems around SMEs during Postwar Period ................................................................................ 143 5.3.3 Postwar SME Finance Policies ......................................................................................................... 144 5.3.4 Modernization Fund by Osaka Prefecture ......................................................................................... 145
5.4 Methodology ............................................................................................................................................. 149 5.4.1 Hypotheses ........................................................................................................................................ 149 5.4.2 Econometric Methods ....................................................................................................................... 149 5.4.3 Data ................................................................................................................................................... 150
5.5 Results ....................................................................................................................................................... 153 5.5.1 Baseline ............................................................................................................................................. 153 5.5.2 Vestige of Wartime Economies ......................................................................................................... 154
5.5.2.1 Prosperity and Demise of Osaka Arsenal .................................................................................. 154 5.5.2.2 Empirical Framework ................................................................................................................ 156 5.5.2.3 Results ....................................................................................................................................... 157
5.5.3 Industrial Heterogeneity .................................................................................................................... 158 5.6 Discussion and Conclusion ....................................................................................................................... 161
6. Conclusion ....................................................................................................................................................... 164 6.1 Summary of Findings ................................................................................................................................ 164 6.2 Policy Implications ................................................................................................................................... 167
5
6.2.1 Network and Interdependence ........................................................................................................... 167 6.2.2 Determinants of (sources of) regional economic resilience .............................................................. 168 6.2.3 Role of local government and its policies ......................................................................................... 169
6.3 Limitations and Future Work .................................................................................................................... 170
Acknowledgments ................................................................................................................................................ 174
References ............................................................................................................................................................ 176
Appendix .............................................................................................................................................................. 189 Appendix for Chapter 3 ................................................................................................................................... 190
A: Different specification of spatial weight matrix .................................................................................... 190 Appendix for Chapter 4 ................................................................................................................................... 191
A: Covariate balance in PSM ..................................................................................................................... 191 A.1 Estimation results of logistic regression to predict propensity scores ............................................ 193 A.2 Covariate valance before/after PSM ............................................................................................... 196
B: Policy effects on total factor productivity ............................................................................................. 199 B.1 Methodology ................................................................................................................................... 199 B.2 Policy Effects on TFP ..................................................................................................................... 199
C: Policy effects on the number of customers in the Greater Tokyo District ............................................ 201 D: Descriptive statistics of the panel data .................................................................................................. 202 E: Goodman-Bacon decomposition ........................................................................................................... 205
Appendix for Chapter 5 ................................................................................................................................... 209 A: Descriptive Statistics ............................................................................................................................. 209 B: Empirical Analysis Using No. of Employees as Outcome .................................................................... 211 C: Propensity Score Matching .................................................................................................................... 215 D: Alternative Specification for the Parallel Trend ................................................................................... 230 E: Effect on Standardization ...................................................................................................................... 234 F: Matching between the IDBC and Borrower List ................................................................................... 239
6
Chapter 1
Introduction
7
1. Introduction
1.1 Preface
In this decade, the local economy has witnessed various catastrophic disasters and
economic crises. In Japan, as in other developed countries, the Great Recession at the end of
the 2000s caused a persistent economic downturn, and the impacts on employment and
consumption level were quite severe. In addition, the frequent occurrence of severe natural
disasters has revealed the vulnerability of the Japanese local economic environment. The most
severe disaster was the Great East Japan Earthquake, 3.11, in 2011; the unprecedented supply
chain disruption caused by 3.11 exposed the vulnerability of complex and intensive sourcing
structures developed for the pursuit of a comparative advantage. In 2020, we are still
experiencing the disruption caused by the pandemic outbreak of COVID-19. This outbreak has
made the world and local economy dysfunctional by restricting the mobility of the labor force,
human capital, and goods. Although the spatial agglomeration of population, production, and
consumption serves as a driver of urban economic growth in peacetime, this advantage turns
out to be a source of a health crisis in this outbreak. More seriously, due to the progress of
globalization, a shock in a specific region can spread with ease, and it can no longer be
considered a “fire on the other side of the river.” These catastrophic events we have experienced
in this decade are cruel enough to show us that there is no uniform growth strategy to sustain
the local economy and that we have to continue to update strategies and to find a new normal
every time we face a difficulty. In addition, a shock is not necessarily a sudden event. We
should also pay careful attention to a slow-burning shock, such as a technological change and
a change in the product cycle.
To properly cope with various shocks jeopardizing the local economy, we need to
understand how a shock affects regional economic agents and systems, how the local economy
bounces back or adapts to a new environment, and how each agent can contribute to the
8
sustainment of the economic system. In other words, we have to accumulate knowledge about
regional economic resilience and adaptivity. Throughout this thesis, I regard resilience as a
series of capabilities to reduce losses from shocks and to adapt if necessary to the post-shock
environment. Based on this notion, this thesis aims to empirically examine three main issues:
(i) the role of networks and the interdependence between regions and sectors in developing
regional economic resilience, (ii) the determinants of (sources of) regional economic resilience,
and (iii) the role of local government and its policies in developing (sources of) regional
economic resilience. Despite the maturity of the conceptualizations of regional economic
resilience in this decade in the field of regional science, there is still a gap between developed
notions in their application of it to real local economies, particularly to regional economic
policies. I attempt to fill this gap by conducting several empirical investigations based on an
impact evaluation and spatial analysis focusing on cases of shocks in Japan, including
(projected) natural disasters and postwar disruptions. I organize this thesis as follows. The
remainder of Chapter 1 describes the basic framework and notions of regional economic
resilience and reviews seminal empirical investigations. After that, I provide the main issues in
this thesis by associating these issues with the results of the literature review. From Chapters 2
to 5, I illustrate original empirical investigations addressing these issues. Chapter 6 concludes
this thesis by summarizing the key results obtained from my examinations, deriving policy
implications, and presenting the limitations and future work.
9
1.2 Framework
1.2.1 Basic Notions of Regional Economic Resilience
The recent occurrence of the Great Economic Depression and devastating disasters has
spurred in regional science a discussion about the resilience of the local economic system.
Particularly, the Great Recession in the late 2000s provided researchers in regional science
(economic geographers in particular) opportunities to conceptualize and define local economic
resilience and examine the impacts of the recession. The fact that the impacts were spatially
heterogeneous brought some new insights in interpreting and revealing from place-based
perspectives the sources, consequences, and systems of resilience. Beginning with the
Cambridge Journal of Regions, Economy and Society’s special issue that covered seminal
studies conceptualizing regional resilience (e.g., Simmie and Martin, 2010; Pike, Dawley, and
Tomaney, 2010; Pendall, Foster, and Cowell, 2010) and through examining the role of
resilience based on case studies (e.g., Treado, 2010; Chapple and Lester, 2010), the
conceptualization of the notions of regional resilience through the resolution of their ambiguity,
the coordination between conflicting concepts, and the clarification of the complementary
relationship between the notions has been conducted in this decade. Following the special
edition, several influential special issues evolving the concept of regional resilience have also
been organized in the Regional Studies (“Regional Economic Resilience” in 2014 and
“Resilience Revisited” in 2016). In particular, the contribution by Ron Martin and his
colleagues has greatly impacted this literature in that these authors have attempted to integrate
concepts, such as ecology and system engineering, scattered across a number of fields,
including regional science. Centering on the pioneering work of Martin and Sunley (2015), I
briefly describe the concepts of regional resilience and adaptivity. In explaining the concepts,
I frequently contrast economic geography and new economic geography (NEG) as disciplines
that more or less address the notion of regional economic resilience or stability.
10
Table 1.1 Interpretations or definitions of regional economic resilience (Source: Table 1 in Martin and Sunley, 2015)
Definition/type Interpretation Main fields of use Resilience as a ‘bounce back’ from shocks
System returns, ‘rebounds’ to pre-shock state or path: emphasizes speed and extent of recovery.
So-called ‘engineering resilience’, found in physical sciences, some versions of ecology; akin to ‘self-restoring equilibrium dynamics’ in mainstream economics?
Resilience as the ‘ability to absorb’ shocks
Emphasizes stability of the system structure, function and identity in the face of shocks; the size of shock that can be tolerated before system moved to new state/form.
So-called ‘extended ecological resilience’; found in ecology and social ecology; akin to multiple equilibrium economics?
Resilience as a ‘positive adaptability’ in anticipation of or in response to shocks
Capacity of a system to maintain core performances despite shocks by adapting its structure, functions and organization. Idea of ‘bounce forward’.
Found in psychological sciences and organizational theory; akin to ‘robustness’ in complex systems theory; can be linked with evolutionary economics?
For example, Martin and Sunley (2015) defined the resilience of the local economy as
“the capacity of a regional or local economy to withstand or recover from the market,
competitive and environmental shocks to its developmental growth path, if necessary by
undergoing adaptive changes to its economic structures and its social and institutional
arrangements, so as to maintain or restore its previous developmental path, or transit to a new
sustainable path characterized by a fuller and more productive use of its physical, human and
environmental resources.” The remarkable point of their definition is that it includes not only
the capability to bounce back to the pre-shock state of equilibrium but also the ability to adapt
to another state. This characteristic of the definition that covers various post-shock possibilities
not limited to bouncing back can be commonly seen in the recent literature about regional
resilience. According to Meerow, Newell, Stults (2016), who systematically reviewed the
concepts of regional resilience by utilizing a bibliometric analysis on 172 papers, 15 out of 25
definitions uniquely observed in these papers adopted the definition covering the adaptation
and adaptivity to another state.
As shown in Table 1.1, the researchers of regional economic resilience (e.g., Martin and
Sunley, 2015; Evenhuis, 2017; Martin, 2018) suggest three main interpretations or definitions
of the concept. Under the first definition, economic resilience is defined as the ability to bounce
11
back a system to its pre-existing state or path (equilibrium state or path in other words); this
definition focuses on the system’s speed of recovery or return to its pre-shock position and is
similar to the notion of a unique equilibrium characterized by the idea of self-restoring
dynamics found in mainstream economics (e.g., Brakman, Garretsen, and van Marrewijk,
2019; Nunn, 2014, related to the field of spatial economics). With Holling (1973) as the point
of departure, the first definition has been adopted in a wide variety of academic fields, such as
civil engineering (e.g., the recovery of a structure such as a building or bridge to its baseline
after it is damaged), psychology (e.g., efficient action against traumatic experience), disaster
management (e.g., the recovery of a system essential for daily life after a catastrophic disaster),
and so on (Zolli and Healy, 2012; Matyas and Pelling, 2015). Due to this emphasis on
efficiency, constancy, and predictability, the first definition is also called engineering resilience.
The second definition, also termed ecological resilience, is defined as the capability of a
system to absorb a shock without changing its structure, identity, and function (Holling 1973;
Holling 1986). In this definition of resilience, it is assumed that a system changes its state to
alternative equilibrium if a shock on it is beyond a certain tolerance of absorption capacity (the
ability to bounce back in other words). As is the case in the first definition, this definition shares
its characteristics with the concept in economics of multiple equilibria; that is, a shock that is
too severe forces the local economy to move its state to a new equilibrium or path by drastically
changing economic structure, behaviors, and expectations (e.g., Chang, van Marrewijk, and
Schramm, 2015). I should, however, note that the notion of multiple equilibria is not
necessarily supported by geographers and economists. For example, considering equilibria in
spatial economic models, Metcalfe, Foster, and Ramlogan (2006) mentioned that under an
evolutionary perspective, it can be unrealistic to assume the existence and form of alternative
equilibria a priori, as the economy is a historical and contingent process. Additionally, Head
and Mayer (2004) remarked that the notion of multiple equilibria should be considered
12
fascinating theoretical “exotica” rather than a relevant element of actual economic geography.
Under the third definition, economic resilience is defined as the ability of a system to
tackle a shock by undergoing a plastic change in some aspects of its structure and components
to maintain or restore certain core performances or functionalities. In this regard, unlike the
first and second definitions (Evenhuis, 2017), the third definition does not necessarily premise
any equilibrium or stable state, but assumes a certain dynamic process that transforms a
system’s structure and components for adaptation (e.g., Pendall, 2010). In the recent literature
on urban resilience, the conceptualization based on the non-equilibrium perspective is slightly
superior (Meerow et al., 2016).
Despite the difference between these three notions of regional resilience, these definitions
are neither mutually exclusive nor a substitute and fall within a certain spectrum. Each notion
of regional economic resilience (partially) corresponds to a process including several mutually
related components, such as vulnerability, shocks, resistance, robustness, and recovery (Martin,
2012), as shown in Figure 1.1.
Figure 1.1 Regional economic resilience as a process
(Source: Figure 1 in Martin and Sunley; 2015)
13
Additionally, each notion has an individual advantage in analyzing different shocks. As
suggested in Evenhuis (2017), the notion of engineering resilience can be applied to analyze
the impacts of (relatively small) emergency and macroeconomic change, that of ecological
resilience can be suitable to understand the effects of catastrophic events or alternations taking
place under a recession, and the notion of evolutionary resilience can be applied to examine
the process of the local economy’s adaptation to economic structural change.
Before reviewing specific approaches to analyze resilience and adaptivity, as pointed out
in Meerow et al. (2016), it is necessary to note the conceptual tension in the distinction between
short-term adaptation and long-term adaptability (adaptivity). According to Pike et al. (2010),
adaptation is a movement, characterized by a strong and tight coupling between social agents
in place, towards a pre-conceived path (i.e., a high level of specialization), while adaptability
is the dynamic capacity to effect and unfold multiple evolutionary trajectories through a loose
and weak coupling between the agents.
1.2.2. Approaches for Analyzing Regional Economic Resilience and Adaptivity
Though a variety of theories conceptualize and analyze regional economic resilience, the
source and role of resilience and the interpretation of resilience and its implications are quite
different. Martin (2018) mentioned five primary theories addressing resilience: new economic
geography theory, geographical political economy, evolutionary-Schumpeterian theory, path-
dependence theories, and institutionalist approaches. In the following sections, I review the
basic theories or concepts in each field and the empirical literature focusing on the NEG theory
and on the evolutionary theory particularly relevant to my empirical analysis in this thesis.
According to the NEG theory, a shock stems from the (exogenous) change in transportation
costs (terms of trade more generally) and/or the labor distribution between regions. The
resilience of a system is evaluated by the stability of the equilibrium. Particularly, a key
14
question is whether a location (city, region, country) returns to an initial stable equilibrium
without moving to another equilibrium. The examination of resilience in this field is often
termed shock sensitivity (Brakman et al., 2019). In evolutionary theory, a shock is associated
with occasional historic shifts in the technological regime, and resilience might equate with the
adaptability to new technologies and industrial transformations (Martin and Sunley, 2015).
1.2.2 1. Equilibrium Approach
The basic idea of spatial economics is to explain the mechanism of agglomeration based
on the following: increasing returns to scale and monopolistic competition in production; love
of variety in consumption; and the existence of transportation costs in inter-regional trade. The
analytical framework of spatial economics relies on a general equilibrium analysis. In this
section, I first review the spatial economics theoretical properties relevant to shock sensitivity.
After that, I briefly explain several empirics that directly or indirectly test the theoretical
prediction.
Following Fujita, Krugman, and Venables (1999), Brakman, Garretsen, and Van
Marrewijk (2009), and Brakman and van Marrewijk (2015), I non-technically outline the core-
periphery (CP) model of spatial economics developed by Krugman (1991). In Figure 1.2, I
show the structure of the CP model, in which there are two regions (regions 1 and 2) and two
sectors (manufacturing and food). Manufacturing workers are mobile between regions, but
farm workers are not. Each region has both manufacturing and farm workers. The
manufacturing sector consists of "! firms in region 1 and "" firms in region 2. One
manufacturing firm produces one differentiated good (variety), whereas farm workers produce
a homogeneous good. Thus, "!varieties are produced in region 1. In this regard, each firm
has monopolistic power and can determine the price of its product on its own. Farm workers
produce under constant returns to scale and perfect competition, whereas manufacturing
15
workers produce under increasing returns to scale. In the manufacturing sector, due to internal
economies of scale, the average cost decreases as the production level increases. Both sectors
use labor as an input. Transportation costs arise only in the inter-regional trade of
manufacturing goods.
Figure 1.2. Structure of two-region CP model (Source: Figure 3.1 in Brakman, et al., 2009)
Consumers have a Cobb-Douglas form preference and spend their income # on
manufactures $ and food % . In addition, the partial utility of manufacturing goods is
represented by the CES function. Thus, utility maximization can be written as follows:
max#,%
) = %!&'$' , 0 < . < 1;
$ = 123()
*
(+!4
! )⁄
, 0 < 5 < 1;
6. 8. # = 9$ + %,
(1.1)
where 3( is the consumption of variety ; and 9 is the price index of manufacturing goods.
Since food is used as the numéraire, the price is 1. From the property of a Cobb-Douglas
16
preference, a consumer spends a constant share.of income on manufactures (callout b in
Figure 1.2) and 1 − .on food (callout a). Thus, the demand on variety ; can be derived below
(callout c), where =(!&- is the price of variety ;.
3( = =(&-[9-&!.#], @ =
1
1 − 5, 9 = A2=(
!&-*
(+!B
! (!&-)⁄
. (1.2)
A fraction1 − Cof total laborDworks in the food sector. Due to the assumption of
constant returns to scale and perfect competition, production in the food sector equals food
employment; i.e., % = (1 − C), and the workers’ wage is 1. Production in the manufacturing
sector can be defined asG( = H + IJ( to simply express the increasing returns to scale, in which
G( is the amount of labor necessary to produceJ( units of variety; (callout d and e). In the
profit maximization assumingK = =J −L(H + IJ)under the assumption of monopolistic
competition and a zero-profit condition due to free entry, the price of a variety=is equal
toIL 5⁄ , the production levelJis equal toH(@ − 1) I⁄ , and the number of firms is equal to
CD H@⁄ (∵ there areH@manufacturing workers). Note that this holds in all firms in the same
region if a symmetry in production technology is assumed.
Based on the setting above, different locations are introduced. Transportation costs take
the form of an iceberg represented with parameterN ≥ 1. Ndenotes the number of goods that
need to be shipped to ensure that 1 unit of a variety of manufactures arrives per unit of distance
(callout f). In addition, in each sector and region, labor is distributed as defined in Figure 1.3.
Figure 1.3. Labor distribution in the two-region CP model (Source: Figure 3.6 in Brakman, et al., 2009)
17
Denoting the food production in the region, there areP!(1 − C)Dfarmers in the food
sector in region 1. Due to transportation costs, the wage income paid to manufacturing workers
in region 1 will be different from that in region 2. To identify the difference, using a subindex,
the manufacturing wage in region 1 is denoted byL!and that in region 2 byL".
In advance of the derivation of the equilibrium characteristics relevant to shock sensitivity,
I briefly describe the properties of several key variables in the short-run and long-run
equilibrium. Since the price of a variety produced in region 2 and consumed in region 1
equalsIL" × N 5⁄ , the price index of manufacturing goods produced in 1 is written as follows:
9! = RI
5S RCD
H@S! (!&-)⁄
[T!L!!&- + T"N!&-L"
!&-]! (!&-)⁄ . (1.3)
Thus, the price index in region 1 can be represented with a weighted average of the price of
locally produced goods and that of the imported goods from region 2.
In short-run equilibrium, the distribution of manufacturing labor is given (exogenous).
Because of the zero-profit assumption in both the manufacturing and food sectors, all income
earned in the economy equals the wages earned in the sectors. SinceP!(1 − C)Dfarm workers
earn at a wage rate 1 andT!CDmanufacturing workers earn at a wage rateL!in region 1, the
total income generated in region 1 equals to #! = T!CDL! + P!(1 − C)D . Equating total
demand to the supply derives the short-run equilibrium wage. This also holds in region 2.
L! = 5I&) R.
(@ − 1)HS! -⁄
[#!9!-&! + #"N!&-9"
-&!]! -⁄ . (1.4)
In the long-run equilibrium, factor (labor) mobility in the manufacturing sector arises,
which implies that the short-run equilibrium changes. In the CP model, it is assumed that
mobile workers react to differences in real wages in each regionU, V0 = L0 90&'⁄ . If the real
wage for manufacturing workers is higher in region 1 than in region 2, it is expected that they
will leave region 2 and settle in region 1. The reverse will happen if the real wage is higher in
region 2.
18
Figure 1.4. Short-run equilibrium relative real wage and (un)stable long-run equilibria
(Source: Figure 4.1 in Brakman, et al., 2009)
Using the simulation results1, I note when a long-run equilibrium is reached. Figure 1.4
shows how in region 1, the relative real wage (V! V"⁄ ) based on the short-run equilibrium
varies as the share of the mobile workforce in region 1 (T!) varies (the transportation cost is set
asN = 1.7). In this graph, three types of long-run equilibrium can be observed. The first is
spreading equilibrium in which the manufacturing workers are distributed equally between
regions (C). The second is complete agglomeration equilibrium in which all manufacturing
workers are concentrated in one region (A and E). The final one is partial agglomeration
equilibrium (B and D). Note that the partial agglomeration equilibria are unstable and only
arise in the simulation. Suppose thatT!equals F. At this point, the share of manufacturing labor
is larger in region 2 despiteV! V"⁄ > 1. In this case, manufacturing workers have an incentive
to move to region 1, and they immigrate until T! reaches C. In this regard, the spreading
equilibrium is stable. We can similarly check the stability for complete agglomeration.
Although V! V"⁄ equals 1 at B and D, the coordination towards stable equilibria occurs once
we slightly moveT!from these points. Thus, the unstable equilibria B and D serve as certain
thresholds between spreading and complete agglomeration equilibria.
1 Under the setting of Krugman (1991), the long-run equilibrium cannot be derived analytically because of nonlinearity. However, the solvable versions of the CP model have been developed by other researchers, for example, Forslid and Ottaviano (2003).
19
Figure 1.5. Tomahawk diagram (Source: Figure 4.3 in Brakman, et al., 2009)
Figure 1.5 summarizes the results corresponding to various values of the transportation
cost N . Because of its shape, this figure is known as the Tomahawk diagram. If N is
sufficiently low, a complete agglomeration of manufacturing activity either in region 1 or in
region 2 will occur. For the area between the sustain point (Y1andY!) and the break point (Z),
both spreading and complete agglomeration are stable, and remarkably, which equilibrium
eventually becomes stable depends on the path reaching equilibrium. In the context of the
spatial economic model, this property is termed path-dependency. Notably, in the CP model,
the transition from spreading equilibrium to complete agglomeration equilibria is catastrophic
in the sense that the model does not have partial agglomeration equilibria as stable equilibria.
However, these unique properties, namely, catastrophic change and path dependency, are not
always observed in other types of spatial economic models. For example, the bell-shaped
diagram was derived from Puga’s (1999) spatial economic model with decreasing returns in
the food sector and without inter-regional labor mobility. In another example, the bell-shaped
curve also emerges from the model with a spreading force that does not weaken when N falls
(e.g., the existence of a non-traded good). As an important example, deriving a (half-cut) bell-
shaped curve, Helpman (1998) developed a model with housing goods and sectors.
20
Since the early 2000s, the empirical literature has tested the theoretical prediction outlined
above by utilizing historical natural experiments2 to causally identify the effects of a shock on
regional economic resilience. The first empirical investigation testing shock sensitivity, as
illustrated in Figure 1.4, was conducted by Davis and Weinstein (2002). Their main question
is whether city size is robust even against large temporary shocks. To empirically answer this
question, they exploited the allied bombings of Japanese cities during WW2. Their empirical
analysis based on the instrumental variable method showed that cities returned to pre-bombing
population sizes. In this regard, they implied the existence of a unique stable equilibrium rather
than multiple equilibria. Similarly, Miguel and Roland (2011) addressed the question by
exploiting the exogenous shock brought by the US bombing on Vietnam, and they hardly found
evidence of the existence of multiple equilibria. Conducting an empirical investigation
focusing more on local shock, Hornbeck and Keniston (2017) tested the robustness of the intra-
urban economic system by exploiting the Great Boston Fire of 1872. They found a convergence
between burned and unburned plots within Boston.
Davis and Weinstein (2008) developed an analytical framework to test a form of
equilibria (e.g., the number and location of stable/unstable equilibria) and applied it to the case
of Japanese cities after WW2 together with the introduction of an outcome related to industrial
development measured with manufacturing share. The concept of their analytical framework
is as follows. First, as shown in Figure 1.6, the dynamic version of Figure 1.4 is assumed, where
∆62is a shock that occurred in period8(change in city size because of the allied bombings in
their context), Ω is the initial stable equilibrium, ∆! and ∆3 are other stable equilibria,
and]!and]3are unstable equilibria as thresholds between stable equilibria.
2 Conducting a comprehensive review on articles about historical natural experiments, Nunn (2014) reviewed the empirics exploiting historical natural experiments for each issue in development economics, while Cantoni and Yuchtman (2020) reviewed more recent literature focusing on historical natural experiments in each subdiscipline in economics.
21
Figure 1.6. Two-period growth representation of a model of multiple equilibrium
(Source: Figure 4 in Davis and Weinstein, 2008)
Suppose that a negative shock∆62occurs around the initial equilibriumΩin this system.
If a shock is sufficiently small (∆62falls betweenΩand]!), it is completely undone in the next
period (∆62 = −∆624!), which implies that the slope of the solid line passing throughΩis 45
degrees. Once a shock is large enough to go beyond]!, however, the economy moves towards
another stable equilibrium (∆62 = −∆624! + 3^_68`_8). Davis and Weinstein (2008) derived
the order condition to empirically test the number of equilibria (three equilibria exist if a
condition ∆!< ]! < Ω < ]" < ∆3 is satisfied) and indeed found evidence supporting the
existence of a unique equilibrium by showing that any order condition assuming multiple
equilibria is violated.
Though the empirical investigations reviewed above supported the existence of a unique
equilibrium rather than multiple equilibria, depending on the context, this is not always the
case. Following the approach in Davis and Weinstein (2002), Brakman, Garretsen, and
Schramm (2004) tested the same question exploiting the bombing of West German cities and
by showing that the recovery of cities was partially contrary to Davis and Weinstein (2002),
22
found empirical evidence supporting the existence of multiple equilibria. In addition, Bosker,
Brakman, Garretsen, and Schramm (2008) empirically showed the existence of two multiple
equilibria by applying the approach in Davis and Weinstein (2008) and found that the evidence
decreased without considering spatial dependence. Conducting an empirical investigation
focusing on a different context, exploiting the example of the division and reunification of
Germany, Redding, Sturm, and Wolf (2011) found empirical evidence supporting the existence
of multiple equilibria in the location of a hub airport. Specifically, based on the difference-in-
differences method, they showed that the largest hub airport moved from Berlin to Frankfurt
when Germany was divided, and that Frankfurt Airport remained the largest even after
reunification. While the literature reviewed above is based on a reduced-form analysis, newer
literature has emerged answering the question by utilizing a structural estimation termed
quantitative spatial economics (QSE)3. As an example, Allen and Donaldson (2018) developed
a QSE model based on NEG theories and structurally evaluated the persistence of regional
shocks. Their counterfactual analysis showed that the location of economic activity in the US
today is highly sensitive to variations in geographically local historical shocks.
While most of the empirical papers reviewed above are linked to a specific theoretical
model and aim to test the properties of the theory, some literature examining shock sensitivity
has remained faithful to the original question on engineering resilience (the system’s speed of
recovery or return to its pre-shock position) rather than focusing on the theoretical model.
Fingleton, Garretsen, and Martin (2012) is milestone research in the literature in the sense that
it is a collaboration between spatial economists and economic geographers. They analyzed the
3 Conducting a comprehensive review of articles on QSE, Redding and Rossi-Hansberg (2017) provide a non-technical summary of the QSE model to implement an inter- and intra-city structural analysis. Redding (2020) additionally includes a detailed description of the structure of each building block of theoretical models and identification strategies regarding structural parameters by integrating QSE with empirical trade models.
23
resilience of UK regions to employment shocks, distinguishing between the engineering
resilience and ecological resilience after four recessionary shocks. Their empirical results
relying on VECM specifications showed that employment shocks can have permanent effects.
In a similar way, exploiting the financial crisis of 2008-2009 as a natural experiment, Fingleton,
Garretsen, and Martin (2015) tested whether the regions in Europe have become similar in their
vulnerability and resilience to economic shocks and found spatial heterogeneity in the process
of recovery from the crisis. Brakman, Garretsen, and van Marrewijk (2015) utilized the crisis
to address another question, such as the association between urbanization and short-term
resilience. They empirically showed that EU regions with a larger share of the population in
commuting areas and/or a larger output share in medium-high tech industries were less affected
by the shock.
1.2.2.2. Non-equilibrium Approach
In the past two decades, economic geography has progressed by incorporating
evolutionary aspects, concepts, and theories, particularly those in evolutionary economics. This
field is termed evolutionary economic geography (EEG). EEG has attracted the economic
geographers’ attention to the field related to the process of territorial innovation and its
propagation based on learning because the process is highly compatible with evolutionary
economics. Since EEG originates from evolutionary economics, scholars in this field criticize
the neoclassical equilibrium perspective characterized by comparative statics based on
optimization and linearity. Instead, the main focus of EEG is the complex system with
dynamism and nonlinearity. I first outline the basic concepts of EEG. After that, I briefly review
the notion of evolutionary resilience and introduce the adaptive cycle model as its
representative example.
Following seminal papers, including Frenken and Boschma (2006), Martin and Simmie
24
(2008), and Boschma and Frenken (2011), I outline the conceptual framework of EEG together
with a comparison to NEG. In addition, I explain the concept of (related) variety as one of the
key issues in the empirical literature of EEG and the implication of this concept in the
discussion on the measurement of agglomeration economies. EEG stems from three notions.
First, path-dependency together with the examination of a self-organized development path
through externalities and increasing returns to scale explains long-term technological industrial
development. Second, by utilizing ideas and metaphors, such as novelty, diversity, selection,
adaptation, inheritance, and retention, in evolutionary biologies, the generalized Darwinism
approach explains the evolution of firms and industries. Finally, a complex system approach
applies the ideas in a complex system (e.g., dissipation, far-from-equilibrium, emergence, self-
organization, criticality, and coevolution) to understand the interdependence between a macro-
level economic organization and the micro-level agent’s behaviors.
EEG shares three factors with NEG. First, both EEG and NEG theorize and conceptualize
that the spatial distribution of economic activities is derived from location choice and market
competition rather than from the difference in factor prices between regions. Second,
competition arises at the firm level with increasing returns to scale. Third, the formation of an
industrial cluster of economic activities is a self-organized process based on an irreversible
dynamic system. Finally, path dependency is one of the determinants of the cluster location.
On the other hand, EEG regards path dependence as a result of historical accidents determining
the location of the cluster, whereas in NEG, path dependence is an initial condition of the
location of industrial agglomeration derived from the general equilibrium system.
The notion of path dependency in EEG comes from technological lock-in, (dynamic)
increasing returns to scale, and institutional hysteresis (Martin and Sunley, 2006). The
researchers have focused on the role of the lock-in of a firm, industry, and region. For example,
negative lock-in interrupts regional entrepreneurship and innovation. The superiority of
25
agglomeration (e.g., industrial atmosphere, dense inter-firm relationships, highly specialized
infrastructure, political support by the local government) ends up causing the trap of rigid
specialization (Grabher, 1993). As described in later sections, the EEG literature has discussed
the policy implications of lock-in, as well as a theoretical illustration of the mechanism.
In EEG, the central discussion stemming from the generalized Darwinism approach is the
notion of diversity. In terms of policy implications, the importance of diversity arises due to
the existence of lock-in. The respect for diversity and the acquisition of diversity through the
transfusion of new routines of firms in other regions are imperative to overcome the problem
of the trap of rigid specialization, as the replication of routines with small modification between
(technologically related) firms contribute to the branching of a regional industrial portfolio.
While the discussion on diversity or variety per se is not necessarily a new concept and is also
seen in other fields of regional science, such as urban economics and fields related to business
activities, such as industrial organization and international management, the contribution of
EEG in this context is that it provides an evolutionary foundation on the notion of “related”
variety. The motivation for related variety comes from the “MAR versus Jacobs” debate
throughout regional science.
As described in McCann and Van Oort (2019), agglomeration economies governing
regional growth can be divided into specialization and diversification. The first component,
specialization, is also termed localization economies or Marshall-Arrow-Romer (MAR)
externalities. MAR externalities are characterized by the spillover caused by the transmission
and exchange of knowledge, ideas, and information (whether tacit or formal) between firms in
a geographically concentrated industry. Duranton and Puga (2004) mentioned three sources of
externalities, specifically, learning (knowledge spillover through face-to-face contact),
matching (higher quality of worker-employer matching), and sharing (indivisible inputs only
available in larger cities).
26
Diversification can be further (imperfectly) divided into portfolio and Jacobs externalities
(Frenken, Van Oort, and Verburg, 2007). The portfolio argument, also termed the unrelated
variety in the field of economic geography, is a regional version of the firms’ product portfolio
and has much in common with discussions in other fields, such as industrial organization and
international management. In a manner analogous to that of the capital asset pricing model
(Sharpe, 1970), regional diversity serves as a portfolio strategy to protect the local economy
from sudden and idiosyncratic shocks to a specific sector (e.g., oil price shocks, trade wars,
and radical innovation) and plays an important role in reducing regional growth volatility
(Baldwin and Brown, 2004). The argument for Jacobs externalities claims that the variety of
industries within a geographic region promotes knowledge externalities through the interaction,
copy, and modification of business customs and innovative behaviors. Unlike proponents of
the MAR theory, Jacobs believes that the most important knowledge transfers come from
outside the firms’ own industry and that inventions in one sector can be incorporated in the
production of another industry.
Although following seminal papers like Glaeser, Kallal, Scheinkman, and Shleifer (1992),
the regional science literature has empirically investigated which component of agglomeration
externalities can enhance regional growth, no consensus has been developed so far (De Groot,
Poot, and Smit, 2016). Some investigations did not find significant evidence for any type of
agglomeration externality, while other investigations found evidence for one or two types. As
reviewed in Beaudry and Schiffauerova (2009), the difference in empirical findings is due to
issues in the measurement and methodology of agglomeration rather than the strength of
agglomeration forces across industries, regions, and time periods. One of the strategies to
overcome the dichotomy between specialization and diversification is to introduce related
variety. The basic concept of related variety is effective communication and mutual learning
based on moderate cognitive proximity (Boschma, 2005) between related (but not the same)
27
sectors. Despite the correlation of shock between related sectors, moderate cognitive proximity
is helpful in alleviating lock-in and in stimulating novelty.
A similar argument can be seen in the literature on industrial organization and
international management. While their main interest is in firm-level sectoral and geographical
diversification, economic geographers share a common interest. While the literature (e.g.,
Grant, 2016) also claims similar advantageous aspects of diversification (e.g., growth, risk
reduction, and comparative advantage benefits, such as economies of scope, economies from
internalizing transactions, and a parenting advantage), there is a curvilinear relationship
between the firms’ performance and industrial/geographical diversification (Palich, Cardinal,
and Miller, 2000). While limited diversification focusing on a single industry can limit
opportunities to utilize resources and capabilities across the division, excessive diversification
increases major costs, such as effort losses, coordination costs, and the transaction costs
incurred in managing a business portfolio. In this sense, there is a certain optimal
diversification level at which firms can maximize their performance. Moreover, related or
moderate diversification enables firms to maximize their performance through the utilization
of an associated pool of corporate resources, the improvement of their opportunity to select
better markets, and an increase in access to different location/sector-specific resources (e.g.,
Nachum, 2004; Qian, Li, Li, and Qian, 2008).
A series of empirical investigations beginning from Frenken et al. (2007) have shown the
association between (un)related variety and economic growth. These studies have proposed
empirically tractable indices to measure (un)related variety based on the entropy measure. The
reason why they utilized this measure is decomposability. This property is advantageous in
evaluating the diversification of economic activities within a related (cognitively close) sector,
while undecomposed entropy can measure diversification across sectors.
In these studies, variables capturing (un)related variety were constructed based on
28
regional-level employment data by industrial sector. Suppose all five-digit sectors ; fall
exclusively under a two-digit sectorY5, wherea = 1,… , c. The two-digit share d5 is the sum
of the five-digit shares =(.
d5 = 2 =((∈7!
. (1.5)
Unrelated variety (UV) is defined as the entropy at the two-digit level. This takes a larger value
if the share of each industrial sector in employment is nearly uniform and thus diversified.
)e = 2d5 log" i1
d5j
8
5+!. (1.6)
From the decomposable property, related variety (RV) is given by the weighted sum of entropy
within each two-digit sector. This decomposition enables us to evaluate whether diversification
is especially strong between five-digit sectors included in a specific two-digit sector.
ke = 2d5l5
8
5+!, l5 = 2
=(d5log" i
1
=( d5⁄j
(∈7!. (1.7)
Utilizing these indices, the empirical literature has shown a positive association between
(un)related variety and regional economic growth4. Frenken et al. (2007) showed a positive
association between employment/productivity growth and RV, as well as a negative association
between UV and unemployment growth in the Netherlands. With Italian NUTS-3 level data,
Boschma and Iammarino (2009) found a positive association between related export variety
and value-added growth, while variety per se did not explain regional growth. More recently,
using West German data, Fritsch and Kublina (2017) investigated the association between
employment growth and UV interacted with R&D intensity or between employment growth
and UV interacted with the start-up rate. They showed a positive association between
employment growth and these interactions, which implies that high absorptive capacity
combined with intensive R&D activities or new business formation can contribute to
4 Content and Frenken (2016) provide a comprehensive review of the association in detail.
29
employment growth.
EEG focuses on the ability to adapt and reconstruct industrial, technological, and
institutional structures in an economic system that continues to evolve. In this regard,
evolutionary resilience is an ongoing process occurring together with the redistribution of
resources rather than a recovery to a (incumbent or new) stable equilibrium (Boschma, 2015).
One of the representative conceptual models developed on (mainly) two components of EEG,
namely, path dependency and diversity, is the adaptive cycle model by Simmie and Martin
(2010). The main components of the model are shown in Figures 1.7 and 1.8. The model
illustrates an adaptive cycle characterized by growth, decline, stability and change in the
regional economy. The model consists of four phases, namely, exploitation, conservation,
release, and reorganization, and each phase is associated with the following three capital types
(Dawley, Pike, and Tomaney, 2010).
Figure 1.7. Four-phase adaptive cycle model of regional economic resilience
(Source: Figure 2 in Simmie and Martin, 2010)
Figure 1.8. Resilience as a process; variations in resilience across the adaptive cycle
(Source: Figure 3 in Simmie and Martin, 2010)
30
• Potential of accumulated resources available, inter alia, the competences of individual
firms, skills, hard and soft (business cultures, etc.) infrastructures.
• Connectedness; patterns of relations, networks and collaborations between firms and
agencies; traded interdependencies (e.g., supply agreements) and untraded inter-
dependencies (e.g., informal knowledge spillovers), informal and formal business
associations, labor mobility between firms and agencies, etc.
• Creative and flexible responses; innovative capacity of firms, new firm formation,
entrepreneurialism, venture capital, institutional innovation, etc.
In the exploitation stage, regional economic growth progresses, and the accumulation of
(production, human, knowledge) capital occurs. In this phase, localization economies and
comparative advantage are the key drivers of local industrial development. In the conservation
phase, although the growth driven by specialization reaches a peak, the connectivity between
local economic agents becomes too strong, and the development pattern suffers from
inflexibility. This decreases the tolerance against potential local shocks. Once a shock occurs,
in the release phase, the structural decline and stagnation in growth continue through the
outflow of the labor force and firms, the loss of connectivity, and thus the shrinkage of
industrial agglomeration. In this process, the old production and institutional system are lost,
and resources are released. This leads to the reorganization phase and to beginning again
through the emergence of a new form of activity. The process from release to reconstruction
resembles the Schumpeterian concept. As mentioned in Martin and Sunley (2015), related
variety can play an important role in the transition from the release to the reconstruction phase.
Related variety enhances the transfer of labor and capital to other related economic activities,
while it serves as a shock absorber (Neffke and Henning, 2013).
31
1.2.3. Policy Debate for Regional Economic Resilience
As reviewed in the previous sectors, the development and interpretation of the notions of
regional economic resilience have roughly converged in this decade. As a witness to this, the
publication of the Handbook on Regional Economic Resilience in 2020 was a milestone, as it
took a comprehensive, panoramic view of the developed concepts and ongoing discussions and
thus looked forward to the future direction of the field. Despite the maturity of the notions of
resilience, there is still a gap between the developed notions and their application to real local
economies, particularly to regional economic policies, because the literature on regional
economic resilience has depreciated or neglected the role of governments (e.g., Hassink, 2010;
Boschma, 2015). To enhance the policy discussion on regional economic resilience, we should
carefully examine the (causal) association between resilience and local economic growth and
the determinants of resilience per se (Martin and Sunley, 2015). In addition, we also pay
attention to the specific role of each local economic agent (e.g., firm, household, and
government) in developing resilience and the interaction between individual actors and the
overall regional economy in the process of considering the form of policy intervention (Bristow
and Healy, 2020).
Related to the discussion on policies and strategies, Ross (2007) claimed that we should
distinguish inherent resilience from adaptive resilience. Inherent resilience refers to the
ordinary ability to handle the crisis (e.g., the ability of the market to redistribute resources by
reaction with price signals and the ability of an individual firm to substitute alternative inputs
for those lost to external shock) and is an ability that can be improved before a catastrophic
event occurs. In contrast, adaptive resilience refers to an ability to sustain a function based on
ingenuity and effort under a disruption (e.g., an improvement in the substitutability of inputs
or an enhancement of the market by providing the information to match suppliers to customers).
One of a few consensuses already established so far is the desirability of preventive
32
policies and strategies rather than supportive measures aimed to recover from existing disasters
and conducted after a shock (MacKinnon and Derickson, 2013; Kakderi and Tasopoulou, 2017;
Di Caro and Fratesi 2018). In addition, the importance of the development of “prospective”
adaptive resilience has been emphasized. Taking regional economic policy as an example,
strategies to encourage or enhance a form of development aiming to reduce exposure to
potential vulnerability and shock are necessary (Martin, 2018). Inspired by the development of
the notions of regional economic resilience, these strategies focus on risk mitigation by
addressing structural vulnerability (e.g., the dependence on a single sector, technology, and
market, centralized governance, and a lack of networks) and on defensive factors to enhance
adaptivity (e.g., endogenous knowledge creation, an association of internal knowledge with
external knowledge, the securement of related or moderate variety, the enhancement of
networking and learning, and the development of an environment for innovation) as ways of
developing resilience (Kakderi and Tasopoulou, 2017). In this regard, the policies and
strategies for peacetime market competition are also helpful in preparation against regional
crises (Di Caro and Fratesi 2018). As shown in Figure 1.9, the formation of a regional industrial
ecosystem that associates various actors and factors contributing to local economic resilience
and the support for it can be the jurisdiction of policy intervention (Martin, 2018).
Figure 1.9. Potential arenas for policy intervention (Source: Figure 45.4 in Martin 2018)
33
The first practical guideline on regional policies towards regional economic resilience is
Economic Crisis: Resilience of Regions by the European Observation Network for Territorial
Development and Cohesion; ESPON in 2013. The project consists of regional scientists and
economic geographers who contributed to the development of the concepts described above,
and the final report is organized by largely reflecting the recent achievements in the field. Based
on scientific reports targeting EU countries, the authors provide several insights for
policymakers in building regional economic resilience. I briefly introduce their policy proposal
by referring to EPSON (2014) and Bristow and Healy (2018). They mentioned the following
four factors that should be considered in developing a local economy with high regional
economic resilience.
• Diversity. More diverse economies tend to be more resilient over time, as they prove more
able to adapt to changing circumstances. Policies that avoid establishing a dependence on
specific firms or market segments tend to develop more resilient economies. Equally,
policies that promote the diversification of markets have also proven beneficial.
• Skills. Policies promoting higher-qualified and higher-skilled labor help build economies
with greater resilience capabilities. This is a long-term foundation of more resilient
economies, and its base is laid through consistent policies implemented over a long period
of time.
• Innovation. Regions with higher levels of innovation activity tend to be able to respond to
economic shocks more positively than those in which innovation capabilities are lower.
Policies promoting firm-level innovation may assist in developing more resilient
economies.
• Good governance. There is a strong correlation between the quality of government in a
region and the region’s observed capacity for resilience to economic shocks. Developing
high-quality governance arrangements is a key component to form resilient economies.
34
Table 1.2. Temporal framework for assessing the effect of regional policies (Source: summarized from Wolman et al., 2017 by author)
Short-term Medium-term Long-run
Properties Difficulty in conducting policies aiming to increase employment and income
Dominant role of factor market, productivity, and product cycle in policy design
Possibility of policies whose duration until they have an impact is long
Specific policies
Public enterprise to absorb employment, safety net
Cluster policy, startup support, SME R&D support
Educational policy for human capital investment
Expected effects
Stability of local economy, shock absorption
Diversification, entrepreneurship, innovation of local industry
Change in the trajectory of a regional economy
Wolman, Wial, Clair, and Hill (2017) illustrated specific regional programs enhancing
regional economic resilience and claimed that careful consideration is required in the timing
related to when the program should be put in place and how long it will take for that program
to have an impact. Based on the findings in previous literature5, their own quantitative analysis
conducted by utilizing MSA-level regional data, and the use of a set of qualitative case studies
targeting six metropolitan regions (Charlotte, Cleveland, Detroit, Grand Forks, Hartford, and
Seattle), they developed and utilized a temporal framework for assessing the effect of regional
policies on the acquisition of resilience. As shown in Table 1.2, they divided various place-
based economic programs into short-term, medium-term, and long-run policies. While this
framework focuses on economic crises rather than man-made or natural disasters, it has certain
implications in common with a policy framework to tackle other regional shocks.
In the short term, most of the policies aiming to increase regional employment and income
take too much time to conduct and to realize their effects because the factor market structure
does not change quickly. Thus, these kinds of policies are regarded as infeasible without
support from the upper-level government. While local economic stabilization policies, such as
public enterprise projects, an increase in employment in the public sector, and safety nets, can
be helpful in this situation, these policies marginally or do not contribute to local economic
5 Neumark and Simpson (2015) provide a more general and comprehensive review of the regional economic policies covering theoretical motivation, policy design, and empirical evaluation results. Note that the place-based policies introduced below largely overlap with those for developing regional economic resilience.
35
development and are not always feasible depending on the financial condition in each region.
Unlike the case in short-term policies, under medium-term policies, at a certain level, it
can be possible to change and improve the stock of factors that serve as economic development
assets, and the factor markets dominate the development path as the supply of assets becomes
more elastic. The improvement in productivity, the adaptivity to the dynamics in the product
cycle, and thus the enhancement of competitiveness is crucial in developing the resilience of
the local industry. As partially discussed above, diversity, entrepreneurship, and innovation are
drivers of regional economic resilience. While industrial policies, such as cluster policies,
startup support, and SME R&D policies, can play an important role in developing or
rehabilitating these drivers, the potential is either limited or extremely difficult to execute
unless these policies are very well planned and implemented.
In the longer term, it can be feasible to implement policies to improve the region’s assets,
which might play an important role in changing the trajectory of a regional economy.
Specifically, human capital strategies (e.g., improvements in the pre-school, elementary,
secondary, and community college formal education systems) can play an important role in
improving an area’s labor force skills and thus contributing to an area’s economic growth.
The actions to mitigate the impact of an exogenous shock on regional economies are
limited in the sense that they are outside the purview of economic development agencies and
beyond their budgets. However, despite various constraints (e.g., budget constraints and lack
of expertise), by utilizing regionally adapted and more flexible support and taking advantage
of better access to information on local economic conditions and development trends, local
governments are expected to play a critical role in enhancing regional economic resilience
through the reduction of various costs and risks that prevent local economic agents from
acquiring tolerance and adaptivity against a shock (e.g., Oates, 1972).
36
1.3. Organization of This Thesis
1.3.1 Main Issues
This thesis is organized into six chapters, including this study. Through these chapters, I
attempt to address several open issues relevant to the discussion in this study. The first issue is
the role of networks and the interdependence between regions and sectors in developing
regional economic resilience. As suggested in other studies, for example, Martin and Sunley
(2015) and Bristow and Healy (2020), resilience cannot be developed in isolation, and we
should refrain from dealing with resilience as a notion that only depends on endogenous and
inherent characteristics and processes in a single region. The careful understanding of the
complex interrelationship between the regional economy and the area outside of the region or
the entire world is particularly critical if the ability of local economic agents, such as firms,
workers, and institutions, is insufficient in developing their own resilience.
Practically, the smart specialization policy in the EU is oriented to cross-sectoral
strategies by focusing not on isolated regions and industries but on the interaction with outside
economic agents and on supporting diversification to incorporate external knowledge and
business resources (e.g., McCann and Ortega-Argilés, 2014; Balland, Boschma, Crespo, and
Rigby, 2019; Radovanovic and Benner, 2019). In addition, again, a breakaway from
dependence on a single sector, technology, and market is one of the fundamental actions to
mitigate the structural vulnerability that prevents a region from developing resilience, and this
breakaway can be partially achieved by the enhancement of networking. Despite this
importance, little is empirically known about the role of vertical or horizontal linkage in
developing resilience and its process because the literature has merely focused on the
description of the inter-regional difference in (source of) regional economic resilience and has
ignored the interdependence.
The second issue is an empirical examination of the determinants of (sources of) regional
37
economic resilience. Notwithstanding the development and maturity in the discussion on the
conceptualization of regional economic resilience and the empirical exploration of the
association between (sources of) resilience and local economic development in this decade, it
is very recent that researchers have begun to address the determinants of resilience per se. The
milestone in a recent discussion of the determinants of resilience is a special edition in the 2018
issue of the Annals of Regional Science. The special issue covers seminal articles exploring the
determinants of economic resilience and targeting overall EU regions (Fratesi and Perucca,
2018; Bristow and Healy, 2018; Rizzi, Graziano, and Dallara, 2018) and specific countries that
share the presence of relevant internal socioeconomic disparities (Kitsos and Bishop, 2018 in
the UK; Angulo, Mur and Trivez, 2018 in Spain; Di Caro, 2018; Faggian, Gemmiti, Jaquet,
and Santini, 2018; Mazzola, Lo Cascio, Epifanio, and Di Giacomo; 2018 in Italy).
Di Caro and Fratesi (2018) concluded the special issue by deriving three implications for
future literature. First, they claimed that the determinants of regional economic growth and
those of resilience mostly overlapped. Second, they emphasized the necessity to understand
factors that explain peacetime regional economic performance because they also tend to be
useful to understand some specific patterns observed during and after the disruption in a
specific region. Finally, they suggested that researchers should conduct empirical examinations
based on more spatially disaggregated units because the variation in regional economic
resilience can be observed only at the geographically detailed level (e.g., NUTS-3).
While the second issue in this thesis is particularly related to the final implication, I
additionally motivate this issue from two arguments. One argument is the necessity of
empirical examination based on agent-level microdata to precisely reveal the individual role of
each economic agent’s (e.g., firms, households, and local government) decision-making and
their interaction (see also Bristow and Healy 2020). The other argument is that the firm
heterogeneity hypothesis assumes that firms in a localized sector/industrial cluster may be
38
heterogeneous (Clausen 2013; Wang 2015; Cainelli and Ganau, 2019). Following these
arguments, I emphasize the importance of explicitly considering or controlling agent-level
heterogeneity in an empirical analysis. By doing so, I can distinguish the managerial issues in
developing the resilience for each economic agent from the issues that arise at the local
industrial level or from policy issues.
The final issue is the role of local government and its policies in developing (sources of)
regional economic resilience. As discussed above, the literature on regional economic
resilience has depreciated or neglected the role of governments. This neglect might be
because of this literature’s specific focus on the local industry’s self-organized process to
acquire economic resilience. As is the case in related fields, in the field of regional economic
resilience, little is known about the actual situation and effects of regional policies conducted
by local governments in particular. I point out the importance of a detailed examination of the
policies under the initiative of local governments from two perspectives: complementarity with
national policies and the place-based nature of the local economic policy. Policies that satisfy
region-specific characteristics and policy demands are necessary (Tödtling and Trippl, 2005),
while nationwide policies (e.g., EU Cohesion Policy) that aim for the convergence of inter-
regional disparity and thus an unequal pattern of development can be complemented; these
policies reflect a motivation in common with that in the discussion of multi-level governance
(e.g., Kitagawa, 2007; Laranja, Uyarra, and Flanagan, 2008).
1.3.2 Outline of Essays
This thesis consists of four essays. The first two essays analyze the actions, conducted by
firms and local governments themselves, towards regional economic resilience. In Chapter 2,
I investigate the question of whether learning from the experience of a local catastrophic shock
regionally diversifies a firm’s supply chain against expected future shocks to mitigate potential
39
damage. In Chapter 3, I analyze the determinants of the early development of municipal pre-
disaster preparation plans and the spatial heterogeneity and dependence of development status.
Later, two essays investigate the cooperative actions to acquire (the source of) resilience
between firms and local governments and their consequences. In Chapter 4, I empirically
evaluate whether local R&D support programs enhance the SMEs’ business diversification. In
Chapter 5, I investigate how the local SME modernization program helps firms bounce back
from war damages and postwar disruption.
In Chapter 2, I address the question above by exploiting the Great East Japan Earthquake
(3.11) as an actual shock and the expected Nankai Trough Earthquake as an expected future
shock. Specifically, based on an empirical analysis conducted by utilizing interfirm transaction
panel data and double and triple differences methods, I examine whether firms diversified their
supply chain networks after 3.11 as a pre-disaster preparation to reduce their structural
vulnerability, such as their dependence on a small number of suppliers or input markets. In
Chapter 3, I quantitatively assess the cross-jurisdictional difference in the development status
of integrated business continuity planning (BCP). Utilizing the spatial probit model, I
investigate whether prefecture-level heterogeneity in municipal BCP development and spatial
dependence in the BCP development status among municipalities existed even after controlling
local financial conditions and risk environments.
In Chapter 4, I compare the effects of local R&D support programs on firm performance
in three neighboring prefectures in the same district in Japan. Particularly, to evaluate regional
innovation policies for higher competitiveness, I focus on the regional and industrial
diversification of transaction networks as the outcome. Through empirical evaluation utilizing
interfirm transaction panel data and the fixed-effect method, reflecting the differences in
industrial structure, geographical positions, and policy schemes, I provide evidence of different
consequences of local innovation policies. In Chapter 5, I investigate the effects of the
40
modernization fund program for small business enterprises, a program implemented by the
Osaka Prefecture in the early 1950s, on the performance of local SMEs. Utilizing firm-level
panel data and double- and triple-difference methods, I empirically show how the
modernization program contributed to the local SMEs’ postwar reconstruction and how the
spatial and organizational proximity to wartime economies differentiated the effects.
These essays can contribute to the examination of open issues in the following way.
Chapters 2, 3, and 4 address the first issue, the role of networks and the interdependence
between regions and sectors in developing regional economic resilience. Chapters 2 and 4 focus
on the local firms’ acquisition of a source of regional economic resilience; this resilience source
is represented by the diversity achieved through the enhancement of inter-regional and inter-
sectoral networking. Chapter 3 focuses on the role of the interdependence between
municipalities in developing resilience. I examine the existence of spatial policy diffusion in
which a city’s action to develop regional economic resilience is associated with the neighboring
cities’ actions. This examination can be particularly related to the concepts of neighboring and
network effects (Topa and Zenou, 2015). Because all chapters analyze the determinants of
(sources of) regional economic resilience based on agent-level data and by using the experience
of a shock, policy intervention, and neighbors’ actions as inputs, they contribute to the second
issue. In addition, I can address the third issue, the role of local government and its policies in
developing (sources of) regional economic resilience, through Chapters 3, 4, and 5.
41
Chapter 2
Firms’ Actions: Supply Chain Diversification6
6 This chapter is revised version of Takano, K. (2019). Does visible shock update firms’ unrelated trade diversity in anticipation of future shock? Evidence from the Great East Japan Earthquake and expected Nankai Trough Earthquake (No. E-2019-01). Teikoku Databank Center for Advanced Empirical Research on Enterprise and Economy, Graduate School of Economics, Hitotsubashi University.
42
2. Firms’ Action: Supply Chain Diversification
2.1 Introduction
The increase in the complexity and fragmentation of sourcing as a result of the pursuit of
comparative advantage sometimes makes the supply chain network vulnerable to unexpected
shocks, including natural disasters, industrial disputes, terrorism, and pandemic outbreaks. The
literature regarding supply chain and international trade has emphasized the importance of a
comprehensive understanding of the vulnerability and resilience of the supply chain network
structure to disruption or idiosyncratic shock (Bernard and Moxnes, 2018). In light of the risk
management against natural disasters, in recent years, there has been an emphasis on the
importance of predisaster planning in the supply chain network as well as postdisaster recovery
(e.g., Ranghieri and Ishiwatari, 2014; Fujita, Hamaguchi, and Kameyama, 2018) because
recent natural disasters have revealed the vulnerability of the supply chain network to great
disruptions. Thus, damage in the supply chain network has been increasingly recognized as one
of the primary causes of a disaster’s economic damage. Fundamental factors underlying the
supply chain network’s vulnerability are supply chain globalization, focused factories and
centralized distribution, and a reduced supplier base (Christopher, 2016). Thus, predisaster
preparation eliminating these factors by means of, for instance, the diversification of suppliers,
relocation to low-risk regions, and the development of a business continuity plan has been
regarded as one of the pressing necessities for every firm’s operations.
Firms facing risk and uncertainty need to take predisaster measures under a tradeoff,
especially when the measures, such as the diversification of suppliers, are intended to increase
redundancy, as mentioned in Barrot and Sauvagnat (2016). A spatially intensive supply chain
network can reduce the transaction cost in peacetime while preventing firms from switching to
alternative suppliers and decreasing the absorption ability to alleviate the effect of shocks in
specific sectors or regions. While a spatially extensive supply chain network can reduce the
43
cost of switching to alternative suppliers and thus entail a higher absorption ability, it can also
increase ordinary transaction costs and make the network structure too redundant. Thus, firms
might decide whether or how they should adopt strategies against catastrophic shocks by
referring to some optimal level under this tradeoff.
However, supply chain disruptions caused by natural disasters are characterized by
uncertainty in the sense that it is very difficult to predict the likelihood and magnitude of
damage due to the rareness of such disasters. Therefore, learning based on both their own
experience and the experience of others is important in terms of efficiency after suffering a
disruption (Ponomarov and Holcomb, 2009; Hopp, Iravani, and Liu, 2012). In this sense, based
on learning, firms facing risk and uncertainty are required to update their subjective perception
and preparedness for supply chain disruption.
In response to this growing policy and practical interest, there has been growing concern
about the effect of uncertainty on value chain structure or sourcing decisions in the literature
of international trade. In particular, recent improvements in access to firm- or product-level
value chain network data have promoted empirical investigations (e.g., Gervais, 2018; Kamal
and Sundaram, 2019) on the effect of uncertainty on geographically diversified/concentrated
sourcing decisions under risk aversion. Despite a growing body of related literature, little is
known about the association between a change in risk perceptions toward expected future
shocks, learning from actual shocks, and sourcing decisions.
However, the recent literature has found a remarkable effect of learning from
idiosyncratic shocks, such as the 2008 Financial Crisis, on economic agents’ investment and
financial decisions (Malmendier and Nagel, 2016; Malmendier, Pouzo, and Vanasco, 2020).
Based on this argument, we should emphasize the behavioral aspects of firms’ operation under
cognitive bias and experience-based learning. Regarding sourcing decisions, careful
consideration of firms’ correction of the subjective probability of a rare negative shock due to
44
suppliers’ proximity to the facilities or the areas at risk of expected future shocks (Naoi, Seko,
and Sumita, 2009; Deng, Gan, and Hernandez, 2015; Zhu, Deng, Zhu, and He, 2016, in
residential markets) might provide us further insight into supply chain management under
uncertainty.
This study examines the impact of changes in risk perceptions regarding the supply chain
network after observed disaster damage in terms of predisaster preparation, focusing on the
spatial diversification of the supply chain. To examine this, I utilize the Great East Japan
Earthquake in 2011 as a case of an actual shock and the expected Nankai Trough Earthquake,
a forthcoming mega earthquake in West Japan, as an expected future shock. Through an
examination of the choices of the firms surrounding the area damaged during the East Japan
Earthquake and the area at risk of the expected Nankai Trough Earthquake after 2011, this
study attempts to show to what extent risk perception can serve as the driving force in
predisaster preparation in the supply chain network and how firms restructure the supply chain
network based on these perceptions under capacity constraints. The exploration of this issue
based on both the Great East Japan Earthquake and the Nankai Trough Earthquake is notable
because of the similarities between these earthquake situations. The area damaged by the Great
East Japan Earthquake suffered from an atomic power accident, a tsunami, and ground
movement, and similar damages are expected in the Nankai Trough hazard zone.
The rest of this study is organized as follows. Section 2.2 reviews the literature on the
economic impacts of natural disasters and diversification in economic activities. Section 2.3
briefly describes the Great East Japan Earthquake in 2011, the expected Nankai Trough
Earthquake and the (projected) damage from each earthquake. Section 2.4 describes the dataset
and the analytical framework of impact evaluation. Section 2.5 reports the estimation results.
Section 2.6 concludes the study.
45
2.2 Literature Review
2.1.1 Natural Disasters and Firms’ Activities
The empirical assessment of the impact of a natural disaster on the supply chain using
firm-level network data has followed two perspectives in recent years. On the one hand, recent
investigations have pursued impact evaluations focusing on the magnitude of the effects of
disaster shock and recovery on firm performance by incorporating the supply chain structure.
Barrot and Sauvagnat (2016) examined whether firm-level idiosyncratic shocks associated with
natural disasters propagate in production networks. They showed that affected suppliers impose
substantial output losses on their customers, especially when they produce specific inputs.
Carvalho, Nirei, Saito, and Tahbaz-Salehi (2016) estimated the overall macroeconomic impact
of the shock of the Great East Japan Earthquake by incorporating the aftermath effects that the
earthquake propagated for both upstream and downstream supply chains. They found that the
propagation over input-output linkages accounted for a 1.2 percentage point decline in Japan’s
gross output the year after the Great East Japan Earthquake. Comparatively, Todo, Nakajima,
and Matous (2015) examined how supply chain networks affected the recovery of firms after
the Great East Japan Earthquake. They found that networks with firms outside the damaged
area contributed to production recovery, while networks within the region contributed to sales
recovery in the medium term. Based on these investigations, it can be inferred that although a
supply chain serves as an aid to the recovery from disaster shock, it also propagates or amplifies
the shock.
On the other hand, there has been an increase in the impact evaluations focusing on the
choice of pre- or postdisaster business activities based on the supply chain, particularly related
to the issue here. Zhu et al. (2016) identified the macro fluctuation of firms' offshoring using
the Great East Japan Earthquake as the exogenous shock and showed a positive effect of the
earthquake on the offshoring of manufacturing. Cole, Elliott, Okubo, and Strobl (2017)
46
examined the extent to which predisaster planning and postdisaster aid played a role in firms’
recovery from the Great East Japan Earthquake. They found that predisaster policies, such as
having alternative transport arrangements and a diversified supplier network, were positively
associated with postdisaster sales recovery.
This study complements the literature above by considering the existence of both an
actual and a future shock and an interaction between them. This study complements the
literature on impact evaluations focusing on the choice of predisaster or postdisaster business
activities by examining firms' actions outside the damaged area rather than inside it.
2.2.2 Business Portfolio under Uncertainty
As mentioned above, one of the main measures against supply chain disruption is the
diversification of the supply chain. However, going beyond supply chain diversification, the
literature particularly discussed the importance of diversification in economic activities. As
described in, for example, Rugman (1979) and Fujita and Thisse (2013), the intersectoral or
regional diversification of the business portfolio can enable firms or regions to alleviate a shock
to a particular sector or region by reducing the dependence on a single market, analogous to
portfolio theory, such as the capital asset pricing model (Sharpe, 1970)7. A recent study by Hsu,
Lee, Peng, and Yi (2018) empirically examined this function, focusing on the occurrence of
disaster shock. Specifically targeting chemical plants in the United States, they showed that a
diversified technological portfolio mitigates the loss of damaged firms’ return on assets. The
aforementioned studies by Todo et al. (2015) and Cole et al. (2017) can also be positioned in
this literature, as they showed a positive association between a (spatially) diversified supply
7 This conceptual framework has been confirmed in a growing body of empirical literature, especially in economic geography (Baldwin and Brown, 2004; Frenken, Van Oort, and Verburg, 2007) and international management (Nachum, 2004; Qian, Li, Li, and Qian, 2008).
47
chain and early postdisaster recovery.
Despite the growing body of empirical literature examining the association between a
diversified business portfolio and the economic resilience of firms or regions, little is known
about what can enhance diversification. Taking uncertainty as an example of the determinants
of a business portfolio, a few empirical investigations in international trade literature have
examined this question. Many of them considered input price uncertainty as a determinant of
sourcing diversification. Recently, Gervais (2018) theoretically showed the regional dispersion
of sourcing under uncertainty (input markets with high price volatility) and risk aversion and
empirically confirmed this proposition with country-product level trade data8.
This study extends the literature by examining the change in risk perceptions toward
future shock by taking a cue from the occurrence of an actual shock—firms’ correction of the
subjective probability of a rare negative shock. As discussed in the grassroots literature on
supply chain management based on conceptual analysis or case study, learning from
disruptions plays an important role in firms’ modification of operations in response to
challenges or opportunities (Ponomarov and Holcomb, 2009; Pettit, Fiksel, and Croxton, 2010).
This improvement might also enable a business to gain a competitive advantage since it
sometimes indicates the firm’s ability to tackle future shocks more efficiently than its
competitors (Fiksel, Polyviou, Croxton, and Pettit, 2015). Learning can also be acquired by
referring to others’ experience; in this case, the learner is motivated to avoid a repetition of the
same disruption within the supply chain (Hopp, Iravani, and Liu, 2012; Scholten, Scott, and
Fynes; 2019). Moreover, there is heterogeneity in the magnitude of learning. Although learning
from the experience of others is important and efficient in increasing security, locating facilities,
and tracking the financial health of suppliers (Richey, Natarajarathinam, Capar, and Narayanan,
8 In contrast, Kamal and Sundaram (2019) showed that sourcing concentration strategies for sharing supplier information are more dominant in countries with weaker institutions.
48
2009; Ramezani, and Camarinha-Matos, 2020), many firms learn only after suffering a
disruption (Hopp, Iravani, and Liu, 2012). Despite the significance of learning in alleviating
shocks, empirical investigations on sourcing uncertainty hardly address this dynamic and
subjective aspect of firms’ preparedness.
In recent literature on investment and financial decisions, there has been a growing
concern about learning from large shocks by explicitly considering cognitive bias. Recently,
Malmendier et al. (2020) introduced a subjective model describing expectations of future asset
price updated by learning. They examined two aspects of experience-based learning—the
overweighting of recent realizations and the long-lasting effect of shock experiences—focusing
on investors’ stock market investment decisions and portfolio choices. Related to this concept,
I examine the reconstruction of a supply chain network by assuming that firms’ subjective
expectations of damage caused by future shocks are heterogeneously updated depending on
their suppliers’ proximity to facilities or areas at risk, analogous to the literature on urban
economics addressing risk perception behavior measured by the price effect in the residential
market (Naoi, Seko, and Sumita, 2009; Deng, Gan, and Hernandez, 2015; Zhu, Deng, Zhu, and
He, 2016).
49
2.3 The Great East Japan Earthquake and the Nankai Trough Earthquake
The Great East Japan Earthquake in 2011, also called 3.11, and the expected Nankai
Trough earthquake and the (projected) damage from each earthquake are briefly described here.
The characteristics of the Great East Japan Earthquake are summarized based on
information from Japan’s Cabinet Office (CAO, 2011). The earthquake occurred on March 11,
2011, with a magnitude of 9.0. It caused extensive damage over a wide range centered on the
northeast coast of Japan as well as widespread disruptions all over Japan due to strong ground
shifts and a tsunami. It led to the loss of 22,626 lives and economic losses reaching 16.9 trillion
yen. In addition, the meltdown accident at the Fukushima Daiichi Nuclear Power Plant caused
by the tsunami led to the evacuation of 146,520 residents within 30 km of the plant. The greatest
difference between this earthquake and other recent natural disasters, such as Hurricane Katrina
in 2005 in the United States and the Great Hanshin-Awaji Earthquake in 1995 in the southern
part of Hyogo Prefecture in Japan, is the economic damage propagated in a wide sphere outside
of the damaged area through the restriction of electricity and supply chain disruption.
After the Great East Japan Earthquake, the potential risk of a substantial earthquake in
the Nankai Trough area began to receive considerable attention. The Nankai Trough has
created large earthquakes at 100- or 200-year intervals over the past 1400 years. As 70 years
have passed since the last earthquake in that area (the Showa Nankai Earthquake in 1946), it
has been assumed that the next large earthquake will occur within a shorter period. Figure 2.1
shows estimated seismic intensity of the Nankai Trough Earthquake. In regions colored with
warmer colors, the expected damage caused by tsunami and ground movement is higher.
Although it is generally difficult to make an exact prediction of the geographic scope of a large
ground shift using current technology, a rough estimation of the maximum seismic intensity
shows that the damage from ground movement could spread over a wide area across Japan.
According to the Cabinet Office (CAO, 2016), the economic loss from a Nankai Trough
50
earthquake could reach approximately 220 trillion yen. This estimated economic loss exceeds
that of the Great East Japan Earthquake, which was approximately 16.9 trillion yen since the
area affected by the earthquake was in the Pacific belt zone. In addition, the area at risk of the
expected Nankai Trough earthquake includes Shizuoka Prefecture, which is one of the
prefectures assumed to be a high damage risk zone and has a nuclear power plant (the Hamaoka
Nuclear Power Plant). In this respect, as in the case of the Great East Japan Earthquake, the
Nankai Trough earthquake may lead to complex disasters, including tsunami damage, a nuclear
plant accident, and ground movement.
Figure 2.1 Estimated seismic intensity of the Nankai Trough Earthquake (Harner, 2012)
51
2.4 Methodology
2.4.1 Empirical Procedure
2.4.1.1 Conceptual Framework
Referring to the literature review in the previous chapter, I set up the following hypotheses.
H1: The geographical diversification of the supply chain network has been promoted
against the Nankai Trough Earthquake after the Great East Japan Earthquake, 3.11, if a firm
had suppliers in hazardous regions of the Nankai Trough Earthquake before 3.11.
H2: The intensity of diversification in H1 is stronger if the firm had suppliers both in the
regions at risk of the Nankai Trough Earthquake and in the area damaged during 3.11.
H1 stems from the need for firms to update their risk perception regardless of the direct
damage to their incumbent suppliers. This hypothesis might be supported if the negative shock
of 3.11 were strong and persistent enough for firms to correct the subjective probability of a
rare negative shock and thus modify their operations in expectation of future shocks, whether
or not they were affected by the disruption. If this modification comes from the perception of
the benefit of supply chain diversification, as empirically confirmed in Todo et al. (2015) and
Cole et al. (2017), the shock of 3.11 might update the spatial distribution of suppliers. In this
sense, based on H1, I examine the mixed effect of learning from the experience of others or
from the firm’s own experience.
On the other hand, Hypothesis H2 examines the updated risk perception of firms
conditioned by the damage to their incumbent suppliers. In other words, I hypothesize the
existence of heterogeneity in the effect tested in H1. As described in Section 2.2, many firms
learn only after they suffer from a disruption, and they often fail to take action utilizing the
learning from the experience of others despite the various benefits of increasing supply chain
resilience against uncertainty. Therefore, it is necessary to distinguish the effect of the learning
based on a firm’s own experience from the mixed effect examined based on H1.
52
Although there is little rigorous econometric evidence about the extent to which the
structure of supply chain networks changed after the Great East Japan Earthquake, Fujita et al.
(2018) provided anecdotal evidence of predisaster preparation among firms motivated by the
Great East Japan Earthquake. Several major manufacturing firms whose production bases were
damaged by the Kumamoto Earthquake, such as Sony and Aisin, recovered effectively because
they implemented steps in their business continuity plans that included the relocation of
factories to other firms and the import of alternative products from overseas plants after the
Great East Japan Earthquake.
2.4.1.2 Estimation Methods
This study focuses on the structural change of the supply chain network among firms in
the context of geography. The inverse of the Herfindahl-Hirschman Index (HHI) is used to
measure the geographic distribution of the supply chain network. While the HHI is a well-
known index for measuring market concentration, recent empirical investigations on
agglomeration economies (De Goort, Poot, and Smit 2016) employed this index to measure
urbanization economies. The HHI of the supply chain network for firm;is defined here as
follows:
ll9( =2Am_^. ^n6o==G;pU6;_Upa;^_9q((8^8`G_^. ^n6o==G;pU6)(
B
":
9+!, (2.1)
where;, rdenote the firm and region, respectively. In this study, 1/HHI is calculated at the
prefectural level and Koiki (wide-regional plan) area level. The spatial scale of a prefecture
and the scale of a Koiki region are close to NUTS-2 and NUTS-1, respectively.
I implemented counterfactual analysis using the difference-in-differences (DD) method
for examining H1 and the difference-in-difference-in-differences (DDD) method for H2 based
on panel data from 2009 to 2017. In the following empirical analysis, only manufacturing firms
53
with a positive number of suppliers, located in neither at-risk regions nor damaged regions
throughout the period, are included in panel data.
In the empirical analysis, the damaged areas in the Great East Japan Earthquake (3.11)
were defined in terms of municipalities damaged by the tsunami. The source of this information
was the Ministry of Agriculture, Forestry, and Fisheries (MAFF, 2011). Similarly, the
hazardous Nankai Trough area was defined according to the municipalities designated the
“Areas for Special Reinforcement of Nankai Trough Earthquake Tsunami Evacuation
Measures” based on “The Act on Special Measures concerning Advancement of
Countermeasures against Disasters of Tonankai and Nankai Earthquakes.” The source of the
information was Japan’s CAO (2015). In Figure 2.2, I show the geographical range of the
regions affected by the tsunami in the aftermath of the Great East Japan Earthquake and the
regions expected to suffer tsunami damage after the Nankai Trough Earthquake defined by
MAFF (2011) and CAO (2015), respectively. The damaged area of 3.11 includes large part of
prefectures in Tohoku Region. The damages of 3.11 were severer in coastal municipalities in
Iwate, Miyagi, and Fukushima. Also, reflecting spatial pattern of estimated seismic intensity
in Figure 2.1, we can see that coastal municipalities in Miyazaki, Kochi, Tokushima,
Wakayama, Mie, Aichi, and Shizuoka are regarded as those under higher risk of the Nankai
Trough Earthquake.
Figure 2.2 Tsunami damaged regions of the Great East Japan Earthquake (green) and tsunami hazard regions of
the Nankai Trough Earthquake (red).
54
The specification corresponding to the DD estimation is as follows:
[1/ll9](2 = 52 + t( + "N( × un8pU2I + v′(2x + @(2 , (2.2)
where ;and8denote the firm and year, respectively; [1/ll9](2is the inverse of the HHI
regarding the supplier, as defined previously, of firm;in year 8; and 52andt( are time fixed-
and firm fixed effects, respectively. v(2is a vector of control variables that includes two-digit
standard industrial classification dummies, based on the TDB Standard Industrial
Classification9, Koiki region (wide-regional plan areas) dummies, industry-year dummies, and
Koiki-year dummies. "N( equals 1 if firm;had suppliers in hazardous regions of the Nankai
Trough Earthquake in 2008. The un8pU2dummy indicates the post-3.11 period; namely, if
8 ≥ 2011, un8pU2equals 1. "N( × un8pU2, the regressor of interest, captures the effect of the
Great East Japan Earthquake on the diversification of the network, regardless of the direct
damage to a firm’s incumbent suppliers. H1 is supported if firms having incumbent suppliers
in the Nankai Trough hazardous area spatially diversified their supply chain network after
3.11—I > 0. In a similar way, the specification for the DDD estimation is as follows:
[1/ll9](2 = 52 + t( + "N( × |}( × un8pU2~ + �(2; Ä + Å(2 , (2.3)
where �(2; Ä = v′(2x + "N( × un8pU2I! + |}( × un8pU2I" . |}( equals 1 if firm ; had
incumbent suppliers in regions damaged by the Great East Japan Earthquake in 2008.
"N( × |}( × un8pU2 , the regressor of interest, captures the effect of the Great East Japan
Earthquake on the diversification of the network, with direct damage to firms’ incumbent
suppliers. H2 is supported if the magnitude of the spatial diversification of the supply chain
network after 3.11 increased in the case that the firm had incumbent suppliers in the area
damaged by 3.11 as well as in the hazard area of the Nankai Trough Earthquake—~ > 0.
9 Available at https://www.tdb.co.jp/lineup/pdf/tic.pdf. This industrial classification roughly corresponds to a general industrial classification, such as the Japan Standard Industrial Classification and International Standard Industrial Classification. Thus, a two-digit code in the TSIC is equivalent to that in the JSIC and ISIC.
55
This study adopts propensity score matching (PSM) to control for selection problem.
Taking the comparative analysis on H2 as an example, the likelihood that a large firm is
included in the treatment group in 2008 increases since large firms tend to have a
geographically broad supply chain network structure. Additionally, regarding both H1 and H2,
whether firms have suppliers in a damaged or at-risk region depends on their location. This
imbalance between the treatment and control groups due to, for example, the location, size, and
sector of firms causes the problem of selection and makes the estimations with the DD and
DDD method biased. PSM tackles this problem by matching each firm in the treatment group
with a firm in the control group that has a similar probability of being assigned to the treatment
group. The probability of the assignment, equivalent to the propensity score, is predicted with
information from the year immediately before the treatment of each firm. Thus, PSM aims to
achieve covariate balancing between the groups. In this study, the propensity score is estimated
by using the logarithm of sales and the number of employees, an interaction between these
variables, Koiki region dummies, two-digit level industrial dummies, and the interaction term
between the continuous and dummy variables in 2008.
2.4.2 Interfirm Transaction Data
The interfirm transaction database was provided by Teikoku Databank (TDB), a major
corporate credit research company in Japan that collects interfirm transactional data through
door-to-door surveys. Approximately 1700 field researchers visit and interview firms to obtain
corporate information in every industrial category and location. The database also includes the
annual transactional relationships among the firms. The database included 39,568,392 records
of transaction relationships among 1,554,850 firms over the period 2008–2017. Specifically,
in 2016, the database included 1,136,203 firms out of a total of 3,856,457 firms in Japan,
according to the Japanese Economic Census in 2016. Thus, during this period, the database
56
captured the interfirm transactional activities of nearly one-third of all firms in Japan. In
addition, the dataset was connected with a corporate information database, COSMOS; hence,
the basic corporate information of each firm, such as sales, number of employees, geographic
location of its headquarters, and industrial category, was also available. In the door-to-door
interviews, each firm reported up to five of its suppliers and clients.
Table 2.1: Ratio of purchases from overseas by firm size (Manufacturing) 2010
No. of employees No. of firms Total purchases Purchases from overseas Overseas ratio [%] 50-99 4275 6000935 436333 7.27 100-199 3968 10097588 617167 6.11 200-299 1618 8463152 740774 8.75 300-499 1286 11852309 1655026 13.96 500-999 926 16612116 1844557 11.10 More than 1000 809 117519043 12826993 10.91
2017 No. of employees No. of firms Total purchases Purchases from overseas Overseas ratio [%] 50-99 3957 4750256 398487 8.39 100-199 3840 9925770 742006 7.48 200-299 1591 8020141 1072431 13.37 300-499 1270 11420462 1447006 12.67 500-999 958 18575654 2692761 14.50 More than 1000 830 118805258 14202167 11.95
Notes: Original source is “Basic Survey of Japanese Business Structure and Activities” by the METI. The unit of total purchases and purchases from overseas is million yen. The respondents of the survey consist of Japanese firms with more than 50 employees. The response rate was 84.6% in 2010 and 84.5% in 2017.
Since this dataset includes both self-reported and other-reported transaction information,
the number of suppliers for each firm usually exceeds five. These annual data are superior
because they capture the dynamism of the disaggregated supply chain network structure, unlike
an input-output table. One limitation of this transaction information database is that it only
covers the domestic supply chain network and does not capture the global scale. However, as
shown in Table 2.1 based on the Basic Survey of Japanese Business Structure and Activities
by the METI, domestic sourcing by Japanese firms accounts for approximately 90% of total
purchases regardless of firm size in the manufacturing sector both before and after 3.11. In this
57
sense, this limitation is not necessarily a serious problem in my empirical investigation10.
10 Other limitations are as follows. First, the database captures the existence of ties among firms, not the strength of the ties. Second, it captures the transaction network among headquarters, not among plants.
58
2.5 Results
2.5.1 Propensity Score Matching
Before the parameter estimation with DD and DDD, the attributes between the treatment
and the control groups are calibrated with PSM. As the first step, I balance the observable
covariates between firms corresponding to "N( = 1 and those corresponding to "N( = 0 ,
regarding the existence of suppliers in hazard regions of the Nankai Trough Earthquake as a
treatment. Subsequently, as the second step, I additionally balance the observable covariates
between firms corresponding to|}( = 1 and those corresponding to|}( = 0, regarding the
existence of suppliers in regions damaged in the Great East Japan Earthquake as a treatment.
The matching method used in the first step is a simple one-by-many matching based on a
genetic algorithm (Leite, 2016), while that used in the second step is propensity score
stratification. The propensity score stratification creates subclasses of similar subjects, for
example, as defined by the quintiles of the propensity score distribution (Stuart, 2010).
Rosenbaum and Rubin (1985), for example, demonstrated that by creating five propensity score
subclasses, at least 90% of the bias in the estimated treatment effect was removed. The reasons
I utilize the stratification method are as follows. First, it may be particularly necessary to
control for the heterogeneous effect of the Great East Japan Earthquake, depending on firm
size, location, and sector. Second, it is quite difficult to maintain covariate balance in the first
step when utilizing other matching methods. Based on this method, samples matched in the
first stage are classified into five subclassses depending on the magnitude of the propensity. In
this process, samples that do not satisfy the common support assumption are discarded. In each
step, I conduct PSM with a caliper restriction, which limits the number of standard deviations
of the propensity distance measure within which to draw control units. Following the previous
literature (e.g., Austin, 2011), I match treated units with the subset of untreated units using
calipers of width equal to 0.2 of the standard deviation of the logit of the propensity score.
59
Table 2.2: Estimation results of the binomial logistic regression on all manufacturing samples corresponding to the existence of suppliers in regions at risk of the Nankai Trough Earthquake (2008)
beta z-val
(Intercept) −3.112 −22.533 *** lnSALES 0.054 2.189 ** lnEMP −0.570 −17.615 *** lnSALES×lnEMP 0.091 24.110 *** lnSALES×KOIKI_Chugoku 0.092 2.754 *** lnSALES×KOIKI_Hokkaido −0.495 −3.842 *** lnSALES×KOIKI_Hokuriku 0.081 1.784 * lnSALES×KOIKI_Kinki 0.115 5.277 *** lnSALES×KOIKI_Shikoku −0.158 −1.988 ** lnSALES×KOIKI_Shuto 0.129 6.896 *** lnSALES×KOIKI_Tohoku 0.071 1.919 * lnEMP×KOIKI_Hokkaido 0.444 2.825 *** lnEMP×KOIKI_Kyushu 0.114 3.452 *** lnEMP×KOIKI_Okinawa −0.546 −2.400 ** lnEMP×KOIKI_Shikoku 0.211 2.231 ** 2-digit dummy YES Koiki dummy YES PseudoR-sq 0.169 N 92308
Notes: Dependent variable is !"#$(&'" = 1), where&'" equals 1 if firm,had suppliers in hazardous regions of the Nankai Trough Earthquake. Significant at the ***1%, **5%, and *10% levels. To obtain the PS prediction model with higher generalization performance, I conducted the stepwise variable selection based on AIC. Candidate covariates include the logarithm of sales and the number of employees, an interaction between these variables, Koiki region dummies, two-digit level industrial dummies, and the interaction term between the continuous and dummy variables in 2008.
Table 2.3: Estimation results of the binomial logistic regression on the matched sample corresponding to the existence of suppliers in regions damaged by the Great East Japan Earthquake (2008)
beta z-val
(Intercept) −8.175 −13.221 *** lnSALES 0.694 7.713 *** lnEMP −0.571 −3.976 *** lnEMP^2 0.058 5.488 *** lnSALES×KOIKI_Shuto −0.212 −1.902 * lnEMP×KOIKI_Chubu 0.156 1.415
lnEMP×KOIKI_Chugoku 0.352 1.885 * lnEMP×KOIKI_Kinki 0.275 2.553 ** lnEMP×KOIKI_Shuto 0.231 1.665 * lnSALES×KOIKI_Tohoku −0.192 −2.203 ** 2-digit dummy YES Koiki dummy YES PseudoR-sq 0.298 N 17734
Notes: Dependent variable is !"#$(-." = 1), where-." equals 1 if firm,had suppliers in damaged regions of the Great East Japan Earthquake. Significant at the ***1%, **5%, and *10% levels. To obtain the PS prediction model with higher generalization performance, I conducted the stepwise variable selection based on AIC. Candidate covariates include the logarithm of sales and the number of employees, an interaction between these variables, Koiki region dummies, two-digit level industrial dummies, and the interaction term between the continuous and dummy variables in 2008.
60
Table 2.4: Chi-square overall test corresponding to the existence of suppliers in regions at risk of the Nankai Trough Earthquake (2008)
Subclass 1 Subclass 2 Subclass 3 Subclass 4 Subclass 5 No. of Treatment (NT=1) 6101 1163 601 385 338
No. of Control (NT=0) 7235 904 378 161 61 Overall Balance (H0: Balanced) p<0.01 p<0.1 n.s. n.s. n.s.
Table 2.5: Chi-square overall test corresponding to the existence of suppliers in regions damaged during the
Great East Japan Earthquake (2008) Subclass 1 Subclass 2 Subclass 3 Subclass 4 Subclass 5
No. of Treatment (EJ=1) 210 210 211 209 209 No. of Control (EJ=0) 13126 1857 768 337 190
Overall Balance (H0: Balanced) p<0.01 p<0.1 n.s n.s. n.s.
The estimation results of the propensity score corresponding to each treatment status are
shown in Tables 2.2 and 2.3. The results of PSM evaluated with the chi-square overall test are
shown in Tables 2.4 and 2.5. Table 2.4 evaluates the overall covariate balance between firms
corresponding to"N( = 1 and those corresponding to"N( = 0, while Table 2.5 evaluates that
between firms corresponding to|}( = 1 and those corresponding to|}( = 0. The sequential
serial number of each subclass is based on the magnitude of the propensity score in ascending
order. In other words, firms included in Subclass 1 have the lowest propensity score, while
those in Subclass 5 have the highest score. These results show that overall covariate balance is
not achieved in Subclass 1, with both NT and EJ at least at the 1% level. Thus, the following
empirical analysis is implemented on the subclasses, except for Subclass 1, owing to potential
violation of a parallel trend in DD and DDD.
As shown in Figure 2.3, the logarithms of both the number of employees and amount of
sales are proportional to the magnitude of the propensity score. Based on Table 2.6, on average,
Subclasses 2, 3, 4, and 5 include SMEs, medium-sized firms, large firms, and leading firms,
respectively. Table 2.7 shows the industrial composition in each subclass. Although there are
some switches of order, the lineup of 2-digits with a large proportion is consistent regardless
of subclass. The spatial distribution of the firms in each subclass is shown in Figure 2.4.
61
Figure 2.3 Box plots in logarithm of the number of employees [persons] (Left) and logarithm of sales [million ¥]
(Right) in each subclass (2008)
Table 2.6: Average number of employees and average sales (inverse log transformed, 2008) Subclass 1
(Micro) Subclass 2
(SME) Subclass 3
(Medium-size) Subclass 4
(Large) Subclass 5 (Leading)
No. of employees [persons] 17.2 92.7 219.9 435.8 1136.5 Sales [million ¥] 449.7 3780.3 10804.9 26566.8 85528.1
Table 2.7: Industrial composition in each subclass (2008)
Subclass Ratio Industry Name (2-digit)
2
0.191 FOOD PROCESSING 0.18 GENERAL MACHINERY MANUFACTURING
0.102 ELECTRICAL MACHINERY AND EQUIPMENT MANUFACTURING 0.085 MISCELLANEOUS MANUFACTURING INDUSTRIES 0.083 METALLIC PRODUCT MANUFACTURING
3
0.157 FOOD PROCESSING 0.143 GENERAL MACHINERY MANUFACTURING 0.112 ELECTRICAL MACHINERY AND EQUIPMENT MANUFACTURING 0.093 CHEMICAL INDUSTRY 0.089 MISCELLANEOUS MANUFACTURING INDUSTRIES
4
0.178 GENERAL MACHINERY MANUFACTURING 0.147 ELECTRICAL MACHINERY AND EQUIPMENT MANUFACTURING 0.137 FOOD PROCESSING 0.092 CHEMICAL INDUSTRY 0.073 STEEL INDUSTRY AND NONFERROUS METAL MANUFACTURING
5
0.19 FOOD PROCESSING 0.178 ELECTRICAL MACHINERY AND EQUIPMENT MANUFACTURING 0.16 GENERAL MACHINERY MANUFACTURING
0.123 CHEMICAL INDUSTRY 0.073 CERAMIC, STONE AND CLAY PRODUCT MANUFACTURING
Notes: Only the five largest 2-digit industries by proportion are shown in this table.
62
Subclass 2
Subclass 3
Subclass 4
Subclass 5
Figure 2.4 Spatial distribution of the firms in each subclass in 2008
Table 2.8: Observed number of firms with each treatment status (2008) Subclass 2 Subclass 3 Subclass 4 Subclass 5
NT=0&EJ=0 822 313 118 35 NT=1&EJ=0 1035 455 219 155 NT=0&EJ=1 82 65 43 26 NT=1&EJ=1 128 146 166 183
Firms included in each subclass are commonly concentrated around Tokyo. In this sense,
based on the property of the dataset, the likelihood that firms’ headquarters were (or will be)
directly damaged by tsunamis may be relatively low. Furthermore, the intensity of
concentration is stronger in Subclasses 4 and 5 than in Subclasses 2 and 3. Therefore,
considering the size of firms in each subclass, the interpretation of the empirical result
regarding Subclass 5 (or even 4) would require close attention because there may be concern
about the gap between the location of headquarters and the location of business enterprises.
Table 2.8 shows the number of firms in each treatment status.
2.5.2 Regression Results with DD
In this section, I describe the estimation results with DD to examine H1. First, I examine
the results using the prefecture level 1/HHI as an outcome. Table 2.9 shows the DD estimation
results for each subclass with the prefecture level 1/HHI. Except for the result in column 4
regarding Subclass 5, which consists of leading firms, the treatment effect in each subclass is
63
not significant. Thus, a positive effect on network diversity averaged throughout the period
after 3.11 is hardly observed for most subclasses.
Table 2.9: DD estimation results with time-invariant treatment variable (Dependent variable: Prefecture level 1/HHI, 2009-2017)
(1) (2) (3) (4) Subclass 2 Subclass 3 Subclass 4 Subclass 5
beta t-val beta t-val beta t-val beta t-val
NT×After −0.020 −0.653 0.009 0.174 0.028 0.397 0.17 1.712 * Firm FE YES YES YES YES Year FE YES YES YES YES 2-digit FE YES YES YES YES Koiki FE YES YES YES YES 2-digit×Year FE YES YES YES YES Koiki×Year FE YES YES YES YES n 18603 8811 4914 3591
Notes: Significant at the ***1%, **5%, *10% levels. Standard errors are clustered at the firm level. Main treatment variable NT takes 1 if a firm had suppliers in hazardous regions of the Nankai Trough Earthquake.
Subclass 2
Subclass 3
Subclass 4
Subclass 5
Figure 2.5 DD estimation results with time-variant treatment variables (Dependent variable: prefecture level 1/HHI, 2009-2017)
Notes: Standard errors are clustered at the firm level. Main treatment variable NT takes 1 if a firm had suppliers in hazardous regions of the Nankai Trough Earthquake. A plot represented with “■” means that the treatment variable NT corresponding to the year is significant at least at the 10% level. Vertical bars added to the plots represent the 95% confidence interval.
64
The estimation results with time-variant treatment variables are shown in Figure 2.5,
where each point on the plot represents a point estimate of the regression parameter in each
period, and points indicated with a filled square mean that the treatment variable corresponding
to the period is significant at least at the 10% level. Vertical bars added to the points represent
the 95% confidence interval. As is the case with the averaged treatment effect, treatment effects
evaluated on each year separately are not or are only marginally significant. Although positive
effects can be observed for Subclass 5, they are significant only for the periods 2011 and 2016.
Thus, a consistent positive effect on network diversity in Subclass 5 may not exist. The
treatment variable in 2010 is insignificant in all subclasses; hence, this result implies that there
is no convincing evidence of the violation of parallel trends.
Second, to capture a broader geographical redistribution of suppliers, the results using the
Koiki region level 1/HHI as an outcome are examined.
Table 2.10: DD estimation results with time-invariant treatment variable (Dependent variable: Koiki region level 1/HHI, 2009-2017)
(1) (2) (3) (4) Subclass 2 Subclass 3 Subclass 4 Subclass 5
beta t-val beta t-val beta t-val beta t-val
NT×After −0.029 −1.993 ** −0.023 −1.105 −0.017 −0.619 0.07 1.905 * Firm FE YES YES YES YES Year FE YES YES YES YES 2-digit FE YES YES YES YES Koiki FE YES YES YES YES 2-digit×Year FE YES YES YES YES Koiki×Year FE YES YES YES YES n 18603 8811 4914 3591
Notes: Significant at the ***1%, **5%, *10% levels. Standard errors are clustered at the firm level. Main treatment variable NT takes 1 if a firm had suppliers in hazardous regions of the Nankai Trough Earthquake.
Table 2.10 shows the DD estimation results with the Koiki level 1/HHI. As shown in
column 4, the treatment effect is positively but weakly significant in Subclass 5 but not in other
subclasses and is even negatively significant in Subclass 2 (column 1). Thus, a positive effect
on network diversity averaged throughout the period after 3.11 is hardly observed for most
subclasses. As with the estimation based on the prefecture level 1/HHI, I also implemented DD
65
based on time-variant treatment variables; the results are shown in Figure 2.6. Significant
treatment effects are observed in Subclasses 2 and 5, but they were exposed after 4–5 years of
3.11. Thus, with these results, it can be difficult to assert the direct causality between 3.11 and
network diversity or concentration. The treatment variable in 2010 is not significant in most
subclasses. In Subclass 3, the treatment variable is significant in 2010 and 2011. However, the
temporal variation in the treatment effects is not as large as that after 2012. These results imply
that there is no convincing evidence of a violation of parallel trends.
Subclass 2
Subclass 3
Subclass 4
Subclass 5
Figure 2.6 DD estimation results with time-variant treatment variables (Dependent variable: Koiki region level 1/HHI, 2009-2017)
Notes: Standard errors are clustered at the firm level. Main treatment variable NT takes 1 if a firm had suppliers in hazardous regions of the Nankai Trough Earthquake. A plot represented with “■” means that the treatment variable NT corresponding to the year is significant at least at the 10% level. Vertical bars added to the plots represent the 95% confidence interval.
2.5.3 Results with DDD
In this section, I describe the DDD estimation results for examining H2. First, I examine
the results using the prefecture level 1/HHI as an outcome. As shown in Table 2.11, the
treatment effect is positive and significant at the 5% level in Subclass 3, which consists of
66
medium-sized firms, but not in others. Thus, a positive effect on network diversity averaged
throughout the period after 3.11 is observed only in a subclass covering specific firm size.
Table 2.11: DDD estimation results with time-invariant treatment variable (Dependent variable: prefecture level 1/HHI, 2009-2017)
(1) (2) (3) (4) Subclass 2 Subclass 3 Subclass 4 Subclass 5
beta t-val beta t-val beta t-val beta t-val
NT×EJ×After 0.085 0.691 0.258 2.155 ** −0.031 −0.196 0.198 1.050
Firm FE YES YES YES YES Year FE YES YES YES YES 2-digit FE YES YES YES YES Koiki FE YES YES YES YES 2-digit×Year FE YES YES YES YES Koiki×Year FE YES YES YES YES n 18603 8811 4914 3591
Notes: Significant at the ***1%, **5%, and *10% levels. Standard errors are clustered at the firm level. NT takes 1 if a firm had suppliers in hazardous regions of the Nankai Trough Earthquake. EJ takes 1 if a firm had suppliers in damaged regions of the Great East Japan Earthquake.
Subclass 2
Subclass 3
Subclass 4
Subclass 5
Figure 2.7 DDD estimation results with time-variant treatment variables (Dependent variable: prefecture level 1/HHI, 2009-2017)
Notes: Standard errors are clustered at the firm level. NT takes 1 if a firm had suppliers in hazardous regions of the Nankai Trough Earthquake. EJ takes 1 if a firm had suppliers in damaged regions of the Great East Japan Earthquake. A plot represented with “■” means that the treatment variable NT×EJ corresponding to the year is significant at least at the 10% level. Vertical bars added to the plots represent the 95% confidence interval.
67
I also estimate DDD based on the specification replacing the time-invariant treatment
variable with the time-variant treatment variables—interaction terms between the treatment
variable "N( × |}( and year dummies. The estimation results with time-variant treatment
variables are shown in Figure 2.7. Similar to the DD results, a point represented with a filled-
in square means that the treatment variable corresponding to the period is significant at least at
the 10% level. Vertical bars added to points represent the 95% confidence interval.
As is the case with the averaged treatment effect, treatment effects evaluated on each year
separately are significant only in Subclass 3. In Subclass 3, consistently positive treatment
effects can be observed relatively soon after 3.11. The treatment variable in 2010 is not
significant in all subclasses; hence, this result implies that there is no convincing evidence of
the violation of parallel trends.
Table 2.12: DDD estimation results with time-invariant treatment variable (Dependent variable: Koiki region level 1/HHI, 2009-2017)
(1) (2) (3) (4) Subclass 2 Subclass 3 Subclass 4 Subclass 5
beta t-val beta t-val beta t-val beta t-val
NT×EJ×After 0.007 0.151 0.105 2.169 ** 0.020 0.370 −0.051 −0.780
Firm FE YES YES YES YES Year FE YES YES YES YES 2-digit FE YES YES YES YES Koiki FE YES YES YES YES 2-digit×Year FE YES YES YES YES Koiki×Year FE YES YES YES YES n 18603 8811 4914 3591
Notes: Significant at the ***1%, **5%, and *10% levels. Standard errors are clustered at the firm level. NT takes 1 if a firm had suppliers in hazardous regions of the Nankai Trough Earthquake. EJ takes 1 if a firm had suppliers in damaged regions of the Great East Japan Earthquake.
Second, I examine the results using the Koiki region level 1/HHI as an outcome. As
shown in Table 2.12, the treatment effect is positive and significant at the 5% level in Subclass
3, which consists of medium-sized firms, but not in other subclasses. Thus, the positive effect
on network diversity averaged throughout the period after 3.11 is indeed limited.
68
Subclass 2
Subclass 3
Subclass 4
Subclass 5
Figure 2.8 DDD estimation results with time-variant treatment variables (Dependent variable: Koiki region level 1/HHI, 2009-2017)
Notes: Standard errors are clustered at the firm level. NT takes 1 if a firm had suppliers in hazardous regions of the Nankai Trough Earthquake. EJ takes 1 if a firm had suppliers in damaged regions of the Great East Japan Earthquake. A plot represented with “■” means that the treatment variable NT×EJ corresponding to the year is significant at least at the 10% level. Vertical bars added to the plots represent the 95% confidence interval.
DDD is based on time-variant treatment variables, and the results are shown in Figure 2.8.
Significant treatment effects can be observed in Subclass 3, and the temporal trend of treatment
effects is very close to that obtained with prefecture level 1/HHI. The treatment variable in
2010 is not significant in all subclasses. These results imply that there is no convincing
evidence of a violation of the parallel trend assumption.
2.5.4 Discussion
The DD estimation results show that the geographical diversification of the supply chain
network toward Nankai Trough Earthquake is not observed after 3.11, even if a firm had
suppliers in hazard regions of the Nankai Trough Earthquake before 3.11, regardless of firm
size and the different geographical units used to measure network diversity. This result might
69
imply that, regardless of the direct damage to their incumbent suppliers, learning did not
necessarily progress or lead firms to update their geographical trade diversity.
However, from the DDD estimation results, only in the group of medium-sized firms is
it observed that the cross-prefectural and cross-Koiki district diversification of the supply chain
network progressed after 3.11 if a firm had suppliers in both hazard regions of the Nankai
Trough Earthquake and regions damaged by 3.11. This result can imply that the recognition of
the visible shock explicitly conditioned by the direct shock to firms’ incumbent suppliers, in
other words, the learning based on their own experience during the disruption, impacted
medium-sized firms’ initiative to update their geographical trade diversity.
As described in Section 2.5.1, we do not observe a remarkable difference in industrial
composition across subclasses or geographical unevenness of the location of firms in a specific
subclass. In this sense, this result may merely reflect the heterogeneity of disaster effects
depending on firm size. On the one hand, larger firms’ supply chain networks were diversified
enough to absorb the shock before 3.11, or the magnitude of spatial diversification was trivial
relative to the size of their incumbent supply chain network even though they implemented
predisaster preparation after learning from the experience. On the other hand, it may be difficult
for smaller firms to find alternative partners because of high search costs even though, having
learned from previous disasters, they want to pursue such predisaster planning. In sum, firms
that are, compared to smaller firms, not bound by capacity constraints due to the search and
maintenance cost incurred to obtain alternative suppliers but that are also, unlike larger firms,
dependent on each supplier updated the diversity of their trade networks.
70
2.6 Conclusion
In this study, I examined the impact of the change in risk perception after a disaster on
the supply chain network by looking at predisaster preparation evaluated by whether firms
update their geographical trade diversity; this aspect has rarely been addressed in previous
studies. In the process of examination, I exploited the Great East Japan Earthquake as an
example of an actual and visual disaster shock and the Nankai Trough earthquake, a
forthcoming mega earthquake in West Japan, as an example of an expected and similar disaster
shock. The impact of risk perception on the predisaster preparation regarding the supply chain
can illustrate firms’ behavior under the tradeoffs around network size, capacity constraints, and
uncertainty.
This study contributes to the literature as follows. First, this study examined risk
perception using a rigorous econometric approach based on a quasi-experiment instead of
anecdotal information. In the empirical analysis, I utilized firm-level and long-term network
data from before and after the Great East Japan Earthquake. Second, from the perspective of
economic geography and international management, this study is one of the few investigations
that examine the determinants of diversification of the local industrial structure. In particular,
this study is remarkable because it revealed the causal effect of a disaster shock on trade
diversity, which should be strictly distinguished from the literature examining only associations.
The results are summarized as follows. There was no statistical evidence of the
geographical diversification of the supply chain network to absorb the shock from the Nankai
Trough Earthquake after 3.11, even if a firm had suppliers in hazard regions of the Nankai
Trough Earthquake before 3.11. This result may imply that the recognition of a visible shock
does not necessarily lead firms to update their trade diversity. However, I observed
diversification of the supply chain in the group of medium-sized firms that also had suppliers
in the regions damaged by 3.11. Thus, the recognition of a visible shock, explicitly conditioned
71
by a direct shock on their incumbent suppliers, can influence firms to update their geographical
trade diversity; such updating was carried out by firms that were relatively not bound by
capacity constraints due to search and maintenance cost but that were still dependent on each
supplier.
In sum, only learning from a firm’s own experience can drive a relatively large change in
firm sourcing strategies, such as a recalibration of the supply chain network, and learning from
the experience of others does not necessarily drive a firm’s actions in anticipation of future
shocks. In addition, predisaster policymaking based on learning might be carried out depending
on the balance between capacity constraints and necessity.
Here, I describe the potential future work that may build on this study. First, the empirical
analysis did not explicitly consider the interaction between firms that arises from the disaster
preparation strategy. Risk perception and predisaster preparation in the supply chain network
might require all firms in the supply chain to participate in preparation. In other words, one
firm’s preparation by itself might have no meaning unless the actions are considered by the
whole chain. In this respect, the preparation behavior needs to be examined by explicitly
considering the strategic interaction among the firms as well. Second, to define treatment and
control groups more rigorously, it would be crucial to consider the determinants of the duration
of the transaction relationship and the magnitude of the randomness of the transaction network.
72
Chapter 3
Local Governments’ Actions: Business Continuity Planning11
11 This chapter is revised version of Takano, K., & Morikawa, S. (2020). A Spatial Analysis of Local Administrative Crisis Management: Evidence from Japan (SciREX-WP-2020-#02). SciREX Center, National Graduate Institute for Policy Studies.
73
3. Local Governments’ Action: Business Continuity Planning
3.1 Introduction
In light of a recent spate of terrorist attacks, natural disasters, and pandemic outbreaks
worldwide, the response mechanisms of public administrative bodies to catastrophic events
that can significantly affect the regional socio-economic environment have attracted
considerable attention from researchers. As a key stakeholder in protecting the population from
the effects of extreme events, there is a constant focus on the performance of public
administrations in terms of the (in)efficiency of their actions in the response and recovery
phases. In fact, past crises, including the 9/11 attacks and Hurricane Katrina in the United States,
have shed light on the problems with existing practices, calling for an improvement in public
crisis management. Following the 9/11 terrorist attack in 2001, particularly, there was increased
interest in the research community regarding matters related to governance capacity and
governance representativeness, revealing failures in crisis management (Terry and Stivers,
2002; Christensen et al., 2016). Hurricane Katrina in 2005, by contrast, promoted discussion
on coping with natural hazards, which again called for a reconsideration of existing systems
(Raadschelders and Lee, 2011). In particular, there has been extensive debate regarding crisis
management to minimize the damage from unexpected and large-scale accidents, and how to
continue their tasks (Henstra, 2010).
While the post-crisis response of the government directly affects the recovery of the
society after the crisis, scholars have also emphasized the significance of planning or
preparedness to the crisis in various sectors (Col, 2007; Keim, 2008; Eriksen and Prior, 2013).
More recently, the global outbreak of COVID-19 poses the problem regarding the operation of
governments at all levels under the severe trade-off given the health, economic and social
challenges it raises (OECD, 2020). In tackling the current crisis around governments caused
by the pandemic, the discussion on local governments’ preparedness for various types of
74
disruption including weather-related natural disasters is quite informative (Dzigbede, Gehl, and
Willoughby, 2020; Entress, Tyler, and Sadiq, 2020).
This study quantitatively reveals the drivers and deterrents in developing municipal
business continuity plans (BCP) based on data obtained from the Survey on Development of
Municipal Business Continuity Planning, covering Japanese municipalities. As the whole
territory of Japan is vulnerable to natural disasters owing to its climatic, geological, and
topographical characteristics, the country is under constant threat of natural disaster. In
particular, the Great East Japan Earthquake, which occurred on March 11th, 2011, was a
particularly complex disaster, with the effects of the earthquake compounded by a tsunami and
a nuclear accident, causing enormous human and economic damage. Additionally, another
huge earthquake, the Nankai Trough Earthquake, is expected to occur within the next 30 years.
Therefore, this study examines the differences in the preparation of various municipalities for
future earthquakes: whereas some municipalities developed their BCP immediately after the
Great East Japan (3.11) earthquake, others did so only under the expectation of future shocks.
Methodologically, we explicitly examine the spatial diffusion of BCP development
among municipalities. Although the existence of policy diffusion, a phenomenon in which a
government’s policy choices are influenced by those of other governments, has been widely
examined in the context of public administration and general political science (e.g., Shipan and
Volden, 2012), to the best of our knowledge, there are few empirical investigations of policy
diffusion in crisis management. This study empirically tests the existence and magnitude of the
diffusion by relying on a spatial econometric technique.
The rest of this study is organized as follows. Section 3.2 reviews the literature regarding
critical management in public administration and policy diffusion. Section 3.3 briefly describes
recent and expected disaster shocks and the institutional background of municipal business
75
continuity planning in Japan. Section 3.4 describes the dataset and the analytical framework.
Section 3.5 reports the estimation results. Section 3.6 concludes.
76
3.2 Literature Review
3.2.1 Crisis management and emergency preparedness
In the literature on public administration, the importance of local government’s roles for
recovery and continuity of administrative functions during emergencies such as natural
disasters and terrorism has been strongly recognized since the early 2000s, centering around
the US. For instance, the Public Administration Review organized a special issue in 2002 with
an article by George H. W. Bush, then President of the US, calling for a revisit of the various
facets of democratic governance, government administration, and public management.
Attention turned to the practical response and preparation in the field—hierarchical and vertical
coordination, roles of the Federal Emergency Management Agency (FEMA), and local
municipalities, etc.—, comprehensively discussing public administration following huge
disasters, motivated by the experience of Hurricane Katrina.
Following seminal articles after Hurricane Katrina, there have been several representative
investigations in the public administration literature that descriptively discuss various types of
local government strategies to deal with catastrophic disasters. Among several facets of crisis
management, emergency preparedness is considered to be an important determinant of crisis
responses. For example, Col (2007) compared crisis management systems and their
consequences around Qinglong in China, which was damaged by the Tangshan earthquake in
1976, and New Orleans in the US, damaged by Hurricane Katrina. It showed descriptively that
there was no loss of life in Qinglong despite the region’s proximity to the epicenter, because
of detailed pre-disaster planning; however, the damage of Hurricane Katrina was magnified by
imprecise and inadequate strategies. Henstra (2010) reviewed past operationalization of
preparedness and conceptually divide emergency management into four areas: preparedness,
mitigation, response, and recovery. It identified 30 elements of a high-quality local emergency
management program, among which 12 elements on preparedness consist of the proposed
77
criteria. From a practical perspective, academics and practitioners have been worked together
to promote emergency planning with risk assessment methodology in local governments. For
example, Somers and Svara (2009) discussed how to develop and operate pre-disaster planning
within a universe of daily operational needs.
While these studies have stimulated many successful frameworks and provided guidance
for grass-root applications, we observe two issues that should be further explored in this field.
First, while scholars admit the importance of municipalities’ roles, there is still a gap in the
evidence on the determinants of municipalities’ policy adaptation and behavior. Mehiriz and
Gosselin (2016) investigated the level and determinants of Quebec municipalities’
preparedness for weather hazards and response to related weather warnings. For preparedness,
they found that municipalities’ capacity, population support for weather-related policies, and
the risk of weather-related disasters are important factors to explain the preparation level while
discussing the direct and indirect effects of these factors. They focused not only on the capacity
to respond (financial, organizational 12 , and human-resource-related) but also
sociodemographic determinants, such as population support for weather-related policies. While
scholars have long argued that municipalities’ mutual collaboration or learning is important
both before and after crisis occurrence (Christensen et al., 2016), they have not yet incorporated
the policy diffusion process of crisis management among local governments, which we
elaborate on below.
Second, quantitative examination of the preparedness in local governments has seldom
been conducted. As one of the exceptions, Okura et al. (2019) assessed the association between
the completion of residential evacuation planning and local socioeconomic characteristics in
12 Wang and Kuo (2017) emphasized the importance of the strategic styles of public managers on organizational capacities in their crisis management.
78
Japan. Our investigation complements this literature by focusing on and spatially analyzing
pre-disaster planning and organizational operations inside local government.
3.2.2 Policy diffusion as an implicit collaboration
Past studies of crisis management in federal states such as the United States and Canada
have been argued that collaboration among stakeholders such as different level of governments
and voluntary organizations are especially important partly because authorities related to crisis
management are decentralized among states and relevant departments (Andrew and Carr, 2013).
Therefore, the studies have operationalized collaboration like government officials’
perceptions on vertical and horizontal coordination (Gerber and Robinson, 2009) or volunteer
involvement in emergency management assistance and planning (Brudney and Gazley, 2009).
Although such an explicit relationship among stakeholders is definitely the basis of crisis
management, for enhancement of preparedness to crisis implicit relationship also works.
Especially, the implicit relationship is important in unitary states, wherein a severe crisis
governmental collaboration is led institutionally by the superior governments, but the local
governments play important roles in local reactions delegated by the central government.
In his study of biotechnology and pharmaceutical industries, Powell (1998) stated that
inter-organizational linkages are not only “formal contractual relationships, as in a research
and development partnerships or a joint venture,” but also “informal, involving participation
in technical communities.” It further states that “both mechanisms are highly salient for the
transfer of knowledge and are reinforcing.” Since crisis management also needs knowledge-
seeking and knowledge-creation attitudes of the stakeholders like these industries (Somers and
Svara, 2009), collaboration must be enforced not only with formal and explicit collaborations
but also informal collaborations.
79
In this study, we observe such informal or implicit collaborations under the framework
of policy diffusion, which has received much attention in political science and public
administration (e.g., O’Toole and Meier, 2014). As is the case in other fields of social science
such as economics, policy diffusion in the context of public administration is defined as a type
of social interaction such that a government’s policy choices are influenced by those of other
governments (Shipan and Volden, 2012). For example, policy diffusion can emerge as the
spillover of antipollution measures, tax competition between municipalities, and pressure on
European Union countries facing debt crises to adopt austerity measures by other member
governments (Brueckner, 2003; Shipan and Volden, 2012). Such informal or implicit policy
collaborations are achieved without formal relationships among governments. Local
governments may want to attract more residents to their towns so that they compete for better
services. In other cases, to achieve the same goal, they may imitate policies from other
municipalities. Likewise, crisis preparation measures can also diffuse among local
governments through these mechanisms even without a formal relationship among them.
The discussion on policy diffusion has progressed mainly from an academic perspective
by examining its mechanism and effect on local government performance. The empirical model
in the policy diffusion literature typically relies on the specification such that a government’s
policy choice as a dependent variable is regressed on independent variables including the
neighboring governments’ choice and a government’s own characteristics. In their classic
quantitative study, Shipan and Volden (2008) revealed the mechanism of policy diffusion,
exploiting the case of a municipal level antismoking policy in the United States. They
decomposed the mechanism of diffusion into four channels—learning from earlier adopters,
economic competition among proximate cities, imitation of larger cities, and coercion by state
governments—and examined the effect of each channel on the duration until the development
80
of policy. Their estimation results showed heterogeneity in the persistence and magnitude of
each channel’s effect on the duration depending on the size of municipalities.
Methodologically, however, if one employs a simple estimation method such as the
simple ordinary least squares (OLS) method to estimate the model based on this kind of
specification, the estimated parameter corresponding to policy diffusion will be biased because
it ignores the simultaneity of policy choice (Anselin, 1988). One of the representative
countermeasures against this endogeneity problem can be the spatial econometric approach.
While applied studies employing spatial econometrics have accumulated in the field of political
science as well as general regional science, from the late 2000s, this technique has seldom been
employed in public administration literature (Cook, An, and Favero, 2019)13. In our study, we
statistically examine the cross-jurisdictional policy diffusion in the development of municipal
BCP using the spatial binomial probit model. In this sense, our empirical investigation also has
a methodological contribution to the literature of public administration, as well as a disciplinary
one.
Finally, in the context of disasters, the fact that they have geographical distributions in
their occurrence, the influence of experience gained through other regions’ disasters attracted
Onuma et al.’s (2017) attention. While they use death tolls during both natural and
technological disasters and their unit of analysis is country, a decrease in death in future
disasters can be interpreted as learning from others’ experiences. They found the adaptation
effect only for natural disasters and its marginal impact was relatively higher for higher-income
countries. Still, the estimated model in their study is a fixed-effects model, from which we
expand the conventional model to the one including spatial correlations.
13 One of the exceptions, Oyun (2017) tested the interstate diffusion of the expenditure on Home and Community Based Services with the spatial panel data model. The author showed positive interdependence in state HCBS expenditures that is contingent on similarities in citizen ideology between states.
81
3.3 Institutional Background
Triggered by recent big disasters including the 3.11, Kumamoto Earthquake in 2015 and
several extreme typhoons, and the future risk of the Nankai Trough Earthquake I described also
in Chapter 2, the importance of the recognition of the fragile nature of the national land against
natural disasters, and policymakers to overcome this vulnerability has been strongly
emphasized (Ranghieri and Ishiwatari, 2014) in Japan. In more recent policy trends, pre-
disaster preparation to parry the damage, as well as post-disaster recovery, have been gathering
remarkable attention (CAO, 2018). While the typical pre-disaster preparation can be business
continuity planning in private sectors, district continuity planning (DCP) has also been
proposed and activated as a planning framework integrating not only firms but also various
organizations that take part in regional economic activities (Isouchi, 2017).
It is urgently necessary for a local government, a key actor of DCP, to develop its own
municipal business continuity planning as a counterpart of private sectors’ BCP. In light of the
fact that 3.11 severely damaged many local government buildings, turning both leaders and
public employees into disaster victims themselves, it is quite important to build their own
business continuity plans to be able to continue performing their ordinary tasks (FDMA, 2015a).
In this context, all the local governments have been required to develop their integrated BCP
including six primary elements, the prescription of surrogate authorities and alternative
facilities, the securement of stockpile and communications, data backup, and the prioritization
of tasks (FDMA, 2015b). To promote the development of municipal BCP, the Japanese central
government has provided various types of support, for example, a manual to develop BCP and
a workshop for municipal government staff. However, despite this support, there were few
municipalities that developed their BCP in the beginning, and some municipalities still have
not developed a BCP even as of 2019.
82
3.4 Methodology
3.4.1 Data
We utilize the data from the Survey on Development of Municipal Business Continuity
Planning conducted by the Fire and Disaster Management Agency in 2015, publicly available
on the web page of FDMA, to investigate the development status of municipal BCP all over
Japan. With this data, we can capture the status of whether a municipality has already
developed its BCP and to what extent the developed plan is precise. The reason why we only
use data from 2015 is that we attempt to mitigate the effect of unobservable (unable to capture
quantitatively, in other words) coercion by higher governments such as a prefectural or national
government. Ideally, it is better to use data observed 2014 or before to avoid this problem.
However, because the survey by FDMA was started from 2015, we cannot observe BCP
development status before 2015. We should note that this is one of the limitations in this study.
As mentioned above, positive intervention toward BCP development by the central government
to lower governments began in earnest after 2015. In this sense, the inclusion of survey datasets
after 2015 can be inadequate to examine municipalities’ crisis management strategies taking
account of their socio-economic conditions and regional disaster risk under the condition that
we cannot precisely capture when a prefectural government strongly ordered the municipalities
to develop their BCP.
In order to examine the drivers and deterrents of municipal BCP, we connect the survey
data to several municipal-level statistics. As shown in Table 1, we construct variables used in
the empirical analysis based on the System of Social and Demographic Statistics covering
fundamental regional socioeconomic characteristics, and the National Land Numerical
Information bundling various geographical information in Japan.
83
3.4.2 Empirical model
3.4.2.1 Probit Model
In our empirical analysis, we regress a binary variable, Q_1_1, which is one if a
municipality answered that its BCP had already been developed before 2015, on several
socioeconomic attributes and regional risk environments defined in Table 3.1 with the binomial
probit model. As below, we briefly explain the definition of each independent variable and its
expected sign.
Firstly, we employ the following four independent variables to explain local governments’
basic capability. To explain the soundness of the local fiscal condition, we employ the financial
capability index, FI. The expected sign of FI is positive because each policy might be
developed and operated more smoothly under sounder fiscal conditions. In addition to FI, we
also explain the flexibility of fiscal management with the ordinary balance ratio,
CURRBALANCE. The ordinary balance ratio becomes smaller if a municipality’s fiscal
management system is more elastic and consequently advantageous in making policies. Thus,
the expected sign of the CURRBALANCE is negative. To control municipality size, we
introduce the natural logarithm of the total population, lnPOP 14 . Not limited to crisis
management, larger municipalities might find it possible to establish diverse and dedicated
departments corresponding to various administrative operations. Thus, the expected sign of
lnPOP is positive. To explain administrative cost, we employ the natural logarithm of
inhabitable land per administrative staff, lnAREAPERGOVEMP. As the geographical range
falling under the jurisdiction of a local government becomes larger, it might be more difficult
14 Another potential variable which can explain municipalities’ capability might be, for example, a dummy variable that distinguishes cities and wards from towns and villages. However, we do not use this dummy variable because it highly correlates with lnPOP (correlation coefficient is nearly 0.8), and lnPOP correlates more with the dependent variable, Q_1_1.
84
to investigate potential regional risk environments and develop pre-disaster planning against
each of them with limited manpower. Thus, the expected sign of lnAREAPERGOVEMP is
negative.
Secondly, we also employ the following three independent variables to explain the
regional risk environment that each municipality faces. To explain geographical closeness to
the sea as a proxy of the magnitude of a tsunami and a high tide risk, we introduce the number
of fishing ports, FISHPORT15. The expected sign of FISHPORT is positive since the coastal
area faces a higher risk of a tsunami and high tides. Likewise, we also introduce a flood hazard
area, FLOODAREA, to capture the risk of floods. The expected sign of FLOODAREA is
positive. We also use a dummy variable indicating municipalities damaged by the 2015
Cloudburst, D_15_RAIN. The 2015 Cloudburst is one of the more recent primary disasters in
Japan, which caused widespread flood damage in the Greater Tokyo district. If damaged
municipalities hurried with developing their BCP, letting the experience of the cloudburst be a
lesson, the expected sign is positive. But the opposite results might occur if they could not keep
up with their administrative operations. To consider the potential risk of a mega-earthquake
and tsunami, we include PROB_EQ, the probability that a given city will experience ground
motion intensity exceeding an upper six on the seven-point Japanese scale within 30 years. The
expected sign of PROB_EQ is positive. Finally, to control the regional heterogeneity of the
development status which cannot be controlled by independent variables we employ,
prefectural dummies are also introduced.
15 It must be more desirable to introduce tsunami hazardous areas as an independent variable as with FLOODAREA. However, unlike flood hazardous areas, geographical information about tsunami hazardous areas is available in only 24 prefectures, half of all prefectures in Japan. Thus, we are constrained to use FISHPORT as a proxy variable. We partially mitigate the impreciseness of this proxy by introducing PROB_EQ, the likelihood of a mega-earthquake.
85
Tabl
e 3.
1 D
efin
ition
of v
aria
bles
Va
riabl
e na
me
Def
initi
on
Sour
ce
Q_1
_1
1 if
a m
unic
ipal
ity h
as a
lread
y de
velo
ped
BCP
FDM
A (2
015)
FI
Fi
nanc
ial c
apab
ility
inde
x M
IC (2
015a
) CU
RRBA
LAN
CE
Ratio
of t
he c
urre
nt b
alan
ce [%
] M
IC (2
015a
) ln
POP
Nat
ural
loga
rithm
of p
opul
atio
n [p
erso
n]
MIC
(201
5b)
lnA
REA
PERG
OV
EMP
Nat
ural
loga
rithm
of i
nhab
itabl
e la
nd [h
a] p
er a
dmin
istra
tive
staff
mem
ber [
pers
on]
MIC
(201
5a),
MIC
(201
5b)
FISH
PORT
N
umbe
r of f
ishin
g po
rts
MLI
T (2
006)
FL
OO
DA
REA
Fl
ood
haza
rd a
rea
[m2]
M
LIT
(201
2)
D_1
5_RA
IN
1 if
a m
unic
ipal
ity w
as d
amag
ed b
y th
e 20
15 C
loud
burs
t CA
O (2
015)
EQ_P
ROB
Prob
abili
ty t
hat
a gi
ven
city
will
exp
erie
nce
grou
nd m
otio
n in
tens
ity e
xcee
ding
an
uppe
r six
on
the
seve
n-po
int J
apan
ese
scal
e w
ithin
30
year
s N
IED
(201
4)
FDM
A (2
015)
: Sur
vey
on D
evel
opm
ent o
f Mun
icip
al B
usin
ess C
ontin
uity
Pla
nnin
g M
IC (2
015a
): A
nnua
l Sta
tistic
s on
Loca
l Pub
lic F
inan
ce
MIC
(201
5b):
Nat
iona
l Cen
sus
MLI
T (2
006)
: Nat
iona
l Lan
d N
umer
ical
Info
rmat
ion
(Fish
Por
ts)
MLI
T (2
012)
: Nat
iona
l Lan
d N
umer
ical
Info
rmat
ion
(Flo
od H
azar
d A
rea)
CA
O (2
015)
: App
licat
ion
Situ
atio
ns o
f Disa
ster R
elie
f Act
N
IED
(201
4): P
roba
bilis
tic S
eism
ic H
azar
d M
aps
86
We exclude municipalities corresponding to the following special cases from our dataset.
Firstly, we exclude all municipalities in Hokkaido (90.4% of all municipalities had already
completed the BCP development before 2015), and those in Tottori (all municipalities had
already completed the BCP). This is because the ineligibility in examining municipalities’
crisis management strategies taking account of their socio-economic conditions and regional
disaster risk, as mentioned in Section 3.4.1, and the infeasibility of the adequate estimation of
standard error of prefectural dummies corresponding to these two prefectures16. Secondly, due
to the unavailability of data, six municipalities within 30km from Fukushima Daiichi Power
Plant which were severely damaged by 3.11 are also excluded. Finally, we exclude 34
municipalities without valid responses17.
3.4.2.2 Spatial Probit Model
Following the discussion in Section 3.2.2, the spatial diffusion of BCP development can
be captured by examining the existence of the similarity of the BCP development status among
neighboring municipalities. We capture this similarity (spatial dependence, in other words)
with the probit model based on the random utility as follows:
!!,# = #!,#$ + &'(!$!$
%
$&'+ )# + *!,#,
(!$ = ,1.!$(
01.!$ < 50[km]
09:ℎ<=(0><
(3.1)
16 Due to multicollinearity problem, the estimate of standard error became extremely large in comparison with that of prefectural dummies for other prefectures. 17 For example, if a municipality answered that its BCP has not been developed in 2016, 2017, or 2018 whereas it answered that its BCP had already been developed in 2015, we regard this response as invalid.
87
where!!,#is a utility of municipality0in prefecture?when it develops municipal BCP,#!,#is
a vector of regional characteristics,(!$ is a weight representing geographical proximity (spatial
weight) between municipality0and@,& is a parameter indicating the magnitude of similarity
among neighboring municipalities in BCP development status, )#is prefecture fixed effect
captured with a prefectural dummy, *! is an error term which follows standard normal
distribution. If!!,# > 0, municipality0develops municipal BCP. It can be assumed that&is
positive if the development status is similar among neighboring municipalities. We
define(!$ as the inverse distance squared by analogy with the gravity model, and the spatial
weight matrix is row-standardized when we estimate the model. For avoiding the imprecise
estimation of parameters due to a dense spatial weight matrix, we set a 50km threshold. We
employ generalized method of moments (GMM) proposed by Pinkse and Slade (1998) because
the GMM is a more robust method than the maximum likelihood in the sense that it does not
require the normality assumption in the process of parameter estimation.
88
3.5 Result
3.5.1 Drivers and Deterrents of BCP Development
We summarize the descriptive statistics of variables used in the probit model estimation
in Table 3.2. To save space, we do not show the descriptive statistics about prefectural dummies.
The average of the dependent variable Q_1_1 is 0.283. This result shows that only less than
30% of all 1504 municipalities had already developed their BCP in 2015. The variance inflation
factor (VIF) is less than three for each variable, so we have little concern about the problem of
multicollinearity. In Table 3.3, we show the regression result of the a-spatial probit model.
Table 3.2 Descriptive statistics n mean sd min max vif q_1_1 1504 0.283 0.45 0 1
FI 1504 0.529 0.28 0.05 2.09 2.802 CURRBALANCE 1504 86.758 6.484 48.9 113.6 1.638 lnPOP 1504 10.22 1.48 5.182 15.131 2.653 lnAREAPERGOVEMP 1504 2.964 0.884 −0.305 5.894 3.178 NUMGYOKO 1504 0.515 1.672 0 22 1.232 SHINSUISUM 1504 16.484 41.48 0 664.063 1.434 D_H27_GOUU 1504 0.015 0.12 0 1 1.253 PROB_EQ 1504 0.053 0.083 0 0.574 2.225
Table 3.3 Estimation result of the a-spatial probit model beta zval
FI 0.09 0.395 CURRBALANCE −0.015 −1.926 * lnPOP 0.252 5.654 *** lnAREAPERGOVEMP −0.128 −1.671 * NUMGYOKO −0.016 −0.648 SHINSUISUM 0.002 2.035 ** D_H27_GOUU −0.381 −1.133 PROB_EQ 0.394 0.617 (Intercept) −1.002 −1.387
Prefecture dummies YES PseudoR2 0.196 AIC 1545.949 n 1504
Notes: The dependent variable is "#$%&'_1_1!,#* , where '_1_1!,# is a dummy variable that takes 1 if municipality + in prefecture ,had already developed its BCP in 2015. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels.
89
We first summarize the results of municipal capability. The regression coefficient of the
financial capability index FI is not statistically significant in any level of significance. On the
other hand, the ordinary balance ratio of the CURRBALANCE is negatively significant at a 5%
level, and this result can indicate the positive association between the likelihood of BCP
development and the flexibility of the local fiscal management system. The natural logarithm
of the population lnPOP is positively significant at a 1% level, so there can be a strong
relationship between the municipality size and the likelihood. The natural logarithm of
inhabitable land per administrative staff lnAREAPERGOVEMP is negatively significant at 10%.
Thus, it might be said that higher an administrative cost is associated with a lower BCP
development status. To summarize, the hypotheses about independent variables related to the
municipal capability that we presented in Section 3.4.2.1 are supported except for the financial
capability index.
Secondly, we explain the results regarding regional disaster risk environments. The
number of fishing ports FISHPORT is not statistically significant. This result shows that there
cannot be a remarkable association between the proximity to the sea and the probability that
municipal BCP had already been developed. On the other hand, the flood hazardous area
FLOODAREA is positively significant at a 5% level, and this result indicates a positive
relationship between a larger flood hazardous area and a higher BCP development status. The
dummy variable indicating municipalities damaged by the 2015 Cloudburst D_15_RAIN is not
statistically significant. Thus, we cannot observe a remarkable association between the actual
experience of a flood hazard and the likelihood of BCP development. PROB_EQ explaining
the potential risk of a mega-earthquake and tsunami is not statistically significant. Eventually,
the hypotheses about regional risk environments in Section 3.4.2.1 are not supported except for
flood hazards.
90
Figure 3.2 Estimated coefficients of prefectural dummies with the a-spatial probit model
Notes: The dependent variable is "#$%&'_1_1!,#* , where '_1_1!,# is a dummy variable that takes 1 if municipality+in prefecture,had already developed its BCP in 2015. The detailed model specification is shown in Table 3.2. The reference group consists of municipalities in Tokyo. A prefecture drawn with thick line has a statistically significant fixed effect at least 10% level.
We plot the regression result of prefectural dummies and their statistical significance in
Figure 3.2. In this estimation we use municipalities in Tokyo (69.4% of all municipalities had
already completed the BCP development before 2015) whose BCP development status was the
highest excluding Hokkaido and Tottori as a reference group. Except for a few exceptions,
estimated coefficients are negative, and this result can imply that there is a regional difference
91
in the BCP development status between Tokyo and other prefectures. Although it is difficult to
find out a consistent spatial trend, it can be prefectures facing the Sea of Japan and the East
China Sea or inland prefectures that have a strongly negative coefficient. On the other hand,
prefectural dummies are not significant in prefectures located in Western Japan and facing the
Pacific Ocean like the Miyazaki, Kouchi, Tokushima, and Shizuoka prefectures, and these
prefectures can be roughly included in the hazardous area of the Nankai Trough Earthquake.
In addition, the estimated coefficient is statistically insignificant in Miyagi and Iwate damaged
by 3.11. However, we can observe a negatively significant coefficient even in prefectures
included in damaged or hazardous areas, which implies that there can be an interesting variation
of BCP development status even within damaged or hazardous areas.
3.5.2 Spatial Diffusion of Municipal BCP Development
3.5.2.1 Spatial Pattern of BCP Development Status
In advance of the regression estimation of the spatial probit model, we visually show the
spatial trend of BCP development status in 2015 with Figure 3.3. As observed, in many cases,
neighboring municipalities developed their BCP in a lump, rather than developing it in isolation.
In particular, we can observe a large spatial cluster of municipalities whose BCP had already
been developed in the coastal area of the Chubu district and in the southern part of the Kanto
district. On the other hand, we cannot observe the case that all municipalities in a specific
prefecture had already developed their BCP.
To test the magnitude of the spatial similarity of BCP status, we utilize join count statistics.
With the statistics, we can test whether same-color joins occur more frequently than would be
expected if the zones were labelled in a spatially random way (Cliff and Ord, 1981; Bivand et
al., 2015). The results of join count test are shown in Table 3.4. The first result shows more
undeveloped/undeveloped municipality joins than would be expected.
92
Figure 3.3 Spatial trend of BCP development status
Notes: Municipalities in yellow had already developed their BCP in 2015, those in blue had not.
Table 3.4: Join count results Same color stat Expectation Std. deviate
Undeveloped 409.286 381.889 10.928 *** Developed 85.028 60.389 12.240 ***
Notes: n=1493. Tested variable is a dummy variable that takes 1 if a municipality had already developed its BCP in 2015. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels. Isolated municipalities with no neighbor based on spatial weight defined in Eq. (3.1) (i.e., -!$ = 0, ∀2) are excluded from the sample. The spatial weight matrix is row-standardized
Similarly, the second result shows more developed/developed municipality joins. These
imply the existence of positive spatial dependence in BCP development status.
93
3.5.2.2 Estimation Result of the Spatial Probit Model
We show the regression result of the spatial probit model in Table 3.5. Compared with
the result in the a-spatial probit model shown in Table 3, there is no remarkable change in the
magnitude, significance, and condition of the estimated regression coefficients except for the
natural logarithm of inhabitable land per administrative staff lnAREAPERGOVEMP.
Table 3.5 Estimation result of the spatial probit model beta zval
FI −0.05 −0.214 CURRBALANCE −0.014 −1.836 * lnPOP 0.245 5.364 *** lnAREAPERGOVEMP −0.061 −0.802 NUMGYOKO −0.004 −0.186 SHINSUISUM 0.002 2.366 ** D_H27_GOUU −0.434 −1.45 PROB_EQ 0.166 0.296
ρ 0.379 2.347 ** (Intercept) −1.067 −1.52
Prefecture dummies YES n 1493
Notes: The dependent variable is "#$%&'_1_1!,#* , where '_1_1!,# is a dummy variable that takes 1 if municipality + in prefecture ,had already developed its BCP in 2015. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels. Isolated municipalities with no neighbor based on spatial weight defined in Eq. (1) (i.e., -!$ = 0, ∀2) are excluded from the sample. The spatial weight matrix is row-standardized.
In Table 3.5, the estimation result of coefficient&corresponding to BCP development
action in neighboring municipalities is additionally shown. In consideration of the property of
parameter&that it takes a value within the range of0 < & < 1if the trend of the outcome is
similar among neighboring samples, the estimates of & is not so large. However, the
estimated&is positively significant at the 5% level, which can imply that there is a remarkable
similarity of BCP development patterns between neighboring municipalities even after
controlling the difference of regional capabilities and risk environments, as well as the
94
prefectural fixed effect. To summarize, the hypothesis we presented in Section 3.4.2.2 is
supported18.
Figure 3.4 Estimated coefficients of prefectural dummies with spatial probit model
Notes: The dependent variable is "#$%&'_1_1!,#* , where '_1_1!,# is a dummy variable that takes 1 if municipality+in prefecture,had already developed its BCP in 2015.The detailed model specification is shown in Table 3.5. The reference group consists of municipalities in Tokyo. A prefecture drawn with thick line has a statistically significant fixed effect at least 10% level.
18 The magnitude and significance of the estimated&does not change remarkably even though we use an alternative cutoff of distance. In addition, the structure of this spatial dependence is explained best by the gravity specification. For more information, see Appendix.
95
As is the case of the a-spatial probit model, we plot the regression result of prefectural
dummies and their significance in Figure 3.4. Although the spatial pattern of estimates does
not change from that with the a-spatial probit model shown in Figure 3.2, the absolute value of
the estimated coefficient becomes entirely smaller. In addition, the number of statistically
significant estimates decreased. This result can imply that the prefecture-specific trend is
partially controlled by the spatial effect additionally introduced by the spatial probit model.
3.5.3 Discussion
Based on both estimation results of the a-spatial probit and that of the spatial probit, two
variables related to regional capability, the total population lnPOP and the ordinary balance
ratio CURRBALANCE, are robustly significant while the financial capability index FI is not
statistically significant in any empirical model. These results can imply that it is the flexibility
of the local fiscal management system and administrative capability in terms of scale that are
significantly associated with municipal BCP development, rather than the soundness of the
fiscal condition.
About variables related to regional risk environments, only the flood hazardous area
FLOODAREA is robustly significant. There might be two reasons why the number of fishing
ports FISHPORT is insignificant. The first reason is that FISHPORT is simply an inadequate
variable to explain the risk of a tsunami and high tide in the coastal area. Unless we can find
an alternative proxy variable, we must dare to accept the loss of sample size and implement the
empirical analysis on limited municipalities where the information of a tsunami hazardous area
is available. The second reason is that the coastal area effect is partially controlled by
introducing the prefecture fixed effect, which is also the case of PROB_EQ. This might be
because the risk of an earthquake and tsunami is generally not local rather than global and is
consequently controlled by prefectural dummies. Although we cannot observe a significant
96
association between the experience of the 2015 Cloudburst D_15_RAIN and BCP development
in 2015, we cannot entirely reject the existence of the lagged effect of the experience. To
examine this lagged effect, we need to further employ the panel data model which can eliminate
the effect of unobservable coercion by a higher government because the coercion can strongly
correlate with BCP development and it is time-varying and unobservable.
From the result of prefectural dummies, it can be implied that there is a cross-prefectural
difference in the BCP development status even after controlling the difference of regional
capability and risk environments. Compared to municipalities in Tokyo as a reference group,
the BCP completion rate is remarkably low in municipalities in prefectures facing the Sea of
Japan and the East China Sea or inland prefectures. These municipalities can correspond to the
regions where a catastrophic disaster has not occurred recently. In addition, they are
geographically distant from 3.11 damaged regions or the Nankai Trough hazardous regions and
have quite a different natural condition. Therefore, these factors might be associated with lower
BCP development status. On the other hand, the BCP completion rate of prefectures located in
Western Japan and facing the Pacific Ocean is not significantly different from that of Tokyo.
Since these prefectures can be roughly included in the hazardous area of the Nankai Trough
Earthquake, it can be implied that earlier crisis management planning toward future disaster
risk has progressed in these prefectures. We can also observe the insignificance in Miyagi and
Iwate that was damaged by 3.11. This result can imply the earlier development of BCP letting
the experience of a huge disaster be a lesson.
Finally, the estimation result of the spatial probit model can show the significant
similarity of the BCP development status among neighboring municipalities even after
controlling the regional attributes mentioned above. One of the reasons why we obtain this
result can be the cross reference among neighboring municipalities. This can support policy
diffusion stemming from the reliance on the precedent in other governments and the necessity
97
of a cross-jurisdictional policy making with the view of efficient and seamless actions against
emergencies. In addition, the manual for local administrative staff to develop BCP, FDMA
(2015b), had already proposed that neighboring municipalities should cooperate to develop
their BCP taking advantage of regional collaboration in daily administrative operations. This
fact might support the existence of early cases of cross-jurisdictional BCP development before
2015. In addition, municipalities might have developed a broad-based BCP following
incumbent network-type crisis management framework like a cross-jurisdictional disaster
prevention agreement for mutual disaster relief (e.g., accommodating for emergency shelter
facilities, supply of relief goods and equipment, and dispatch of medical and administrative
staffs) between (neighboring) cities toward.
98
3.6 Conclusion
The literature on crisis management has grown in the past couple of decades as the
vulnerability of public administrative bodies in the face of catastrophic events has attracted
considerable attention following large-scale natural disasters and terrorist attacks in the 2000s.
However, the literature has been relying on descriptive analysis by exploiting a few cases, and
thus, a quantitative examination of crisis management and its operation by local government,
remains scant. To fill this gap, we empirically investigated the drivers and deterrents in
developing municipal business continuity planning with data obtained from the Survey on
Development of Municipal Business Continuity Planning, covering Japanese municipalities.
This study makes two additional contributions to the literature. First, it analyzes the case of
Japan, among the most disaster-prone countries in the world, but one that has hardly been
examined in the literature regarding public administration. Second, it comprehensively
analyzes two issues, crisis management and policy diffusion, which constitute part of the
central debate in recent public administration literature.
The results obtained with the (spatial) probit model can be summarized as follows. First,
we empirically showed the positive association between the likelihood of BCP development
and local capability, based on municipality size and the flexibility of fiscal management.
Regarding regional risk environments, we also showed a positive association between the
likelihood and the magnitude of flood hazards. Second, there was a remarkable prefecture-level
difference in the BCP development status among municipalities even after controlling for
differences in regional capabilities and risk environments. In particular, a relatively higher level
of BCP development in the Nankai Trough hazardous prefectures might be reflected in earlier
BCP completion, and that of the 3.11 damaged prefecture might be due to lessons learnt from
the experience of 3.11. In contrast, in inland prefectures and those geographically distant from
these damaged or hazardous prefectures, the level of BCP development was significantly lower.
99
Finally, even after controlling for differences in regional characteristics as mentioned above,
we could observe significant spatial dependence in the BCP development status among
neighboring municipalities. This result might support the existence of spatial policy diffusion
in the context of crisis management by the local government.
The examination of policy diffusion in the context of crisis management can be of
significance because of its peculiar feature, namely, the necessity of cross-jurisdictional BCP
to ensure efficient and seamless actions in an emergency. If negative externalities liked to the
extreme event are so high, no single jurisdiction, or country, can manage these on its own. In
this regard, coordination across regions including the quick sharing of information about a
cooperative agreement in supporting post-disaster recovery is essential to avoid disjointed or
contradictory responses, which might incur a collective risk to a region’s population (OECD,
2020). Given local governments’ reliance in their policymaking on the precedent set by other
governments, a driving factor influencing policy diffusion in the public administration
literature, the mechanism should be enforced in tandem with supports for explicit collaboration
among governments.
Future investigations will focus on the following. First, an in-depth examination of the
quality of the developed BCP is necessary, as well as that of the completion of BCP
development. As mentioned earlier, we can also access information on how many primary
elements are included in each municipality’s BCP. To precisely analyze the quality of the
developed BCP, further discussion regarding the priority ranking of the primary elements is
required. In addition, an examination of the quality in terms of feasibility might be also
necessary. Second, there is significant potential to rigorously identify policy diffusion. Another
reason for spatial dependence in BCP development status is the effect of unobservable
characteristics common to neighboring municipalities. This problem in identification is
inevitable as long as we employ a weight that only relies on geographical distance (Topa and
100
Zenou, 2015). Network information capturing social proximity among municipalities, for
example, through some agreement for cooperative administrative operations such as disaster
prevention agreement or the inter-municipal transaction network, can be a good alternative to
illustrate the relationship among municipalities related to policy making.
101
Chapter 4
Acquisition of Source of Resilience towards Industrial Structural Change:
Place-based Innovation Policies19
19 This chapter is revised version of Takano, K., & Okamuro, H. (2019). Local R&D support as a driver of network diversification? A comparative evaluation of innovation policies in neighboring prefectures in Japan (No. E-2019-02). Teikoku Databank Center for Advanced Empirical Research on Enterprise and Economy, Graduate School of Economics, Hitotsubashi University.
102
4. Acquisition of Source of Resilience towards Industrial Structural Change: Place-
based Innovation Policies
4.1 Introduction
Public support to R&D activities of local industries is expected to enhance their
performance because of both process and product innovation. To assist with evidence-based
policymaking, quantitative evaluation of innovation policies under the national initiative with
firm-level data has been conducted intensively in this decade. However, similar policies by
local governments have rarely been evaluated. Variation in public support of policy formation
and its effects across prefectures, as well as the different roles of national and local
governments, is yet to be investigated. Additionally, most studies only evaluate productivity-
related effects of public policies and ignore other aspects of treated firms’ business activities.
This study evaluates the effect of local R&D support programs on firm performance,
comparing three neighboring prefectures in the same district in Japan, by utilizing a large
corporate database based on credit investigation. Our investigation makes three main
contributions. Firstly, we specifically examine the effects of support policies on the regional
and industrial diversification of transaction networks. The inclusion of these network-based
outcomes provides a more in-depth mechanism of the contribution of support policies to local
firms’ market competition.
Secondly, cross-prefectural comparative analysis can reveal how the effect of R&D
support on firms’ performance varies regionally, depending on the support systems and each
prefecture’s industrial and geographical conditions. One of the three target prefectures, A, has
a large industrial agglomeration around world-leading manufacturers. Although the other two
prefectures, B and C, do not have such an agglomeration, they have several manufacturing
sectors of comparative advantage, as well as better access to the largest metropolitan areas in
Japan.
103
Thirdly, we utilize large firm-level panel data which cover over 10,000 manufacturing
firms in each prefecture. The extrapolation of these data enables an empirical evaluation on the
effects of such policy with sufficient statistical power while controlling on firm-level
heterogeneity. Additionally, since this dataset was constructed independently from policy
evaluation, unlike empirical investigations based on survey techniques, there is no concern
about the overestimation of the impact of programs due to the tendency that firms receiving
money are likely to exaggerate the scheme’s benefits (Criscuolo, Martin, Overman, and Reenen,
2019).
104
4.2 Literature Review
4.2.1 Place-based Policies
In the last two decades, place-based R&D support policies, represented by cluster policies,
have been promoted for the benefit of agglomeration externality such as knowledge spillover
between co-located and tied economic agents (Duranton and Puga, 2004). Following this
framework, particularly in the regional innovation systems literature (e.g., Cooke, 2002), the
evolution of local industry’s capability developed by the interaction between the key agents in
a cluster like local firms, research institutes, and local authorities has attracted increasing
attention (Nathan and Overman, 2012).
Based on this argument, firm-level quantitative evaluations on cluster policies under the
national government initiative have been carried out in this decade, targeting, for example,
France (Martin, Mayer, and Mayneris, 2011), Japan (Nishimura and Okamuro, 2011a, 2011b),
Korea (Doh and Kim, 2014), and Italy (Bronzini and Iachini, 2016). However, empirical
investigations on the role and effect of R&D support policies under local government initiatives
are scant (Neumark and Simpson, 2015), though several studies examine this issue in a specific
region (Falck, Heblich, and Kipar, 2010 in Germany) or incorporate a multi-level governance
perspective (Lanahan, 2016 in the US; Fernandez-Ribas, 2009 in Spain; Okamuro and
Nishimura, 2020a in Japan).
An in-depth understanding on the role of local authorities provides insights to illustrate
the changing balance between the central and local governments towards greater
decentralization by contrasting centralized countries with federal countries. The significance
of local authorities’ innovation policies can be justified from the view of multilevel innovation
policy governance (Kitagawa, 2007; Laranja, Uyarra, and Flanagan, 2008; Flanagan, Uyarra,
and Laranja, 2011). As stated in the principles of fiscal federalism (e.g., Oates, 1972), local
policies complement those of central governments by providing regionally adapted and more
105
flexible support, taking advantage of better accessibility to information on local economic
conditions and development trends (Fernandez-Ribas, 2009). Beginning with a special edition
in the 2007 issue of Regional Studies that covered seminal studies examining scientific policy
in each country with program-level (Salazar and Holbrook, 2007) or regional case studies
(Crespy, Heraud, and Perry, 2007; Sotarauta and Kautonen 2007; Koschatzky and Kroll, 2007)
and a historical review (Perry, 2007; Kitagawa, 2007), qualitative analyses with case studies of
local R&D support have been conducted worldwide including a comparison of local
management of biotechnology clusters in Germany, France and Japan (Okamuro and
Nishimura, 2015a).
We complement this literature by incorporating a horizontal comparative analysis
between different R&D support programs conducted by three neighboring prefectures, in
contrast to the vertical comparison addressed in the multi-level governance literature. As
argued by Tödtling and Trippl (2005), no innovation policy can fit all regions due to a wide
variety of regional characteristics. Thus, it is important to compare regional variety in the
implementation, design, and consequence of innovation policies by local authorities.
We also contribute to the place-based R&D policy literature by introducing outcomes
based on inter-firm transaction information. The literature consensus is that the inter-firm
networking and industrial-university cooperation are crucial outcomes of the support programs
(Nishimura and Okamuro, 2011b). However, most studies only focused on networking
activities to create an R&D environment, except for some examining networking for general
business activities (Okubo, Okazaki, and Tomiura, 2016 targeting transaction network
expansion; Freel, Liu, and Rammer, 2019 targeting export behavior).
The primary motivation to examine the effect on transaction networks or export activities
stems from the informality of innovation activities by small business enterprises (Kleinknecht,
1987). Particularly, the inter-firm transaction relationship—namely, exports—can serve as a
106
path of tacit knowledge transfer which may be used to spur innovative activity in the future
through exchanges of goods and services (Golovko and Valentini, 2011; Ganotakis and Love,
2012). In addition, innovation is often an intermediate output to achieve broader goals like the
enhancement of market competitiveness and is rarely an objective in itself (Borrás and Edquist,
2013; Cannone and Ughetto, 2014). We complement this literature by evaluating R&D support
programs under the local government initiative from the viewpoint of diversifying transaction
networks and incorporating an in-depth examination on the compositional shift of supported
SME (small and medium-sized enterprise) s’ business portfolios.
Previous studies on the effects of public R&D support focus on input and output
additionality, which refers to the increase in R&D input and output resulting from public
subsidy (Aerts and Schmidt 2008; Czarnitzki and Lopes-Bento 2013; Czarnitzki and Hussinger
2018). Empirical research on behavioral additionality of public subsidy, that is, changes in the
behavior and strategy of recipient firms, is relatively scarce but diverse. Clarysse et al. (2009)
investigate the changes in internal R&D management process after receiving public funding,
while Okamuro and Nishimura (2015b) find a positive impact of public R&D subsidy on trust
formation in collaborative R&D projects. In this regard, our study may contribute to the
empirical research on public R&D support by considering the diversification of major
customers as behavioral additionality, since the diversification of local firms is especially
important from the viewpoint of local governments that aim to enhance the competitiveness of
local industry.
4.2.2 Diversification of Economic Activities
Another objective of this study is motivated by the discussion on the positive relationship
between diversity in local industry and firm or local economic growth, as shown in the
economic geography and international management literature (e.g., Frenken, Van Oort, and
107
Verburg, 2007; Palich, Cardinal, and Miller, 2000). The literature has emphasized the
importance of both inter-industrial or inter-regional diversity, and intra-industrial or intra-
regional diversity. On one hand, inter-industrial or inter-regional diversity is motivated by a
sector’s absorption ability of idiosyncratic shocks and its adaptability to technological changes
(Rugman, 1979; Fujita and Thisse, 2013). On the other hand, intra-industrial or intra-regional
diversity is motivated by learning from related (but not the same) others (Jacobs, 1969;
Feldman and Audretsch, 1999), and by taking advantage of cognitive proximity, similar
institutions and norms shared between industrially or geographically close firms. The literature
has empirically shown the remarkable relationship between diversity and performance of firms
or regions. The economic geography literature, for example, showed a positive association
between industrial or regional diversity and regional growth (Boschma and Iammarino, 2009;
Boschma, Minondo, and Navarro, 2012), while that on international management showed a
similar association based on firm-level examinations (Nachum, 2004; Qian, Li, Li, and Qian,
2008).
Despite a growing body of empirical literature examining the association between
diversity and economic performance of firms or regions, little is known about what can form
this diversity on a causality level rather than on a correlation level. We fill this gap by
empirically examining whether R&D support under local government initiatives can affect the
regional and industrial diversity of transaction networks of subsidized firms with a rigorous
micro econometric approach. In other words, we examine the feasibility of transaction network
diversification based on local governments’ policy intervention, as Okubo et al. (2016) and
Freel et al. (2019) illustrated.
108
4.3 Investigated Prefectures and Programs
4.3.1 General Characteristics of Each Prefecture
We select the following three prefectures as target regions because they provide
sufficiently large samples of supported firms. In addition, we select neighboring prefectures
located in the same district to avoid comparison between completely different regions. Firstly,
regarding administrative and geographical characteristics, Prefecture A covers one of the three
major metropolitan areas in Japan. Prefecture B has a mountainous topography, so the usable
land area for economic activities is relatively small. Prefecture C has two half-million cities.
These three prefectures are directly connected to the Greater Tokyo Metropolitan District by
the Shinkansen super-express.
We also summarize industrial characteristics of these prefectures based on the latest
results of the Census of Manufactures in 2017. Prefecture A’s manufacturing sector is highly
specialized in transportation machinery, which accounts for approximately 50% of the total
value of manufacturing shipments in A and nearly 40% of that in Japan due to a huge industrial
agglomeration around world-leading manufacturers. Since A’s manufacturing shipment
accounts for about 15% of that of Japan, the total value of shipments of the manufacturing
industries in this prefecture aside from transportation machinery is also quite large.
Whereas Prefectures B and C do not have such a huge industrial agglomeration, both have
some specific industrial characteristics. Compared to the national share, B is specialized in
electronic parts, devices, and circuits (13% of its total value of manufacturing shipments) and
ICT equipment (9%). C is also specialized in transportation machinery (26% of C’s
manufacturing shipments) and electronic sectors (12%), and it also has one of the most famous
industrial clusters of chemical sectors in Japan. In addition, C’s manufacturing sector accounts
for about 5% of Japan’s total value of manufacturing shipments.
109
4.3.2 Overview of R&D Support Programs
4.3.2.1 Decentralization of Japanese Innovation Policy
We firstly review the recent trend of Japanese innovation policy referring to Kitagawa
(2007) and Okamuro et al. (2019). The key actors of the innovation policies for local SMEs in
Japan have been gradually transferred from the central government to local authorities. With
the aim of local economic development, several programs that support local SMEs’ innovation
activities have been implemented by the Japanese government, including the Consortium R&D
Program for Regional Revitalization starting in 1997, the “Science and Technology Basic Plan,”
the “Industrial Cluster Project” in 2001 by the Ministry of Economy, Trade and Industry and
the “Knowledge Cluster Initiative” in 2002 by the Ministry of Education, Culture, Sports,
Science and Technology. The objective of these programs is to promote R&D consortia
involving universities, industry, and governments at the regional level. R&D subsidies for
SMEs was jointly provided and implemented by the Small and Medium Enterprise Agency
under the METI and local governments, particularly the prefectures.
More recently, under the flag of regional revitalization (“Chiho Sosei”), local authorities
have been required to design and implement their own regional growth strategies. This is also
the case for innovation policies: increasing decentralization of the design and implementation
of R&D support programs is expected, which would take advantage of the accessibility to local
demand and conditions.
4.3.2.2 Local R&D Subsidies in the Three Prefectures
Based on the information from applicant guidelines of the three prefectures and the results
of semi-structured interviews with policy managers, we summarized the basic information of
R&D subsidy programs under the local government initiative in Table 4.3.1. Prefecture A’s
ongoing program was founded in 2012 under the direct control of the prefectural government.
110
Table 4.3.1 Support programs provided by each prefecture A B C
Established by prefecture alone Yes No
(supported by SMRJ) No
(supported by SMRJ)
Supports R&D, experiment New businesses and products R&D for practical use
Requirements of collaboration pattern
Collaboration with public research institutes None None
Constraint on firm size None Only SMEs Only SMEs Start-end 2012-present 2007-2017 2007-2017
Maximum budget per project [Yen] 100 million for SME 7 million 5 million
Support after acceptance None Business consultation None Additional adoptions
per firm Available for up to 3 years Available up to 3 times Unavailable for 3 years after previous receipt
Cumulative number of subsidized projects
521 (including non-SME firms) 96 90
Source: Applicant guidelines of the three prefectures and the results of semi-structured interviews
Although Prefectures B and C’s programs were conducted from 2007 to 2017 based on joint
funding by the prefectural government and the Organization for Small and Medium Enterprises
and Regional Innovation in Japan, SMRJ, under the jurisdiction of METI, the fund was
managed at each prefecture’s own discretion.
The objective and design of each prefecture’s subsidy is different in terms of the stage of
R&D procedures that it supports. Prefecture A’s subsidy program is specialized in R&D
support and demonstration experiment subject to cooperation with public research institutes
and universities, whereas new market development is beyond its purpose. In contrast,
Prefecture B’s subsidy supported general matters connected to new businesses and products.
In addition, this subsidy was jointly provided with business consulting support that began one
or two years before the subsidy by a project team organized by specialists of new market
development. Prefecture C’s subsidy supported R&D activities that are particularly connected
to practical use, and new market development is not a purpose. The maximum subsidy per
project in Prefecture A’s program amounts to 100 million yen for SMEs, while that in
Prefectures B and C’s program is 7 million and 5 million yen, respectively. Thus, the maximum
R&D subsidy is much larger in Prefecture A.
111
In all prefectures, public subsidy is limited to within one year (by the end of the fiscal
year), while R&D expenditures for subsidized projects are settled at the end of the projects
after submitting the final report, which may mitigate crowding-out and moral hazard problems.
Moreover, in Prefectures B and C, subsidy recipients are subject to final project evaluation.
These conditions suggest that subsidized projects would be expected to demonstrate some
short-term outcomes such as patents or product prototypes, and that such projects tend to obtain
local subsidies.
112
4.4 Methodology
4.4.1 Firm-level Panel Data
We utilize the inter-firm transaction dataset provided by Teikoku Databank, Ltd. (TDB)
available through the TDB Center for Advanced Empirical Research on Enterprise and
Economy. TDB is a major corporate credit research company in Japan that collects corporate
data through door-to-door surveys. Around 1,700 field researchers visit and interview firms in
every industrial category and location.
The inter-firm transaction database comprises annual transactional relationships among
firms. In 2016, the database included 1,136,203 firms out of 1,629,286 incorporated companies
in Japan, according to the latest Japanese Economic Census in 2016. Thus, during this period,
the database captured inter-firm transactional activities for nearly 70% of all incorporated
companies in Japan. In addition, the dataset is connected to a corporate information database,
COSMOS, so that basic corporate information of each firm is also available. In the face-to-face
interviews, each firm reports up to five suppliers and customers. Since this dataset eventually
includes transaction information from both partners, the number of customers for each firm
often exceeds five. Although it does not capture the transaction amounts and the international
transaction information, this database is superior to similar ones in developed countries, as it
captures the dynamism of the disaggregated supply chain network structure.
In constructing panel data, we firstly collected the lists of subsidized firms from each
prefecture’s website. After cleansing the firm list, we matched it with the inter-firm transaction
databases. Finally, we selected only the manufacturing SMEs with less than 300 employees.
4.4.2 Empirical Procedure
4.4.2.1 Conceptual Framework
We propose the following hypotheses.
113
H1: R&D support enhances industrial diversity of recipient firm’s customer composition.
H2: R&D support enhances regional diversity of recipient firm’s customer composition.
We set up these hypotheses relying on the previously discussed additionality concepts.
Firms that receive public R&D subsidy may increase R&D expenditures (input additionality)
and R&D (output additionality), and therefore exploit new market opportunities in new regions
or industries through new technology and products. Thus, the diversity of customer
composition of recipient firms may be enhanced due to successful innovation. Moreover,
public subsidy may directly affect recipient firms’ business strategy (including that for existing
products and services) toward higher diversification (behavioral additionality) through
technological and marketing support, which is incorporated at least in the support programs of
Prefectures B and C. Hence, public R&D may increase recipient firms’ industrial and regional
diversification of customer composition both indirectly through input and output additionality
and directly through behavioral additionality.
Additionally, we presume several regional differences in the effects on these outcomes.
Firstly, the effect of Prefecture A’s program on network diversity may be the weakest among
the three prefectures since this program is specialized in the support of recipient firms’ R&D
and demonstration experiment rather than new market development. In contrast, the effect of
Prefecture B’s program on network diversity may be the strongest since it supports both R&D
and new market development. Finally, the effect of Prefecture C’s program may be weaker
than that of B’s program because the former solely supports R&D for practical applications.
4.4.2.2 Definition of Outcomes
We define transaction outcomes which capture industrial or regional diversification of
the recipient’s customer composition based on Frenken et al. (2007). The outcomes are
constructed using an entropy measure based on the industrial or regional composition of
114
customers for each firm. With regard to the entropy index, we define diversification outcomes
corresponding to two dimensions: regional or industrial and inter-class (unrelated) or intra-
class (related).
The formal definition of firm0’s inter-industrial customer diversification is as follows:
BCDEFBCG! ='H!) log( L1H!)M
), (4.1)
whereH!)is the share of firm0’s customers included in 2-digit sector, N (customers’ shares
are measured by the number, and not by the volume, of transactions). In a similar way, we
define firm0’s inter-regional customer diversification in the following way:
BCDEFFEO! ='?!* P9N( Q1?!*R
*, (4.2)
where ?!* represents the share of firm 0 ’s customers located in prefecture = (roughly
corresponding to NUTS-2 region). These outcomes take larger values if the share of customers
categorized into each industrial sector or located in each prefecture is nearly uniform, and
smaller values if the share of customers is higher in a specific region or sector. In this sense, a
firm with a large BCDEFBCG or BCDEFFEO has its customers in various sectors or regions.
Intra-class diversity is derived from the decomposable nature of entropy measure. This
decomposition enables us to evaluate whether industrial/regional diversity is especially strong
between subclasses included in a specific parent class, while inter-class diversity simply
evaluates the extent to which the customers are distributed to different classes. Formally,
suppose all five-digit sectorsSfall exclusively under a two-digit sector T) , firm0’s intra-
industrial customer diversification is defined as follows:
BCDFUBCG! ='H!) '?!+H!)+∈-%
P9N( L1
?!+ H!)⁄ M)
, (4.3)
where ?!+ represents the share of firm0’s customers included in 5-digit sector S. In a similar
115
way, let all prefecturesfall exclusively under a Koiki district (roughly corresponding to NUTS-
1 region) which includes four or five prefectures on average, T. , firm 0 ’s intra-regional
customers diversification is given by the following index:
BCDFUFEO! ='H!. '?!+H!.+∈-&
P9N( Q1
?!* H!.⁄ R.
, (4.4)
where H!. is the share of firm0’s customers in Koiki district .. In these indices, since the
entropy measure is calculated based on the composition of subclasses weighted by the share of
their parent class, the diversity within an industrial sector or region with a large share is
emphasized.
4.4.2.3 Estimation Methods
To rigorously evaluate the policy effect in each prefecture, we should exercise caution
when evaluating the effect on treated firms’ outcomes in comparison with counterfactual
situations. Without this comparison, we would face serious empirical concerns due to
confounding factors, which would prevent us from accurately generating counterfactual
situations and eventually implementing the appropriate evaluation.
The first confounding factors are unobservable omitted variables. Not controlling for
firms’ unobservable characteristics in relation to R&D activities, such as high motivation to
develop the business and R&D-friendly corporate culture, and the surrounding environment
can cause upper bias on estimated policy effects because these unobservable variables are
positively correlated with both firms’ outcomes and the likelihood of receiving R&D subsidies.
The second confounding factor is the difference in the attributes between the treatment and
control groups. The likelihood of applying and being chosen for the support program is not
exogenously but endogenously decided. From the perspective of local government, it often
promotes R&D activities of firms that engage in specific business sectors such as high-tech
116
industries. Thus, there can be a difference in firms’ characteristics between the treatment
(adopted) and control (not adopted) groups. Ignoring this difference in empirical evaluation
can lead to biased estimations.
To tackle the first confounding factors, we employ the fixed effect model (FE). With FE,
unlike OLS, we can eliminate the estimation bias due to time-invariant unobservable variables
such as corporate culture and firm location. Formally, we specify the estimation equation based
on FE as follows:
W!/ = &/ + X! +'Y+ZG0+1'_0+10[!/
(
+&3+ \UGG!/ + ]^_`DB!/ + aBCGEb!/
+ #′!/d + *!/ ,
(4.5)
where0represents the firm and :represents the year. &/is the year FE and X! is the firm FE.
With&/, we can control for unobservable factors that are common in year:for all firms, and
all time-invariant unobservable characteristics of firm 0are controlled with X! . W is each
outcome variable defined in section Definition of Outcomes. G0+1'_0+10 , the treatment
variable of our interest, represents the dummy variable corresponding to the duration after the
first adoption for each firm0and period:; for example, if one to three years have passed since
firm0received a subsidy in year:, ZG'_0[!/ = 120. This variable takes zero for all non-adopted
firms in year:. UGGis a dummy variable coded one if the firm received a subsidy more than
once during the observation period, and zero otherwise. ^_`DBis coded one if a firm receives
a subsidy for more than one year in the period, and zero otherwise. BCGEbcaptures the change
in the number of received subsidies. #′!/dis the linear sum of control variables including 2-
20 We use three-year binned treatment variables instead of the treatment variables defined with those multiplied by year dummies. This is because the number of treated firms became quite small (one or two depending on prefecture), especially when counting the number of treated firms whose duration after the first adoption exceeds seven years. This may cause the problem such that the estimated effects only depend on the change in a few firms’ performances.
117
digit industry dummies and interactions between industry and year dummies. The observation
period starts two years before the focal subsidy programs began. Thus, the minimum value
of:is 2010 in Prefecture A, and 2005 in B and C. Eventually, we estimate a two-way fixed
effects model in a DID design such that each observation in the treatment group receives a
treatment at different points in time, as shown in Equation (5)21.
To cope with the second confounding factor, we utilize propensity score matching (PSM).
PSM enables a comparison between the treatment and control groups controlling for the
differences in firms’ characteristics by selecting a subset of untreated firms similar to the
treated firms, based on the likelihood of receiving subsidies. One of the advantages of PSM
compared with OLS multiple regression is that we can exclude the observations that do not
satisfy the common support. In other words, PSM allows us to implement a comparison
discarding the inadequate observations such as those that are absolutely treated or untreated.
Firstly, the likelihood of firm 0 receiving an R&D subsidy is predicted conditional on
observable characteristics. After that, each treated (subsidized) firm is matched with a control
(not subsidized) firm based on the proximity of predicted likelihood. This approach enables us
to construct a virtual situation wherein the treated firms were not subsidized, based on the
matched control firms’ sample. To obtain the likelihood, we estimate a logit model based on
the following specification:
21 It should be noted that recent studies (Goodman-Bacon, 2018; Callaway and Sant’Anna, 2019) highlight potential problems in the interpretation of two-way fixed effects estimators with variation in treatment timing. For instance, if one uses a simple two-way fixed effects model in a DID design ignoring the difference in the treatment timing, the estimated DID parameter captures not only the before-after difference between treated and untreated observations but also the difference between early and late treatment. Goodman-Bacon (2018) developed a decomposition method to distinguish the first difference from the second and to examine the contribution of each difference for the balanced panel data. As a result of the decomposition utilizing this method, we confirm that the before-after difference between the treated and the untreated firms mostly contributed to the estimated DID parameters in our analysis. For more details, see Appendix E for Chapter 4.
118
H=9e(D=<g:! = 1|i!) = i′!k + !! , (4.6)
The dependent variableD=<g:is a dummy variable taking a value of one ifG0+1'_0+10takes
one, and zero otherwise. The independent variables i! include basic firm characteristics
representing their capability of R&D activities such as sales and the number of employees in
natural logarithm (ln TU`ET,ln E^H), and firm age (UOE). We also include industry dummies
to distinguish between low-tech and high-tech industries. For improving the fitting of the logit
model, we further introduce square terms of each quantitative variable and interaction terms
between each variable. In PSM, we use covariatesi! in three years before the support program
started in each prefecture. Thus, covariatei! in 2009 is used in PSM for Prefecture A, and in
2004 for Prefectures B and C. Prior to PSM, we only selected firms whose transaction
information can be observed throughout the period captured in panel data. To obtain the PS
prediction model with higher generalization performance (i.e., to avoid the over-fitting of the
prediction model), we conduct the stepwise variable selection based on Akaike's Information
Criterion, AIC.
After the prediction, we match the treatment firms with the control firms based on 10
nearest-neighbor matching (10NN). 10NN matches each treatment firm with approximately ten
control firms for closer distance measured with predicted propensity. We discard the firms that
did not satisfy the common support assumption.
119
4.5 Results
4.5.1 Descriptive Statistics
We briefly explain the results of PSM in each prefecture22. The PSM results are given in
Table 4.5.1. Since the absolute value of standardized bias is smaller than 0.1 for most covariates
in all prefectures, there is no convincing evidence for remarkable differences between the
treatment group and the control group within the limitation of observable covariates. Table
4.5.2 shows the sample size of manufacturing SME panel data with and without PSM, and the
number of the observations whoseG0+1'_0+10is equal to one. “Unbalanced” indicates for
unbalanced panel data without PSM, “Balanced” indicates that with PSM.
Table 4.5.1 Summary of PSM results Prefecture A B C
Matched with covariates in 2009 2004 2004
Predicted prob. with logit
Adoption in 2012-2016
Adoption in 2007-2016
Adoption in 2007-2016
Pseudo-R2 0.195 0.151 0.169 # of covariates with |standardized bias|
larger than 0.1 (# of all covariate)
0 (22) 2 (9)
No covariate with |standardized bias| > 0.15
0 (17)
Notes: Balanced covariates are selected based on the forward-backward stepwise method in logistic regression.
Table 5.2 Sample size of manufacturing SME panel data with/without PSM Prefecture A Prefecture B Prefecture C (1) (2) (3) (4) (5) (6) Unbalanced Balanced Unbalanced Balanced Unbalanced Balanced
3'_) = 1 190 162 85 75 84 66 3*_+ = 1 20 18 56 42 40 28 3,_- = 1 22 18 10 5
N 57614 6041 32875 3744 52049 3120 Notes: “Unbalanced” stands for unbalanced panel data without PSM, “Balanced” stands for that with PSM. 3)./'_)./), 4 = 0,1,2is a dummy variable that takes 1 if 34 + 1, 34 + 2, 34 + 3 years have passed since a firm was first adopted. The 3)./'_)./)rowsshow the number of records corresponding to each duration.
22 We present the detailed procedures and the results of the PSM in Appendix 1.
120
4.5.2 Regression Results
4.5.2.1 Policy Effects on Industry Diversification of Customer Composition
We show the estimation results of the policy effects on subsidized firms’ industry
diversification of customer composition. Table 4.5.3 shows the results of the FE model whose
outcome measure is BCDEFBCG . Common to all regression results, standard errors are
clustered at the firm level to deal with the serial correlation within estimation, and main
treatment variable G0+1'_0+10, S = 0,1,2is a dummy variable that takes 1 if 3S + 1, 3S + 2,
3S + 3 years have passed since a firm was first adopted. We observe positive and statistically
significant lagged effects of the subsidy both in Columns (5) and (6). This result indicates that
the magnitude of inter-industry diversification of subsidized firms became robustly larger in C.
In contrast, we cannot find such a significant change in the magnitude of diversification in
Prefectures A and B.
Similarly, in Table 4.5.4 we show the estimation results of FE models
using BCDFUBCG as the outcome variable. In each prefecture, we find that estimated
parameters and their significance are inconsistent before and after PSM. Due to these
inconsistent results, we cannot find robust evidence about whether the magnitude of intra-
industry diversification became larger in all prefectures. In sum, it can be concluded that H1
holds only for Prefecture C’s results onBCDEFBCG.
4.5.2.2 Policy Effects on the Regional Diversification of Customer Composition
We show the estimation results of the policy effects on subsidized firms’ regional
customer diversification. Table 4.5.5 shows the results of the FE model whose dependent
variable is BCDEFFEO. We observe positive and statistically significant lagged effects of the
subsidy in Column (5). In Column (6), we indeed see a positive effect ofG4_5, whereas the
significance ofG6_7vanished after PSM.
121
Table 4.5.3 E
stim
ation results of F
E (D
ependent variable: IN
TE
RIN
D)
Prefecture A
P
refecture B
P
refecture C
(1)
(2)
(3)
(4)
(5)
(6)
beta
t val
Beta
t val
beta
t val
beta
t val
beta
t val
beta
t val
D_1_3
0.041
0.698
0.083
1.348
−
0.058
−0.427
−
0.035
−0.241
0.013
0.189
0.041
0.532
D_4_6
0.160
1.551
0.151
1.457
0.112
1.026
0.059
0.523
0.360
2.413
**
0.290
1.712
*
D_7_9
0.022
0.167
0.007
0.047
0.290
3.125
***
0.442
3.907
***
MU
LT
I
−0.075
−0.493
0.038
0.339
−
0.060
−0.548
−
0.084
−0.700
IN
DE
C
0.018
0.796
0.041
1.608
0.019
0.26
0.010
0.124
AD
D
−0.071
−1.012
−
0.092
−1.16
0.004
0.031
0.116
0.722
Firm
F
E
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Year F
E
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
2-digit F
E
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
2-digit×
Year F
E
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
PS
M
NO
Y
ES
N
O
YE
S
NO
Y
ES
n
57614
6041
32875
3744
52049
3120
Notes: S
ignificance: ***1%
, **5%
, *10%
. S
tandard errors are clustered at the firm
level. M
ain treatm
ent variable ! !
"#$_!"#!,#=0,1,2
is a dum
my variable that takes 1 if
3#+1, 3#
+2, 3#
+3
years have passed since a firm
w
as first adopted. T
he m
ethod of P
SM
is 10 nearest-neighbor m
atching.
122
Table 4.5.4 E
stim
ation results of F
E (D
ependent variable: IN
TR
AIN
D)
Prefecture A
P
refecture B
P
refecture C
(1)
(2)
(3)
(4)
(5)
(6)
beta
t val
beta
t val
beta
t val
beta
t val
beta
t val
beta
t val
D_1_3
−0.006
−0.133
0.005
0.112
0.108
1.224
0.139
1.695
*
0.040
0.862
0.063
1.078
D_4_6
−0.118
−1.658
*
−0.048
−0.708
−
0.112
−1.331
−
0.028
−0.384
−
0.145
−2.003
**
−0.011
−0.135
D_7_9
−0.131
−1.256
−
0.139
−1.344
−
0.177
−1.592
−
0.278
−2.074
**
MU
LT
I
0.080
2.905
***
0.037
0.625
−
0.127
−0.99
−
0.149
−1.144
IN
DE
C
−0.031
−1.706
*
−0.025
−1.212
0.000
−0.002
0.051
1.330
AD
D
0.025
0.516
0.060
1.095
−
0.126
−1.082
−
0.226
−2.211
**
Firm
F
E
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Year F
E
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
2-digit F
E
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
2-digit×
Year F
E
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
PS
M
NO
Y
ES
N
O
YE
S
NO
Y
ES
N
57614
6041
32875
3744
52049
3120
Notes: S
ignificance: ***1%
, **5%
, *10%
. S
tandard errors are clustered at the firm
level. M
ain treatm
ent variable ! !
"#$_!"#!,#=0,1,2
is a dum
my variable that takes 1 if
3#+1, 3#
+2, 3#
+3
years have passed since a firm
w
as first adopted. T
he m
ethod of P
SM
is 10 nearest-neighbor m
atching.
123
These results indicate that the magnitude of inter-prefectural diversification of recipient
firms C became robustly larger at least seven years after subsidization. In contrast, no
significant changes in the magnitude of the diversification can be robustly observed for the
other prefectures.
In Table 4.5.6, we show the estimation results of the FE model using !"#$%$&'as an
outcome variable. We observe positive and statistically significant sustained effects of R&D
subsidy both in Columns (3) and (4). From these results, we argue that the magnitude of the
intra-Koiki diversification in B became robustly larger for subsidized firms. To specify which
Koiki district B’s subsidized firms diversified their business, we drew tree maps detailing the
regional composition of customers. These maps are drawn using matched panel data obtained
in PSM. The first layer represents Koiki district level regional share of customers, and the
second represents prefectural level regional share of customers. The area of a rectangle
corresponding to the second layer is proportional to the square of regional share of customers.
As shown in Figure 4.5.1, intra-regional diversification of customers, especially within the
Greater Tokyo and Chubu district, can be observed. In Greater Tokyo, customer’s share in the
prefectures surrounding Tokyo (13 in tree map) such as Kanagawa (14) and Saitama (11)
increased, while that in Aichi (23) increased in Chubu. On the other hand, we observe negative
and statistically significant effects on!"#$%$&'in the results on A’s program within three
years of the subsidization, while no significant effects can be observed in the results on C’s
program. Therefore, it can be concluded that H2 holds in all the prefectures except for A. H2
is supported by the estimation results on !"#$%$&' in Prefecture B and those on
!"#&$$&'in Prefecture C.
124
Tabl
e 4.
5.5
Estim
atio
n re
sults
of F
E (D
epen
dent
var
iabl
e: IN
TERR
EG)
Pref
ectu
re A
Pr
efec
ture
B
Pref
ectu
re C
(1)
(2)
(3)
(4)
(5)
(6)
be
ta
t val
be
ta
t val
be
ta
t val
be
ta
t val
be
ta
t val
be
ta
t val
D
_1_3
−0
.060
−0
.970
−0.0
62
−0.9
36
0.
041
0.31
8
0.01
4 0.
113
0.
006
0.09
6
−0.0
12
−0.1
38
D
_4_6
−0
.057
−0
.437
−0.0
84
−0.6
72
−0
.048
−0
.299
−0.0
85
−0.5
01
0.
221
1.71
1 *
0.04
3 0.
466
D
_7_9
−0
.078
−0
.374
−0.0
87
−0.3
81
0.
346
2.97
6 **
* 0.
311
2.23
5 **
M
ULT
I −0
.101
−2
.175
**
−0
.072
−1
.154
−0.0
47
−0.6
05
0.
033
0.35
8
IND
EC
−0.0
25
−0.8
23
−0
.030
−0
.934
0.06
2 1.
046
0.
032
0.53
2
AD
D
0.10
8 1.
379
0.
195
2.51
1 **
−0
.160
−1
.035
−0.2
27
−1.5
21
Fi
rm F
E Y
ES
YES
Y
ES
YES
Y
ES
YES
Y
ear F
E Y
ES
YES
Y
ES
YES
Y
ES
YES
2-
digi
t FE
YES
Y
ES
YES
Y
ES
YES
Y
ES
2-di
git×
Yea
r FE
YES
Y
ES
YES
Y
ES
YES
Y
ES
PSM
N
O
YES
N
O
YES
N
O
YES
N
57
614
6041
32
875
3744
52
049
3120
N
otes
: Sig
nific
ance
: ***
1%, *
*5%
, *10
%. S
tand
ard
erro
rs a
re c
luste
red
at th
e fir
m le
vel.
Mai
n tre
atm
ent v
aria
ble ! !
"#$_!"#!,#=0,1,2
is a
dum
my
varia
ble
that
take
s 1
if 3#
+1, 3#
+2, 3#
+3
year
s hav
e pa
ssed
sinc
e a
firm
was
firs
t ado
pted
. The
met
hod
of P
SM is
10
near
est-n
eigh
bor m
atch
ing.
125
Tabl
e 4.
5.6
Estim
atio
n re
sults
of F
E (D
epen
dent
var
iabl
e: IN
TRA
REG
)
Pr
efec
ture
A
Pref
ectu
re B
Pr
efec
ture
C
(1
) (2
) (3
) (4
) (5
) (6
)
beta
t v
al
beta
t v
al
beta
t v
al
beta
t v
al
beta
t v
al
beta
t v
al
D_1
_3
−0.0
64
−1.8
66
* −0
.073
−1
.846
*
0.22
9 3.
322
***
0.19
2 2.
945
***
−0.0
16
−0.3
84
−0
.038
−0
.917
D_4
_6
−0.1
02
−1.8
48
* −0
.077
−1
.143
0.18
9 2.
017
**
0.19
6 2.
119
**
−0.0
21
−0.2
43
−0
.075
−1
.096
D_7
_9
0.20
3 1.
693
* 0.
222
1.72
6 *
0.03
4 0.
223
0.
172
0.98
6
MU
LTI
−0.1
60
−0.6
71
−0
.107
−0
.484
−0.0
31
−0.6
25
−0
.065
−1
.193
IND
EC
−0.0
38
−2.3
58
**
−0.0
49
−2.5
17
**
0.06
9 1.
911
* 0.
063
1.70
0 *
AD
D
0.08
1 1.
714
* 0.
113
1.85
9 *
−0.1
85
−2.2
51
**
−0.1
15
−1.2
63
Fi
rm F
E Y
ES
YES
Y
ES
YES
Y
ES
YES
Y
ear F
E Y
ES
YES
Y
ES
YES
Y
ES
YES
2-
digi
t FE
YES
Y
ES
YES
Y
ES
YES
Y
ES
2-di
git×
Yea
r FE
YES
Y
ES
YES
Y
ES
YES
Y
ES
PSM
N
O
YES
N
O
YES
N
O
YES
N
57
614
6041
32
875
3744
52
049
3120
N
otes
: Sig
nific
ance
: ***
1%, *
*5%
, *10
%. S
tand
ard
erro
rs a
re c
luste
red
at th
e fir
m le
vel.
Mai
n tre
atm
ent v
aria
ble ! !
"#$_!"#!,#=0,1,2
is a
dum
my
varia
ble
that
take
s 1 if
3#
+1, 3#
+2, 3#
+3
year
s hav
e pa
ssed
sinc
e a
firm
was
firs
t ado
pted
. The
met
hod
of P
SM is
10
near
est-n
eigh
bor m
atch
ing.
126
Regional share of customers of firms in B satisfying
!!"#$_!"#! = 0, ∀& (# of firms: 312)
Regional share of customers of firms in B satisfying
!$_! = 0 (# of firms: 29)
Regional share of customers of firms in B satisfying
!&_' = 0 (# of firms: 18)
Regional share of customers of firms in B satisfying
!(_) = 0 (# of firms: 8) Figure 5.1 Tree maps based on regional share of customers of firms in B
Notes: !!"#$_!"#!, & = 0,1,2is a dummy variable that takes 1 if 3& + 1, 3& + 2, 3& + 3 years have passed since a firm was first adopted. These maps are drawn with matched panel data obtained in PSM. The First layer represents Koiki district level regional share of customers, and the second represents prefectural level regional share of customers Area of a rectangle corresponding to second layer is proportional to square of regional share of customers. Shuto represents Greater Tokyo District.
4.5.3 Discussion
Firstly, we observe significant industry or regional diversification for the subsidized firms
in Prefectures B and C, though only partially, but not in Prefecture A. Specifically, the
enhancement of inter-industrial diversity of the recipient firm’s customer composition,
corresponding to H1, was observed only in Prefecture C although a significant effect could not
127
be observed immediately, only four years after the subsidization. The enhancement of intra-
regional diversity, corresponding to H2, was observed only in Prefecture B. Although we could
observe inter-regional diversification in Prefecture C, a robust effect was observed with a time
lag of seven years after the subsidization. In summary, these results imply that significant
effects of local R&D subsidies on the new market development were not necessarily confirmed
in all prefectures and in all dimensions of market development. In this sense, we should be
cautious in discussing the role of local R&D support as an instrument to gain new business
opportunities. Despite these limited results, we may interpret these results from the viewpoint
of policy design features and local industrial characteristics in each prefecture as follows.
As described in Section “Local R&D Subsidies in the Three Prefectures,” new market
development is beyond the scope of Prefecture A’s subsidy program. Thus, subsidized firms
might have had no motivation to diversify or expand their market when relying on the subsidy.
In addition, Prefecture A has a huge industry agglomeration, which forms an integrated
industry structure within Prefecture A from the upstream to the downstream sectors. Hence,
satisfying the demand of incumbent customers may be enough for them to expand the sales of
new products23.
Secondly, we observe significant and sustained positive effects on subsidized firms’
regional diversification of customers in Prefecture B, and on industrial diversification of
customers in Prefecture C. The results for Prefecture B with the FE model might confirm our
benchmarking results demonstrated by the tree maps. Even after controlling for various FEs,
we observe regional diversification of customers in specific Koiki-districts, including Greater
Tokyo and Chubu. This might be because Prefecture B’s subsidy enabled subsidized firms to
promote the market development of new products as final goods mainly in Greater Tokyo, the
23 Interestingly, only in Prefecture A, we observed positive effect of the subsidy on total factor productivity within certain recipient firms. See Appendix B for Chapter 4.
128
largest consumption area in Japan, and as intermediate goods mainly in Chubu, the largest
industrial area. Regarding Prefecture C, we observe positive effects on subsidized firms’ inter-
industry customer diversification. This result suggests that subsidized firms promoted new
market developments in various sectors, taking advantage of a wide variety of sectors existing
in this prefecture.
Again, it is noteworthy that these empirical results only partially and weakly support our
hypotheses. Hypothesis 1 is supported only for a prefecture and an outcome measure, while
Hypothesis 2 receives support only for an outcome measure in Prefectures B and C,
respectively. This partial support may be due to the trade-off between the advantages and
disadvantages of local governments in designing and implementing public R&D support. On
the one hand, we may expect local authorities to be able to better target subsidies where they
are needed and effective (Oates 1972; Fernandez-Ribas 2009). On the other hand, designing
effective R&D support programs requires a lot of expertise and policy-making skills that may
not be sufficiently available at the local level, especially in Japan, where innovation policy has
traditionally been centralized.
Indeed, a recent study (Okamuro and Nishimura 2020b) focused on Japanese cities
empirically demonstrates that the implementation of local R&D support program depends
significantly on supply-side factors, especially the local authorities’ administrative capability.
Although the R&D subsidy programs in Prefectures B and C were funded by a national public
institution, program design and implementation were entrusted to prefecture governments.
Thus, our empirical results may at least indirectly reflect the trade-off between the advantages
and disadvantages of place-based policies by local governments in a country where top-down
decentralization of regional development policies is ongoing (Okamuro et al. 2019).
We directly estimate the effects of local public subsidy on the recipients’ diversification
of customer composition without considering their innovation outcomes. We miss the measure
129
of innovation outcomes in our estimations due to data constraints. At the end of the section
“Local R&D Subsidies in the Three Prefectures,” we have demonstrated that the payment of
R&D subsidy is conditioned on the final report and evaluation of R&D projects, at least in
Prefectures B and C. Hence, subsidy recipients are expected to show some short-term R&D
outcomes such as patents and prototypes by the end of the same fiscal year. Indeed, we find
that 49.8% (Prefecture A, excluding large firms with more than 300 employees)24, 61.5%
(Prefecture B), and 54.5% (Prefecture C) of subsidy recipients have applied for patents after
the first support, which is a distinctively high ratio for Japanese SMEs. Moreover, subsidy
recipients often obtain not only financial support but also technological and marketing
consultation, which applies at least for the firms in Prefectures B and C. Therefore, due to
behavioral additionality, subsidy recipients may be motivated and induced to engage in new
market not only for new, but also for existing products.
24 This share is measured for the subsidy recipients that are classified as SMEs with the current data. Focusing on 191 subsidy recipients that can be identified as SMEs at the timing of subsidization, the share of subsidized firms that have applied for patents increases to 57.1%.
130
4.6 Conclusion
In this study, we empirically evaluated the effects of R&D subsidy programs under the
local government initiative on firms’ performance, comparing three neighboring prefectures in
the same district in Japan but with different regional characteristics. We also examined whether
local R&D support can drive regional and industrial diversification of inter-firm transaction
networks, as a measure of new market development (both for new and existing products), due
to input, output, and behavioral additionality.
Utilizing large firm-level panel data of local manufacturing SMEs based on credit
investigation, we showed that the effects of subsidy programs were quite heterogeneous
between these three prefectures. Particularly, in sum, we observed significant and sustained
positive effects on the subsidized firms’ regional or industrial customer diversification in
Prefectures B and C, but not in Prefecture A. However, more detailed explanation reveals that
Hypothesis 1 is only supported for one prefecture and one outcome measure, while Hypothesis
2 also only receives support for one outcome measure in prefectures B and C, respectively.
These results as a whole suggest that the subsidized firms in Prefectures B and C
attempted to develop new markets in various sectors and regions based on R&D subsidies, but
not those in Prefecture A. The obtained results imply that the consequences of local innovation
policies may vary, reflecting the industrial structure and geographical positions of each region.
In particular, whether supported firms adopt strategies for business portfolio diversification
largely depends on the structure of the input-output market and its size within the region.
However, since our hypotheses are only partially supported for specific diversification
variables and for specific years after subsidization, we can only draw very cautious
implications from our estimation results.
Our evaluation analysis has the following implications. Firstly, we should pay more
attention to the role of local R&D support as a driver of regional and industrial diversification
131
of local firms’ business activities. Secondly, the duration in which each support program bears
fruit is different among regions. Thus, we should evaluate the policy effects from a long-term
perspective that takes into account the regional economic environment.
Despite these contributions, our study has some limitations. Firstly, we do not explicitly
consider the differences in program design between the prefectures, such as the amount of
subsidy. To overcome this disadvantage, we need more detailed information about each public
support program. Secondly, the identification strategy of policy effects employed in this study
is still insufficient to control time-variant firm characteristics and the selection problem due to
imperfectly exogenous assignment of treatment. We can alleviate this problem with detailed
information about selection procedures and criteria. Such information could enable us to
conduct more rigorous evaluation with, for example, the regression discontinuity design.
Finally, due to lack of data, our empirical estimations miss innovation outcomes as critical
intermediaries for diversification of customer composition. Although we attempted to cope
with this caveat by referring to direct effects of public subsidy via behavioral additionality,
explaining the conditions of focal subsidy programs, and demonstrating patent application
records of recipient firms, our empirical results would be more rigorous and persuasive if we
had been able to consider innovation outcomes explicitly in our models.
In spite of these limitations, our investigation contributes to the literature by providing an
in-depth examination and comparison of the effects of R&D support programs under local
government initiatives, considering inter- and intra-regional/industrial diversification of
transaction networks.
132
Chapter 5
Consequences of Local Policy for Postwar Reconstruction:
Historical SME Financial Policy25
25 This chapter is revised and English version of Takano, K., & Okamuro, H. (2020). Place-based SME finance policy and local industrial revival: Empirical analysis on directed credit program after WW2 (No. J-2020-06). Teikoku Databank Center for Advanced Empirical Research on Enterprise and Economy, Graduate School of Economics, Hitotsubashi University.
133
5. Consequences of Local Policy for Postwar Reconstruction: SME Financial Policy
5.1 Introduction
Access to finance plays a crucial role in the development of local economies. The
financial market development affects the outcomes of regional industry like the increase of
start-up rate and new firm entry (Guiso, Sapienza, and Zingales, 2004), and higher firm survival
rate (Arcuri and Levratto, 2020), as well as more general regional outcomes such as
urbanization (Bodenhorn and Cuberes, 2018), the increase of investment and saving (Carbó
Valverde, López del Paso, and Rodríguez Fernández, 2007), and the accumulation of human
capital (Kendall, 2012)26. In the process of the improvement of access to finance, the policy
intervention by the government is often justified. The access to finance of SMEs and venture
firms tends to be limited because of the problems like credit constraint, which prevents them
from launching and expanding their business (Carpenter and Petersen, 2002; Fritsch and Storey,
2014). Thus, the role of SME finance policies to mitigate the constraint has been emphasized
(Calomiris and Himmelberg, 1993; Karlan and Morduch, 2010). Despite the importance of the
policy interventions to improve the access to finance, little is empirically known about the
historical aspects of the SME finance policies by the local governments and their effects on the
development of the local industry. Although the form of access to finance depends on local
economic environments, these place-based factors have been ignored in the empirical literature
so far (Ughetto, Cowling, and Lee, 2019).
In this study, we fill this gap by empirically examining the effects of the modernization
fund program for small business enterprises implemented by Osaka Prefecture in the early
1950s. This study contributes to the literature mainly from two viewpoints. Firstly, we
26 Arestis, Chortareas, and Magkonis (2015), for instance, provides a more comprehensive review of the empirical investigation testing the association between access to finance and economic growth.
134
contribute to the literature about (SME) finance policy and local industrial development. This
study is one of few investigations empirically examining the SME finance policy by the local
government, whereas the previous literature has mostly focused on that implemented by the
central government. Besides, we investigate not only the effects of the policy on average but
also the heterogeneity of them depending on the historical environment of Osaka, the spatial
unevenness of the vestige of wartime economies in particular. We also contribute to the
literature about (SME) finance policy and local industrial reconstruction. The frequent
occurrence of natural disasters and economic crises in this decade has led regional economists
to investigate the role of policy-based finance in the process of the local industrial
reconstruction after the exogenous shock. Despite the rapid growth of this literature, few have
examined the role of policy-based finance in the recovery of local industry from more
catastrophic shocks like a war. Furthermore, due to the limitation of data, the accumulation of
empirical investigations about the activities of SMEs during the postwar period is even
insufficient in Japan. In this study, we construct the firm-level panel data tractable in
microeconometric evaluation by exploiting SME microdata collected shortly after the Second
World War based on the business credit survey combined with the detailed internal data about
the place-based finance policy. Our empirical analysis is valuable in the sense that we evaluate
the role of local finance policy in the postwar reconstruction by constructing and utilizing the
panel data.
135
5.2 Literature Review
5.2.1 SME Finance and Local Industrial Development
The development of econometric methods for the program evaluation and the
improvement of the access to firm-level microdata and spatially detailed regional data from the
2000s enabled the researchers to empirically examine the effects of the finance policies for
local SMEs on their outcomes. For example, Bernini and Pellegrini (2011) tested the effects of
discretionary grants in Southern Italy on the recipients’ production and employment level, and
productivity, Lee (2019) evaluated the effects of guarantee loan program by the U.S. Small
Business Administration on the employment and wage level of metropolitan areas, and
Rupasingha, Crown, and Pender (2019) investigated the effects of the USDA’s Business and
Industry (B&I) Guaranteed Loan Program to support businesses located in rural areas on the
recipients’ survival, respectively27.
It is quite important to discuss the design and effects of local industrial policies, including
SME finance policies, considering their interaction with socioeconomic factors peculiar to each
region. This is because the fact that one industrial policy does not fit all regions makes it
challenging to consider the policy design rooted in the characteristics each region has (Tödtling
and Trippl, 2005). However, the literature has hardly addressed this issue although the
investigations on the regional heterogeneity of the effects within a country and those on local
governments’ own programs can offer an angle on the issue.
As an example of the empirical investigation on within-country heterogeneity, Briozzo
and Cardone-Riportella (2016) examined the effects of Spanish SMEs’ subsidized and
guaranteed credit during the economic crisis. They showed that the effects on the firms in
Catalonia and Basque (with socioeconomic environment different from other Spanish regions)
27 About further remarkable empirical investigations on discretionary grants, see, for example, Neumark and Simpson (2015).
136
were quite different from those on firms in other Spanish states. There have been several
investigations testing the effects of local governments’ own SME finance policies recently.
Cannone and Ughetto (2014) evaluated the effects of the public financing program DOCUP
2000–2006 (Documento Unico di Programmazione) in the Piedmont region of Italy. They
showed the increase of indebtedness and of total fixed assets in subsidized firms thought they
could not find any evidence of any impact on firm profitability. Similarly, Martín-García and
Santor (2019) investigated the effects of public credit guarantees on SME business activity and
investment. They found a positive effect on the mitigation of credit constraints and the
enhancement of investment in supported firms.
This study is placed in later literature. We complement this literature by empirically
revealing the historical aspects and the effects of local governments’ own SME finance policy.
As stated in the decentralization theorem (Oates, 1972), the superiority of local government
relative to central government is that they can collect the information about the local
environment more, and thus they can design and manage the flexible policies more suitable to
regional economic conditions (Fernandez-Ribas, 2009). Therefore, the policies implemented
by the local government, as well as those by the central government, are remarkably important
and it is necessary to accumulate the knowledge about them furthermore. Despite this
significance, the most of targeted policies in previous literature were under the initiative of the
central government or for the large firms, taking the investigations about rationalization
policies as examples (e.g., Kiyota and Okazaki, 2005; Nakamura and Ohashi, 2012). We fill
this gap by focusing on the policy implemented by a local government, Osaka Prefecture during
the postwar period.
5.2.2 SME Finance for Local Industrial Reconstruction
The policy-based finance plays a more important role under the disruption like economic
137
crisis, war damage, and natural disaster. As described above, the primary role of policy-based
finance is the mitigation of credit constraints due to information asymmetry. This role becomes
more crucial under the economic disruption because the difficulty of the securing of working
capital of SMEs obviously increases. Also, the role of policy-based finance as a cowbell to
induce lending to SMEs by the private financial sector is particularly important under the
disruption (Vittas and Cho 1999; Shimada 2016).
The recent occurrence of the great economic depression and devastating disaster has
spurred the discussion about the resilience of the regional economic system in the field of
regional science (Martin and Sunley, 2015; Boschma, 2015). Along with this trend, the role of
SME finance in the process of industrial revival has been empirically examined. Here we
briefly review the investigations focusing on natural disasters. As a representative investigation,
Davlasheridze and Geylani (2017) analyzed the impacts of floods on businesses and the effects
of disaster loans provided by the U.S. Small Business Administration. Utilizing county-level
panel data, they showed that SMEs are extremely vulnerable to flood disasters because of the
lack of business adaptation to extreme events and further found that the loans are significantly
important for the post-disaster recovery of smaller firms. In contrast, more recent literature has
shown that the effects of financial support are not always straightforward. For example, Cole,
Elliott, Okubo, and Strobl (2019) examined the case of the Great Hanshin-Awaji Earthquake
and showed that the post-disaster financial aid contributed to the plant survival and sustainment
of employment level in the short term, but the effects were eventually offset by the increase of
debt combined with a sluggish Japanese economy. Similarly, Kashiwagi (2019) found that the
effects of post-disaster loans were quite limited in the manufacturing sectors, which was
because of the recovery through the inter-firm cooperation in advance of the loans.
Despite the recent progress in this literature, little has empirically examined the effects of
SME finance policy under the postwar reconstruction. The war damage is quite different from
138
the economic disruption like a natural disaster in that it generally incurs the catastrophic
changes in domestic institution and industrial structure, as well as the physical damages. Thus,
the investigation on the contribution of SME finance policy to overcome the postwar economic
disruption is remarkably valuable in expanding the insight of the policy studies about regional
economic resilience.
As one of a few empirical investigations, Bianchi and Giorcelli (2019) tested the effects
of the Marshall Plan implemented post-WW2 Europe utilizing Italian prefecture-level panel
data. They showed that the prefectures receiving large postwar aid achieved the growth of
employment level in the manufacturing and agriculture sector, and the production level growth
in the agriculture sector. This study complements this literature by carrying out the empirical
evaluation utilizing firm-level microdata, focusing on the prefecture-level policies rather than
nation-level policies, and examining within-region heterogeneity of the effects due to the
spatial unevenness of the dependence on the wartime economies. As descriptive studies related
to this study, Sawai (2017) revealed the actual state of factory diagnosis program and R&D
activities in local public technology center by Osaka Prefecture and Osaka City in the late
1940s, and Spadavecchia (2005) implemented the cross-regional comparison of the subsidy
programs for Italian industrial districts after the 1950s.
139
5.3 Institutional Background
5.3.1 Industrial Characteristics of Postwar Osaka
In advance of describing the historical background of Osaka during and postwar period,
we briefly review the characteristics of manufacturing sectors in Osaka. We show the name
and location of cities and wards that we frequently mention in this study in Figure 5.3.1. Figure
5.3.2 roughly shows the spatial distribution of manufacturing plants in Osaka City and Sakai
City, the central part of Osaka Prefecture based on the Osaka Chamber of Commerce and
Industry; OCCI (1950). Firstly, many plants in the metal and machinery industry agglomerated
around the waterfront area and riverside area of Yodogawa. Particularly, the metal and
machinery SMEs concentrated around the former state-operated arsenal, the eastern side of
Osaka City (e.g., Higashinari Ward and Joto Ward). While the agglomeration introduced above
still exists now, there are several industrial districts no longer extant. For example, the
agglomeration of the textile industry in the northern part of Osaka City and that of the wood
industry in the riverside area of Kizugawa (middle west area of Osaka City) almost vanished
due to the urbanization after the 1960s. Also, there were several large industrial districts of the
textile industry that still exist now in the Senshu Region the southern area of Osaka Prefecture,
though they are not shown on the map. In sum, Osaka Prefecture was (is) one of the largest
industrial districts in Japan that has a variety of manufacturing industries.
The industrial agglomeration beginning from the end of the 19th century played an
important role in the production of weapons and munitions during WW2. We describe Osaka’s
wartime economies and their demise following Osaka City (1953), Osaka Prefecture (1968),
Takebe (1982). During WW2, Osaka accounted for approximately 30% of the domestic
production of weapons in Japan. Particularly, the size of the metal and machinery industry
related to the production of weapons and munition became larger and larger thanks to
preferential assignment of the labor force, materials, capital equipment, funds, and food.
140
Figure 5.3.1: Name of primary cities and wards in Osaka Prefecture in 1950
In addition, Osaka had 6 large state-operated arsenals and about 70% of their production
were outsourced to private firms. In this sense, the manufacturing firms in Osaka formed a
company town of the state-operated arsenals and large private weapon plants. In Figure 5.3.3
we show the geographical distribution of subcontract munition plants called “cooperating
factories” (Kyoryoku Kojo) in 1943.
141
Figure 5.3.2: Industrial Districts in Okasa in 1948
Source: drawn by the authors based on OCCI, 1950
Figure 5.3.3: Geographical distribution of Cooperating Plants in Osaka and Sakai City
Source: drawn by the authors based on Kinki Region Cooperating Industry Council and Osaka Cooperating Industry Union. (1943)
Tokyo
Machinery & metal
Machinery& metal SME
Paper& textilenon-SME
Machinery, metal, chemical non-SME
WoodMetal
non-SMEMachinery& textile
142
Figure 5.3.4: Damaged area (drawn with red) of bombings in Osaka City
Available in https://www.oml.city.osaka.lg.jp/index.php?page_id=1147 [last accessed on September 18, 2020]
We can confirm that the distribution is similar to that of the industrial districts of metal
and machinery while the magnitude of concentration is stronger in specific wards like
Higashinari, Minato, and Nishi-Yodogawa. In contrast, the daily necessities sectors like textile
and food regarded as nonessential and non-urgent declined during this period. Also, the
chemical industry grew less than the metal and machinery industry due to the severe shortage
of materials.
The prosperity of Osaka as an agglomeration of the munition industry, however, began
to collapse as the war situation became deteriorated. The bombing on mainland Japan
beginning in 1944 and 1945 incurred massive damages on Osaka economies. We show the
damaged area by the bombings in Figure 5.3.4. Particularly, the damages of the bombings were
larger in Osaka City and Sakai City having many manufacturing plants. The burned area
amounted to 52 sq km, the number of destroyed residences was 310,955, and the number of
victims was 13,888 in Osaka City. The main target industrial districts of the bombings on Osaka
143
were Konohana having large private weapon plants like Sumitomo Metal Industries, and
Higashinari and Joto neighboring Osaka Arsenal. Similarly, the bombings on Sakai City
damaged the region around large private weapon plants. In Sakai City, the damaged area
amounted to 5 sq km and about 30,000 residences were destroyed.
5.3.2 Problems around SMEs during Postwar Period
We review the problems around Japanese SMEs, including those in Osaka, during the
postwar period following Nakamura, Akiya, Kiyonari, Yamazaki, and Bando (1981) and
Takebe (1982). The losses of production equipment and materials mainly due to the bombing
incurred a serious shortage of goods. Although there were several economic policies to tackle
the postwar problems and several big opportunities of revival around manufacturing sectors in
the late 1940s and the beginning of the 1950s, they did not necessarily mitigate and solve the
economic disruption and damage on Japanese SMEs. The first economic policy characterizing
the postwar period is the austere fiscal policy under the Dodge Line. While this austerity
contributed to the convergence of the postwar hyperinflation, the following stabilization crisis
caused an increase in unemployment and bankruptcies. In this regard, Japanese economies
declined because of the austerity on the contrary. The second policy was the priority production
system (and intensive production system following it). The priority production system
preferentially distributed funds and materials to key industries like the iron and steel industry
and coal industry, and the intensive production system distributed them to superior firms with
higher production efficiency compared to other ones in the same sector, which was quite
disadvantageous for most SMEs. Especially, because both SMEs and large firms coexisted in
targeted sectors of the intensive production system, the system was seemed harmful for SMEs.
The biggest event around Japanese manufacturing sectors was the special procurement
brought by the Korean War and the export boom that parallelly progressed with the
144
procurement. Under the procurement, massive demands for the production and repair of
weapons, and final goods for the military and their families emerged. The procurement
remarkably increased the production level in munition sectors like metal, machinery, and
textile industry. Although the procurement generally contributed to the post-war reconstruction,
it is said that the benefit of the procurement on SMEs was limited. According to the Basic
Survey of Small Business Finance conducted in 1950, only 32.1% and 23.9 % of Japanese
SMEs could achieve an increase in production and sales in this period. Also, only 9% of SMEs
could receive orders from the procurement. Furthermore, the procurement was temporary
because the demands for munitions dramatically decreased after the end of the Korean War.
The problem around SMEs that parallelly emerged with problems above was the shortage
of funds for their operations and capital investments. As described above, the materials and
funds distributed to SMEs were quite limited because of the preferential production systems,
which made SMEs disadvantageous in terms of access to finance. More than 70% of SMEs
answered that their financing was poor in the Basic Survey of Small Business Finance
conducted in 1948. Consequently, this lack of access to finance made the introduction and
replacement of production equipment by SMEs quite difficult, and they had no choice but to
rely on the production using overused, obsolete, and less efficient machine tools. Despite the
Japanese market with an excess of demand, SMEs could not massively supply their goods
because their production level remained quite low.
5.3.3 Postwar SME Finance Policies
We briefly explain the progress of Japanese SME finance policies by the beginning of the
1950s relying on Nakamura et al. (1982), Ministry of International Trade and Industry; MITI
(1963), and Ueno, Murakoso, and Hirai (2006) in advance of the description about the scheme
of the modernization fund program in Osaka that we analyze in this study. Due to the
145
debilitation of SME and local financial institutions incurred by the forced bank consolidations
during WW2 and the vanishment of the wholesale finance system, the postwar financial
difficulty of SMEs was more serious than that during the prewar period. Although the
Reconstruction Finance Bank (Fukko Kinyu Kinko) was launched in 1947 as an institution
providing loans for private firms, its contribution to the revival of SMEs was very small
because its main targets were large firms and it was eventually abolished due to the austere
fiscal policy under the Dodge Line. Similarly, the SME finance program based on the collateral
funds from US aid to Japan launched in 1950 as a subsequent program of the Reconstruction
Finance Bank did not function well because of the complexity in its operation for the financial
institutes and the strong restriction on the ways to use for SMEs.
To overcome this shortage, several policy-based financing institutions were established
around 1950. After the establishment of the Small and Medium Enterprise Agency in 1948, the
People’s Finance Corporation and Japan Finance Corporation for Small Business were
launched in 1949 and 1953, respectively. The aim of these institutions was, however, the relief
and preservation of SMEs, and it was the middle 1960s that the policy-based finance aiming
the enhancement of SMEs’ competitiveness began to be established. Nevertheless, we cannot
necessarily say that there was no policy-based finance for the modernization of SMEs. For
example, the subsidy program for capital investment by business cooperatives was
implemented from 1947. However, it was in 1954 that the direct loan program for the
modernization of SMEs started throughout Japan under the initiative of the Japanese
government.
5.3.4 Modernization Fund by Osaka Prefecture
Despite the immature SME support policies under the initiative of the Japanese
government at the beginning of the 1950s, Osaka Prefecture provided various support programs.
146
According to Osaka Prefecture (1952), it provided programs including the business and
technical consulting, factory diagnosis, lending of high-performance machine tools, as well as
directed credit for the modernization. This directed credit program was provided by Osaka
Prefecture individually from 1951 to 1953. Thus, the remarkable characteristic of this program
is that it was implemented antecedent to the program under the initiative of the central
government.
We describe the scheme of this program following the internal document of Osaka
Prefecture used in the actual operation then (Osaka Prefecture, 1951). The basic framework of
the program was as follows: firstly, Osaka Prefecture deposited 1.6 billion yen to the designated
7 private banks, and secondly, the fund for the modernization was lent to SMEs from the
deposit at low interest (0.35 yen per day). The business cooperatives which had the main office
and SMEs (with the capital of a million yen or less, or 300 or fewer employees) in Osaka
Prefecture were applicable to this modernization fund program. In addition, the designated
banks could claim compensation if they suffered from a loss due to the lending under the
program. Thus, this modernization fund program could mitigate the difficulty in managing
lending for the private banks and that in acquiring access to finance for SMEs.
Osaka Prefecture gave priority to the objectives of the program as listed below. According
to it, we can presume that the program mainly aimed to modernize local SMEs through the
replacement of old equipment or production system.
1. The improvement of production capability through the replacement of old and inefficient
equipment.
2. The introduction of a more efficient machine tool as the substitution of poorly made
equipment or manual operation.
3. The rise of production level by the extension of equipment.
4. The maintenance of facilities combined with the introduction of extension of equipment.
147
Based on the final report inside Osaka Prefecture (Osaka, 1954), we show the specific
achievements of the program. Firstly, the total number of supported firms was 482 (326 firms
when we limited the firms with the capital of a million yen or more and in manufacturing
sectors or wholesale sectors related to manufacturing). The acceptance rate was 61.2%, so the
receipt of the fund was moderately competitive. Since approximately 99.6% of 1.6 billion yen
was lent to SMEs, this program seemed to be operated well. The average amount of lending
was about 2.7 million yen (approximately 19 million yen in present value in 2017). the total
number of replaced or introduced machine tools was 4,805 and that of improved facilities was
327.
Table 5.3.1 summarizes the average value of capital, number of employees, and loan
amount for the applications by individual firms. The number of applications was large in metal,
machinery, and textile, which strongly reflected the industrial structure of Osaka. From the
average value of capital and the number of employees, the main targets of this program seemed
to be medium firms rather than micro-enterprises. Excluding a few exceptions, the variation of
loan amount between sectors was not so large.
Table 5.3.1: Average lending status by industrial sector Sector Capital [M ¥] No. of employees [person] Loan amount [M ¥] Adoptions Textile 3.524 60.2 3.034 85 Metal 2.419 59.267 2.895 116
Metal product 2.402 92.148 3.507 27 Machinery 2.049 75.059 2.572 101
Miscellaneous 4.813 93.484 3.484 31 Wood 2.492 65.118 6.488 17
Chemical 6.468 66 3.236 22 Chemical product 2.785 52 2.058 20
Medical 9.213 78.25 1.75 4 Other 2.653 37.968 2.755 31
Notes: source is Osaka Prefecture (1954). The aggregation for the adoptions by individual firms. We exclude the firms with unknown capital or no. of employees, and cooperatives.
Osaka Prefecture concluded the achievements of the program remarking “each supported
plant achieved the rationalization and modernization through the increase of production amount,
148
improvement of technologies, reduction of cost, improvement of product quality.” Our
objective is to objectively examine whether these effects subjectively observed on site can be
quantitatively supported and what kind of heterogeneity exists in the effects, exploiting
microeconometric methods for the program evaluation.
149
5.4 Methodology
5.4.1 Hypotheses
In this study, we mainly test the following two hypotheses.
• H1-a: Modernization fund improved supported SMEs’ production level.
• H1-b: Modernization fund improved supported SMEs’ production efficiency.
We assume the following mechanism in H1-a. Provided that each SME has production
function represented with " = $(&, (, )) , where Y is production level, A is a parameter
capturing technological progress, L is labor input, and K is capital input. As described in the
previous section, for SMEs, the financial restriction for capital investment was quite large
during the postwar period. Because the modernization fund program aimed to increase capital
investment (increase K in other words) through the mitigation of this restriction, the policy
intervention might have improved the production level Y. In H1-b, we presume the mechanism
such that higher production level was achieved keeping the amount of labor force constant
through the replacement of old and inefficient equipment or the introduction of automation.
5.4.2 Econometric Methods
We evaluate the effects of the modernization fund on the supported SMEs’ performances
utilizing the difference-in-difference method. The use of DD mitigates the problem in
identifying causal effects of the program due to time-varying unobservable variables whose
trend is similar between the treatment group (firms that receive fund) and control group (those
did not receive fund). Furthermore, by combining DD with fixed effect estimation, we control
various confounding factors that prevent us from a precise estimation of the effects. The
regression specification of DD in our analysis is as follows:
"!" = +" + -! + ./01.! × 1$.0/"3# + 456./57!" + 8!" . (5.1) "!"is firm:’s outcome in period.. +"is time fixed effect controlling macroeconomic trends
150
common to all firms included in our dataset. -! is the individual (firm) fixed effect controlling
time-invariant firm-specific unobservable factors such as its corporate culture and history.
./01.! × 1$.0/" is the variable of interest. ./01.! is a dummy variable taking 1 if firm:was
a borrower, and1$.0/"takes 1 if period.is after the policy intervention. 456./57!"is a set of
other control variables including sector-time fixed effect and city-time fixed effect. Particularly,
the inclusion of sector-time fixed effect is important to control the sector-specific macro shocks.
The representative shock specific in our analysis is the special procurement because it
positively affected the specific sectors like the metal and machinery industry. Though the effect
of the special procurement on SMEs was limited as described above, we cannot completely
reject the likelihood of indirect effects through, for example, keiretsu. 8!" is a stochastic
disturbance. Our hypotheses will be supported if3# > 0.
Due to the specification of our panel data of Osaka SMEs, we implement DD analysis
using the observations in 1951 and 1957. We use the logarithm of annual sales [million yen]
as an outcome to measure the production level and the logarithm of annual sales per capita
[million yen/employee] as an outcome to measure the production efficiency.
5.4.3 Data
To identify the borrowers, we utilize the list of supported plants (Osaka, 1954) an internal
document summarizing the achievement of the program. In this list, detailed information about
loans is available, as well as basic information about each borrower like company name, name
of the president, capital, number of employees, and main products.
• Lending bank
• Loan amount
• Start and end date of lending
• List of equipment introduced or improved with a fund
151
We match this list with the corporate information database named “Imperial Directory of
Banks and Companies”; IDBC (Teikoku Ginko Kaisha Yoroku) published by Teikoku
Koushinjo Co., Ltd. (the present Teikoku Databank, Ltd.) in 1951 and 195728. In the empirical
analysis using DD, 4 categories of data capturing the observation in both before and after the
intervention and in both treatment and control groups are necessary. To the best of our
knowledge, in Japan, there is no SME micro-dataset except IDBC. 3,112 firms and 10,400
firms are observed in 1951 IDBC and 1957 IDBC in Osaka Prefecture, respectively. In both
1951 and 1957 IDBC, the following attributes can be identified for firms with the capital of a
million yen or more.
• Company name
• Full address
• Foundation year and month
• Business objectives
• Capital
• List of executives
• Number of employees
• Annual or monthly sales
• Bankers
• List of plants, offices, and facilities (only a part of firms)
Although we can identify each firm’s business object on IDBC, this information is
difficult to directly convert into variables tractable in empirical analysis. Thus, we match the
industrial classification by using the following two supplemental databases.
28 The records corresponding to firms in Osaka in 1957 IDBC will be available via the TDB Center for Advanced Empirical Research on Enterprise and Economy, Hitotsubashi University in Excel format for free, limited to academic use by March 2021.
152
• 1951 Osaka Commerce and Industry Directory (edited by OCCI)
• 1949 and 1952 List of Factories in Japan (edited by MITI)
The firms observed in our two-wave panel data satisfy the following conditions:
• Number of employees and sales were observed in both 1951 and 1957.
• Industrial sector is identified by supplemental datasets.
• Located in same sector and city with borrowers if a firm was non-borrower.
• Number of employees is less than or equal to 300.
• Capital is less than or equal to 1 million yen.
• Never experienced company split-up from 1951 to 1957.
153
5.5 Results
5.5.1 Baseline
In advance of regression analysis, we show the average trend of outcomes observed in
both before (1951) and after (1957) the intervention and in both treatment (borrower) and
control (non-borrower) group in Table 5.5.1. By taking the difference between a before-after
change of an outcome in the treatment group and that in the control group, we can calculate
naïve estimates of the effects of the program. While the average level of annual sales in the
treatment group was lower than the control group, this gap was reduced after the policy
intervention. In contrast, the gap did not decrease so much about sales per capita.
Table 5.5.1: DD table
ln(sales) ln(sales per capita) Before After Before After
Control 4.955 5.877 1.127 1.721
Treatment 4.341 5.576 0.123 0.778
DD 0.313 0.061
To confirm whether the results above are statistically validated and robust, we estimate
the regression DD specified in Eq. (1). Table 5.5.2 shows the estimation results. About annual
sales, the estimated DD represented with the regression coefficient of treat×after is positive
and statistically significant at 1% level. About sales per capita, however, the estimated DD is
not statistically significant though the sign is as expected. One of the reasons for the result of
sales per capita is the increase in employment level parallel to that of the production level. In
the same table, we show the estimation result using the logarithm of the number of employees
as an outcome, and we can observe the positive DD statistically significant at 10% level.
Eventually, H1-a is supported while H1-b is not necessarily supported according to our
estimation. In the following sections, we examine the baseline results more in detail by
checking whether and how the spatial and industrial heterogeneity of the effects are observed.
154
Table 5.5.2: Estimation results of DD (baseline) ln(sales) ln(sales per capita) ln(emp)
beta tval beta tval beta tval treat×after 0.42 2.794 *** 0.204 1.29 0.216 1.876 * Treated firms 45 45 45 n 1196 1196 1196
Notes: statistically significant in ***1%, **5%, *10%. The estimation results of two-way fixed effects model. Standard error is clustered by firm level. All models include time-sector and time-city fixed effects.
5.5.2 Vestige of Wartime Economies
5.5.2.1 Prosperity and Demise of Osaka Arsenal
In this section, we evaluate within-regional spatial heterogeneity of the effects focusing
on the historical milieu of Osaka local industry, the spatial heterogeneity of the vestige of
wartime economies specifically. The subject of our analysis is the former Osaka Arsenal, the
largest arsenal in 6 state-operated weapon plants in Osaka, and its company town. We briefly
review the historical background about the Osaka Arsenal following Miyake (1993), a
representative case study about the arsenals in Osaka.
The Osaka Arsenal was established in 1870 around Osaka Castle. The arsenal fulfilled a
central role in the production of weapons used in the Japanese Army during the period from
the Russo-Japanese War to WW2. While it engaged in weapon production relying on imported
technologies at first, it established the unique system and technologies for the mass production
of guns, tanks, about munitions as the time passed. Particularly, its technology of casting and
metal processing was leading in Japan, and they were even diverted into the production of
civilian goods like water pipes. At the end of WW2, the arsenal was the largest in the East
having about 64,000 engineers, 20,000 machine tools, and a site area of 6 million sq m.
The arsenal influenced the Osaka manufacturing sectors. Due to the nature of the arsenal
that it was specialized in weapon production, it executed the massive lay-off every time the
war ended. This massive release of engineers with high-quality technology acquired in the
arsenal to the private sectors had a big impact on the formation of industrial agglomeration in
155
Osaka, the Joto region in particular (Abe, 2006). Many spin-off firms having the advantage in
steel and aluminum processing and machinery were established by ex-employees of the arsenal
(Matsushita, 2012). Also, as explained in Section 4.3.1, the outsourcing rate to the private
sectors of the arsenal was high. This formed the keiretsu and company town engaging in
weapon production around Joto regions during the period from the 1930s to WW2 (Ueda, 2004).
At the end of WW2, nearly 600 plants were under the control of the arsenal.
However, the defeat in WW2 marked the end of the Osaka Arsenal. The arsenal and
surrounding industrial agglomeration became the targets of the bombings by the US, and they
damaged the region around the arsenal again and again from 1944 to 1945. The largest bombing
was executed on August 14th, 1945, one day before the unconditional surrender. About 650
one-ton bombs exactly hit their targets and they destroyed most of the facilities. Due to the risk
of unexploded bombs, it was difficult to redevelop the demolished area, and it was kept idle
until the 1960s. Although the area was eventually transformed into the business district after
the 1970s, one of the industrial cores in the Joto region was lost forever.
After WW2, SMEs around the former arsenal faced a trial like other Japanese SMEs. The
special procurement by the Korean War, however, might have changed this situation ironically.
As described in Section 5.3.2, the benefit of the special procurement on SMEs was generally
limited. But we should be cautious in the discussion on whether this is also the case in Osaka
and is true even if we consider the indirect impact through the keiretsu relationship. The recent
case studies have investigated the anecdotal evidence of the impact of this special procurement.
For example, Sawai (2018) revealed the historical fact that the orders of munitions like
cannonballs from the special procurement were concentrated to the large manufacturers having
main plants in Osaka like Komatsu, OKK, and Daikin. According to the Osaka Research Center
for Industry and the Economy; ORCIE (1953), the main reason for this concentration was
thanks to the subcontract plants existing even after WW2 in former keiretsu for cannonball
156
production. In this sense, we cannot necessarily ignore some effect of former Osaka Arsenal as
a vestige of wartime economies on the industrial agglomeration in postwar Osaka.
5.5.2.2 Empirical Framework
Following the discussion in the previous section, we empirically investigate the effects of
the modernization fund on the borrowers around former Osaka Arsenal. Specifically, we
additionally examine below following the hypotheses set in Section 5.4.1.
• H2: The effects of the fund on SMEs’ performances were larger for the borrowers around
former Osaka Arsenal.
We assume possible 3 mechanisms and their interaction in H2. The first mechanism is
the special procurement. As described above, the industrial agglomeration around the arsenal
was specialized in metal and machinery industry strongly related to the production of munitions.
Considering the anecdotal evidence, the modernization fund might have contributed to the
enhancement of the production of munitions. The second mechanism is the agglomeration
externality. It has been theoretically and empirically pointed out that various types of
externality including matching, learning, and sharing work inside the agglomeration (Duranton
and Puga, 2004). This mechanism might have worked together with the first mechanism.
Although not all of metal and machinery SMEs are necessarily associated with the production
of munitions, there might have been some technological and knowledge spillover from
geographically close subcontract firms engaging in the production. The third mechanism is
larger room to recover conditioned by the policy intervention. The decline of the agglomeration
around the former arsenal might have been severer though we cannot directly confirm due to
the lack of data in the prewar period. Unlike other private large munition plants, the arsenal
157
was lost forever because it was operated by the Japanese Army29. Due to this abolishment, a
part of the technological and knowledge spillover brought by the large plants (Greenstone,
Hornbeck, and Moretti, 2010) might have been lost. But the effects of the modernization fund
might be larger if the borrowers originally had high potential based on the knowledge and
technology they acquired as subcontract plants of the arsenal. In contrast, if the shock of the
abolishment was beyond a certain tolerance, the decline of the agglomeration was persistent
despite the policy intervention.
To examine H2, we estimate the regression DDD model specified below.
"!" = +"$ + -!$ + ./01.! × 1$.0/"=# + >5?ℎ5! × 1$.0/"=%+ ./01.! × >5?ℎ5! × 1$.0/"A# + 456./57!" + B!" . (5.2)
>5?ℎ5! is a dummy variable taking 1 if firm:was in wards neighboring former arsenal (Joto
and Higashinari, or simply Joto region). Compared Eq. (2) with Eq. (1), this DDD tests if
effects were heterogeneous between the borrowers inside the agglomeration around the arsenal
and those outside by introducing triple interaction term ./01.! × >5?ℎ5! × 1$.0/" . H2 is
supported if A# > 0.
5.5.2.3 Results
Like Section 5.5.1, in advance of regression analysis, we show the average trend of
outcomes observed in each group in Table 5.5.3. The rows with Treat=1 show the results in
treated firms, and those with Kosho=1 show the results in the firms in the region neighboring
the arsenal in 1951. Each outcome of firms inside the agglomeration was smaller than that of
firms outside.
Table 5.5.3: DDD table ln(sales) ln(sales per capita)
29 Many of the former private plants engaging in weapon production had to temporarily cease their operation after WW2 and their continuity was uncertain due to the postwar compensation plan (Compensation Agency, 1948). However, the plan was hardly implemented, and most of these plants could survive by transforming into the peace industry.
158
Treatment Kosho Before After DD Treatment Kosho Before After DD
0 0 5.005 5.952 0 0 1.198 1.827
1 0 4.462 5.584 0.176 1 0 0.135 0.718 −0.046 0 1 4.526 5.233 0 1 0.507 0.802
1 1 3.967 5.551 0.878 1 1 0.085 0.965 0.585
DDD 0.702 0.631 Notes: the rows with Treat=1 show the results in treated firms, and those with Kosho=1 show the results in the firms in wards neighboring the arsenal in 1951
Table 5.5.4: Estimation results of DDD (former arsenal)
ln(sales) ln(sales per capita)
beta tval beta tval after×kosho −0.203 −1.05 −0.152 −0.906 treat×after 0.232 1.748 * 0.066 0.622 treat×after×kosho 0.68 1.743 * 0.498 1.035 Treated firms 45 45
n 1196 1196 Notes: statistically significant in ***1%, **5%, *10%. The estimation results of two-way fixed effects model. Standard error is clustered by firm level. All models include time-sector and time-city fixed effects.
Looking at the trend of each outcome after the intervention focusing on inside the
agglomeration, annual sales became almost the same with that of firms outside the
agglomeration, and sales per capita became larger. By taking the difference between DD inside
the neighboring wards and that outside, we can calculate naïve estimates of the additional
effects of the program brought to the borrowers inside the agglomeration.
To confirm whether the results above are statistically validated and robust, we estimate
the regression DDD specified in Eq. (2). Table 5.5.4 shows the estimation results. About annual
sales, the estimated DDD represented with the regression coefficient of treat×after×kosho is
positive and statistically significant at the 10% level. This result implies the additional effect
of the fund on annual sales in borrowers inside the agglomeration around the arsenal. About
sales per capita, however, the estimated DDD is not statistically significant though the sign is
as expected. Eventually, H2 is partially supported in only the production level.
5.5.3 Industrial Heterogeneity
The effects estimated in Section 5.5.1 are the overall average effects for the supported
159
firms. However, the additional analysis might be necessary to check whether the effects were
different between industrial sectors. One of the objectives of this analysis is the impact of the
special procurement. As described above, the keiretsu around the former arsenal was exploited
in the production of munitions during the Korean War, and the impact through the keiretsu
might have had the industrial extent, as well as the spatial one examined in Section 5.5.2. The
subcontract firms under the Osaka Arsenal and large private munition plants were located all
over Osaka Prefecture centered around the Joto region and the waterfront area (Kinki Region
Cooperating Industry Council and Osaka Cooperating Industry Union, 1943; United States
Strategic Bombing Survey, 1947). If this hypothesis is supported, the stronger effects might be
observed in borrowers in the metal, machinery, and textile industry following the similar
mechanism assumed in the previous section. To examine this, we decompose the effects of the
program by estimating the regression model specified below.
"!"& = +"$$ + -!$$ +C./01.! × 1$.0/" × ?04.5/!&D&&
+ 456./57!"& + E!"&, (5.3)
?04.5/!&takes 1 if firm:was included in sector?.
Table 5.5.5: Estimation results of DD (industrial heterogeneity)
ln(sales) ln(sales per capita)
beta tval beta tval treat×after×chemical −0.048 −0.163 −0.072 −0.414 treat×after×wood 0.235 0.578 −0.073 −0.183 treat×after×machinery 0.474 1.791 * 0.049 0.152 treat×after×textile 0.465 1.707 * 0.09 0.287 treat×after×metal 0.952 2.196 ** 1.011 2.499 **
treat×after×other 0.326 1.724 * 0.214 0.936
Treated firms 45 45
n 1196 1196 Notes: statistically significant in ***1%, **5%, *10%. The estimation results of two-way fixed effects model. Standard error is clustered by firm level. All models include time-sector and time-city fixed effects.
The estimation results of the regression model are shown in Table 5.5.5. About annual
sales, we can observe positive and significant DDs in the metal, machinery, and textile industry.
160
On the other hand, we can observe positive and significant DD only in the metal industry. In
sum, the effects of the program are heterogeneous between industrial sectors, and the results
might reflect the impact of the special procurement.
161
5.6 Discussion and Conclusion
The development of the financial market fulfills an important role in regional economic
development, and the policy intervention for improving SMEs’ access to finance is often
justified. Considering the nature of SMEs such that they are extremely vulnerable to the
socioeconomic disruption like the natural disaster and economic crisis, the significance of
policy-based finance increases more. In this study, we examined the effects of the
modernization fund for SMEs carried out by Osaka Prefecture during the postwar disruption,
the early 1950s. Despite the importance of the local industrial policies for mitigating SMEs’
financial constraints, the previous literature hardly addressed the historical aspects and effects
of local governments’ own SME finance policies carefully considering the region-specific
factors influencing the form of access to finance. In this sense, this study might contribute to
the literature because we provide new insights into this open issue.
We summarize the research findings in this study. Overall, the empirical results imply
that the modernization fund program contributed to the improvement of borrowers’ production
levels relative to non-borrowers. These results might support the policymakers’ subjective
evaluation of the achievements of the program though we evaluated the effects based on the
modern method of program evaluation, counterfactual analysis. On the other hand, we could
not necessarily observe the effect on supported SMEs’ production efficiency though we should
be cautious about the evaluation based on an outcome like sales per capita. This result, however,
reflected the rise of the employment level in the borrowers. In this regard, the program worked
well in terms of the regional industrial activation rather than industrial rationalization.
In an additional analysis, we found that the effects of the modernization fund were
spatially and industrially heterogeneous. Particularly, the local historical factor unique to Osaka
such as geographical or organizational proximity to former Osaka Arsenal as a vestige of
postwar economies, the exogenous shock like the special procurement by the Korean War, and
162
the existence of industrial agglomeration might have been the causes of the heterogeneity. Our
empirical results suggest that the effects of the program were larger in the specific industrial
sectors like metal and machinery industry and the regions specialized in these sectors due to
the historical background.
Before concluding this study, we mention the future issues and limitations of this study.
Firstly, we cannot observe the effects of the program on smaller firms with the capital of 1
million yen or less, because the firm-level database utilized in our analysis did not cover them.
Also, the coverage of the database might have been limited to the firms with established
reputation considering the background that the corporate information used in our analysis was
collected shortly after WW2. Thus, we cannot reveal the truly complete picture of the program
unless we can find SME micro-data more like an exhaustive survey. Secondly, further analysis
to reveal the detailed mechanism in the effects of the program is required. Although we
empirically showed the heterogeneity of the effects due to various regional-specific factors in
Osaka like wartime economies and the special procurement, each estimated effect is still just a
compound. Thus, we should unravel this complex result, for example, by additionally
analyzing detailed data about wartime economies like the keiretsu information of munition
production during WW2.
Despite these future issues and limitations, this study can contribute to the literature in
that we showed the role of the local government’s effort in the postwar reconstruction. The
discussion in this study on the role and importance of the place-based industrial policy by the
local government under the disruption might provide some insight into the industrial revival
after the local shock like the natural disaster and economic crisis, not limited to the war damage.
163
Chapter 6
Conclusion
164
6. Conclusion
6.1 Summary of Findings
This thesis aimed to address three main issues in the literature of regional economic
resilience. Throughout this thesis, I regarded resilience as a series of capabilities to reduce
losses from shocks and if necessary, to adapt to the post-shock environment. The first issue is
the role of networks and the interdependence between regions and sectors in developing
regional economic resilience. Since resilience cannot be enhanced in isolation by a single
economic agent or region, it is imperative to understand the complex interrelationship between
different economic agents or regions as a determinant of the development of regional economic
resilience and to understand its sources. In addition, cross-regional and sectoral networking
plays an important role in not only mitigating the economic agents’ and regions’ structural
vulnerability, such as the dependence on a single sector, technology, and market, but also
creating external knowledge linkages.
The second issue is an empirical examination of the determinants of (sources of) regional
economic resilience. Despite the maturity of the conceptualizations of regional economic
resilience and the empirical exploration of the association between (sources of) resilience and
local economic development in this decade, little is still known about the determinants of
(sources of) resilience per se. In addition, in the process of examining the determinants, an in-
depth understanding of the individual role of each economic agent’s (e.g., firms, households,
and local government) decision-making and their interaction is needed.
The final issue is the role of local government and its policies in developing (sources of)
regional economic resilience. While the literature on regional economic resilience neglects the
role of governments, the discussion on resilience has just spiraled off into the policy realm due
to the recent frequent occurrence of socioeconomic disruptions, such as economic crises,
natural disasters, and pandemic outbreaks. In the acquisition of regional economic resilience,
165
the local government fulfills a crucial role in terms of implementing policies in response to the
spatial unevenness in the effect of a shock and to the need to establish complementarity with
nationwide actions. In this sense, in applying notions of regional economic resilience to place-
based policies, through the empirical evaluation of previous programs, we should understand
the actual situation, effects, and limitations of the economic policies by local government.
Based on these issues, Chapters 2, 3, and 4 focused on prospective adaptive resilience,
preventive policies and strategies, while Chapter 5 focused on a policy, conducted after a shock,
that aims to recover from an existing disaster.
First, in advance of providing policy implications, I summarize the findings obtained from
four essays. Chapters 2 and 3 examine the local economic agents’ own actions to acquire (the
source of) resilience. In Chapter 2, I focused on the firms’ actions. Specifically, I investigated
the question of whether the learning gained from experiencing a local catastrophic shock
enables a firm to regionally diversify its supply chain against expected future shocks and
thereby may mitigate potential damage. To address this question, I exploited the Great East
Japan Earthquake as an actual shock and the expected Nankai Trough Earthquake as an
expected future shock. The results showed that the experience of a local catastrophic shock
would not necessarily drive firms to change their sourcing strategies in preparation for future
shocks. However, I found that the medium-sized firms diversified their supply chain if they
had suppliers that were both in regions at risk of a projected shock and had been damaged by
an actual shock. This suggests that the heterogeneity in the firms’ characteristics and in the
surrounding business environment may matter in the actions taken towards the reduction of the
firms’ vulnerability due to an excessive dependence on a small number of local input markets.
In Chapter 3, I focused on the local governments’ actions. I analyzed the determinants of
the early development of municipal pre-disaster preparation plans (business continuity
planning; BCP) and the spatial pattern of the development status. In this status, I particularly
166
examined the existence of a spatial dependence motivated by the concept of (spatial) policy
diffusion: I tested whether the decision in BCP development by a municipality was associated
with the decision in BCP development made by the neighboring municipalities. The
interdependence of the BCP development status between cities matters in enhancing efficiency
and the seamless actions taken against emergencies. This is because no single jurisdiction or
country can manage the damages caused by an extreme event if negative externalities linked
to it are so high. The results showed a positive spatial dependence in the development status
between municipalities even after controlling for differences in regional characteristics (e.g.,
local administrative capacity, the risk environment, and prefecture-level spatial heterogeneity).
Chapters 4 and 5 examine the interaction between local industry and government. I
empirically evaluated the association between the local government’s efforts and the local
industry’s acquisition of (the source of) resilience. In Chapter 4, through a comparative
assessment of the effects of local R&D support programs in three neighboring prefectures in
the same district in Japan, I empirically evaluated whether the local R&D support programs’
enhancement of the SMEs’ business diversification was a source of regional economic
resilience. Specifically, I focused on the regional and industrial diversification of transaction
networks as an outcome. Network diversification enhances the firms’ competitiveness through
the improvement of their opportunity to select better markets and technologies and the increase
in their access to different location/sector-specific resources, thus alleviating negative lock-in.
An empirical evaluation based on a firm-level dataset confirmed only weak and partial effects
of local R&D support on business diversification. However, it provided evidence of the
different consequences of local innovation policies. While under the program, a certain pattern
of diversification progressed and included new market development as an objective, this was
not the case in the prefecture whose program was specialized in the development of the R&D
environment and in which the local input-output market was self-contained.
167
In Chapter 5, I investigated the effects of the SME modernization fund program
implemented by Osaka Prefecture in the early 1950s on the performance of local SMEs.
Utilizing the firm-level panel data, I empirically showed how the modernization program
contributed to the local SMEs’ post-war reconstruction and how the spatial and organizational
proximity to wartime economies differentiated the effects. In this regard, an empirical analysis
in Chapter 5 examined whether the effort of the local government helped the local industry
bounce back from war damages and post-war disruption. The results showed an improvement
in the production levels in the recipients. In addition, I showed that the effects were
heterogeneous between sectors and regions depending on their geographical and organizational
proximity to local wartime economies. Specifically, the recipients in sectors relevant to
munitions production or characterized by specialized agglomeration achieved additional or
larger improvement in the production level; this outcome might be explained by special
procurement by the Korean War and positive externalities between firms.
6.2 Policy Implications
I organized this thesis based on three issues: (1) the role of network and interdependence
between regions and sectors; (2) the determinants of (sources of) regional economic resilience;
and (3) the role of local government and its policies. Based on these issues, I accordingly state
several policy implications derived from a series of empirical investigations on the
development of resilience policies and strategies in the local economy.
6.2.1 Network and Interdependence
Chapters 2 and 4 examined the determinants of the local firms’ network diversification,
which was analyzed as a potential driver of economic resilience. While Chapter 2 tested the
association between the experience of past disasters and the use of diversification as a sourcing
168
strategy to mitigate structural vulnerability in relation to trade uncertainty, Chapter 4
investigated the association between policy interventions by local governments and the use of
diversification as a way to develop new markets and increase the presence of local SMEs.
Despite this difference in the focus of these chapters, they both showed that the policies and
strategies to enhance the firms’ network diversification did not necessarily work well.
Thus, the investigations conducted in this thesis revealed that it was almost impossible to
find unified and silver-bullet policies and strategies aimed to acquire and enhance regional
economic resilience through network diversification. The potential implication from these
results is that we should be cautious when considering policies and strategies to strengthen
diversification. We may have to keep in mind that whether the acquisition of prospective
adaptive resilience through networking can be achieved depends on various conditions, such
as the firms’ or the regions’ characteristics, capabilities, and surrounding constraints.
In Chapter 3, I found a significant similarity of the BCP development status among
neighboring municipalities. Though based on my framework, it is difficult to distinguish cross-
jurisdictional collective actions explicitly and sufficiently from the effect of unobservables (e.g.,
localized risk environments only common to neighboring cities), this result at least suggests
the importance of the interdependence between regions in the development of regional
economic resilience. This result implies that we should carefully consider whether a
government’s decisions are influenced by those of other governments, the extent to which this
mutual process spatially diffuses, and what kind of mechanism (e.g., a reliance on the precedent
and magnitude of externalities caused by a shock) seems to work in this process.
6.2.2 Determinants of (sources of) regional economic resilience
One of the motivations to examine the determinants of (sources of) regional economic
resilience by focusing through the use of microdata, on the individual role of each economic
169
agent is to reveal the managerial issues in addition to the issues that arise at the local industry
level or policy issues. In particular, I focused on the roles of local (small and medium-sized)
firms through Chapters 2, 4, and 5. From the results obtained in these chapters, I state the
managerial implications associated with my policy implications. The implication common to
the results in these chapters is that consistent with the recent discussion on the determinants of
regional economic resilience described in Section 1.3.1, the firms’ fundamental capabilities
and the constraints on them also matter in developing resilience.
In Chapter 2, it was only in the medium-sized firm group that we could observe network
diversification, and diversification did not progress in the more vulnerable group organized by
smaller firms perhaps because of capacity constraints (e.g., excessive search costs for finding
alternative suppliers). In addition, representing one of the consequences of an excessively high
cost of connectedness that prevents SMEs from being more resilient, the results in Chapter 4
imply that the surrounding environments around local SMEs might restrict the opportunity for
new market development despite the availability of subsidies. Although the results in Chapter
5 imply the success of the evaluated program contrary to the findings in Chapters 2 and 4, we
can observe geographical and industrial heterogeneity in the magnitude of the effects, possibly
reflecting the difference in the elasticity of productivity or the production level with respect to
policy intervention. In light of these obtained results, the local firms’ corporate attributes that
enhance or weaken their market competition in times of peace might indeed be associated with
the development of resilience and thus with the preparation against crises.
6.2.3 Role of local government and its policies
The results in Chapters 4 and 5 imply a trade-off between the advantages and
disadvantages of local governments in implementing industrial policies and provide insight
into how policymakers can address the trade-off. The results in Chapter 4 only partially and
170
weakly support our hypotheses that R&D support can enhance new market development,
measured by transaction network diversification, in various regions and sectors. Despite the
policy design based on the local economic circumstances in each prefecture, the consequences
of policy intervention were not necessarily successful. These results might reflect a shortage of
know-how and financial constraints that the local government tends to face in conducting
industrial policies. The results in Chapter 5 might be suggestive when we consider how to
tackle this trade-off, although we should be careful about the difference in time and place. As
described above, the recipients that could make the best use of the modernization fund were
the ones who due to historical reasons (the existence of an arsenal and special procurement in
Osaka’s case), potentially had a higher elasticity of productivity or production level with
respect to policy intervention. In this regard, to maximize the effects of a policy towards the
development of regional economic resilience within areas with limited resources, it is not
necessarily effective to distribute the resources too evenly across local SMEs, and some sort of
targeting strategy taking advantage of the ability to collect information about the local
socioeconomic conditions, including the local areas’ historical background, might be more
productive30.
6.3 Limitations and Future Work
To show future directions for this thesis, I mention the limitations and future work. In this
thesis, I have two main unaddressed issues specific to resilience and policy study. One is that
I could only focus on a portion of the process of regional economic resilience in any chapter.
In conducting empirical investigations consistent with the definition of resilience stated above,
I had to cover the observations before, during, and after an exogenous shock. Despite this
30 The recent literature, such as Tingvall & Videnord (2020), emphasizes the necessity and advantage of targeting strategies in regional innovation policies.
171
necessity, Chapter 3 only focused on pre-shock actions. In Chapter 4, the existence of a specific
shock was not presumed, or my targeted shock and technological change was too slow-burning
to analyze by utilizing only ten years’ worth of data. Chapter 5 only focuses on the process of
bouncing back after a shock (from the early to late 1950s), so the assessment of shock
sensitivity based on a before-after comparison could not be conducted. Although Chapter 2
covers the process of regional economic resilience (both before and after an earthquake), the
existence and effects of a shock were merely indirect and based on the observations of interest.
Thus, an empirical investigation covering both direct and indirect shock effects and
incorporating a before-after comparison could be imperative for a more comprehensive
resilience analysis.
The other limitation is a lack of an examination of the vertical interrelationship between
economic agents in developing (in the sources of) regional economic resilience. From a local
industrial perspective, an explicit consideration of the keiretsu relationship or the relative
position of a firm can be necessary, though due to the complexity in analysis and lack of
information, I ended up by regarding the firms’ positions as parallel. Taking Chapter 2 as an
example, whether a firm updates sourcing strategies might depend on its relative position in
the supply chain network. There can be a difference in the importance and form of sourcing
strategies between upstream and downstream firms. The relative position also matters in the
issue in Chapter 4 because according to interviews with local policy managers, some SMEs
conduct R&D activities depending on the demand from prime contractors.
From a policy perspective, an in-depth examination of the relationship between local and
central governments can be essential. In Chapter 3, coercion by the upper-level government
could be a driver of the early development of municipal BCP, although I could not explicitly
consider it because of the difficulty in analyzing this with the spatial econometric model and
the unobservability of the cases of coercion. In Chapter 4, by using a comprehensive policy
172
database covering various aspects, for example, firm-level recipient information for both local
and central governments, I could have explicitly examined the complementarity of the local
governments’ policies with respect to those by the central government. In this regard, it can be
particularly crucial to evolve my investigations in this thesis further by incorporating both
horizontal and vertical linkages.
From the perspective of a general empirical analysis, there are two main unsolved issues.
The first issue is the insufficient identification of the causal relationship. The treatment
variables analyzed in Chapters 2 (the existence of suppliers in specific regions), 4, and 5
(competitive selection of the recipients of a subsidy or fund) had selective disposition, which
made the adequate estimation of causal effects quite difficult. By utilizing several
methodologies, such as fixed effects, propensity score matching, and the DD(D) method, I
could partially address the problem caused by confounding factors in these chapters. However,
I could not tackle the identification concern about reverse causality due to the absence of
(quasi-)random variation in the treatment variables. In Chapters 4 and 5, to solve this problem,
detailed information about the selection and scoring process would be necessary in order to
apply methodologies, such as regression discontinuity design and instrumental variables, that
could address reverse causality, as well as confounding factors.
The second issue is the absence of a theoretical foundation and thus an in-depth
understanding of the mechanism underlying the estimated associations. In this thesis, I showed
that the estimated results had considerable heterogeneity depending on the following: the
firms’ characteristics and the surrounding business environment (Chapter 2); the local
industrial structure, geographical positions, and policy schemes (Chapter 4); and historical path
dependency (Chapter 5). However, the explanation of the heterogeneity in this thesis only
relied on anecdotal evidence or conceptual supposition, which cannot necessarily be
convincing in light of the recent empirical literature in the field of economics. As one of the
173
ways to make the obtained results more convincing and to examine the validity of the estimated
effects, a theoretical prediction and empirical framework consistent with the prediction could
be required.
174
Acknowledgments
This thesis would not have been completed without the help of many people. I have
greatly benefited from the constant support of my committee. My supervisor Professor Morito
Tsutsumi has supported me generously during the time in which I was an undergraduate student
for six years. He has educated me very thoughtfully and flexibly, considering my aptitude, my
career path, my way of thinking, and so on. I am also grateful to two mentors at Hitotsubashi
University, Professor Kentaro Nakajima and Professor Hiroyuki Okamuro. Through the
research projects at the TDB Center for Advanced Empirical Research on Enterprise and
Economy; TDB-CAREE, they taught me how to conduct and present empirical analysis in the
field of regional science and small business economics from scratch. The members of my
committee at the University of Tsukuba, Professor Tomokazu Arita, Professor Nobuyuki
Harada, and Professor Mitsuru Ota, carefully read this study and provided many helpful
comments. All of them also served as members of the advisory group of my Master’s and/or
graduation thesis.
The chapters in this thesis are part of our achievements at the TDB-CAREE and the
SciREX Center in the National Graduate Institute for Policy Studies. I would also like to take
an opportunity to thank the many faculty members at these institutions. In TDB-CAREE,
members including Takeo Goto, Yasushi Hara, Yoshiki Hiramine, Shinya Kitamura, and
Makiko Koto have supported our projects since the foundation. The joint research project
conducted with So Morikawa in the SciREX Center during my summer internship provided a
new direction on my research from the perspective of public administration.
Finally, and most importantly, I am grateful to my family. My parents showed a deep
understanding of my decision to go on to a doctoral course despite the economically and
socially severe situation surrounding Ph.D. students in Japan and dedicatedly supported me.
175
My younger brother, an alumnus of Hitotsubashi, supported my life in Kunitachi and as an
economic historian, provided me with research assistance in the empirical analysis.
176
References
• Abe, T. (2006). Modern Osaka Economic History[近代大阪経済史]. Osaka University Publisher.[in Japanese]
• Aerts, K., & Schmidt, T. (2008). Two for the price of one? Additionality effects of R&D subsidies: A comparison between Flanders and Germany. Research Policy, 37(5), 806-822.
• AIST. (1964). JIS-Certified Plants List[JIS表示許可工場名簿].[in Japanese] • Allen, T., & Donaldson, D. (2018). The geography of path dependence. Unpublished
manuscript. • Andrew, S. A., & Carr, J. B. (2013). Mitigating uncertainty and risk in planning for
regional preparedness: The role of bonding and bridging relationships. Urban Studies, 50(4), 709-724.
• Angulo, A. M., Mur, J., & Trívez, F. J. (2018). Measuring resilience to economic shocks: an application to Spain. The Annals of Regional Science, 60(2), 349-373.
• Anselin, L. (1988). Spatial econometrics: methods and models. Springer Science & Business Media.
• Arcuri, G., & Levratto, N. (2020). Early stage SME bankruptcy: does the local banking market matter?. Small Business Economics, 54(2), 421-436.
• Arestis, P., Chortareas, G., & Magkonis, G. (2015). The financial development and growth nexus: A meta‐analysis. Journal of Economic Surveys, 29(3), 549-565.
• Baldwin, J. R., & Brown, W. M. (2004). Regional manufacturing employment volatility in Canada: The effects of specialisation and trade. Papers in Regional Science, 83(3), 519-541.
• Balland, P. A., Boschma, R., Crespo, J., & Rigby, D. L. (2019). Smart specialization policy in the European Union: relatedness, knowledge complexity and regional diversification. Regional Studies, 53(9), 1252-1268.
• Barrot, J. N., & Sauvagnat, J. (2016). Input specificity and the propagation of idiosyncratic shocks in production networks. The Quarterly Journal of Economics, 131(3), 1543-1592.
• Beaudry, C., & Schiffauerova, A. (2009). Who's right, Marshall or Jacobs? The localization versus urbanization debate. Research policy, 38(2), 318-337.
• Berkes, F. (2007). Understanding uncertainty and reducing vulnerability: lessons from resilience thinking. Natural hazards, 41(2), 283-295.
• Bernard, A. B., & Moxnes, A. (2018). Networks and trade. Annual Review of Economics, 10, 65-85.
• Bernini, C., & Pellegrini, G. (2011). How are growth and productivity in private firms affected by public subsidy? Evidence from a regional policy. Regional Science and Urban Economics, 41(3), 253-265.
• Bianchi, N., & Giorcelli, M. (2018). Reconstruction aid, public infrastructure, and economic development. Available at SSRN 3153139.
• Bivand, R., Altman, M., Anselin, L., Assunção, R., Berke, O., Bernat, A., & Blanchet, G. (2015). Package ‘spdep’. https://cran.r-project.org/web/packages/spdep/spdep.pdf. Last accessed: July 22, 2020.
• Blind, K. (2019). Standardization and Standards as Science and Innovation Indicators. In Springer Handbook of Science and Technology Indicators (pp. 1057-1068). Springer, Cham.
• Bodenhorn, H., & Cuberes, D. (2018). Finance and urbanization in early nineteenth-century New York. Journal of Urban Economics, 104, 47-58.
177
• Borrás, S., & Edquist, C. (2013). The choice of innovation policy instruments. Technological Forecasting and Social Change, 80(8), 1513-1522.
• Boschma, R. (2005). Proximity and innovation: a critical assessment. Regional studies, 39(1), 61-74.
• Boschma, R. (2015). Towards an evolutionary perspective on regional resilience. Regional Studies, 49(5), 733-751.
• Boschma, R. A., & Frenken, K. (2006). Why is economic geography not an evolutionary science? Towards an evolutionary economic geography. Journal of economic geography, 6(3), 273-302.
• Boschma, R., & Frenken, K. (2011). The emerging empirics of evolutionary economic geography. Journal of economic geography, 11(2), 295-307.
• Boschma, R., & Iammarino, S. (2009). Related variety, trade linkages, and regional growth in Italy. Economic geography, 85(3), 289-311.
• Boschma, R., Minondo, A., & Navarro, M. (2012). Related variety and regional growth in Spain. Papers in Regional Science, 91(2), 241-256.
• Bosker, M., Brakman, S., Garretsen, H., & Schramm, M. (2007). Looking for multiple equilibria when geography matters: German city growth and the WWII shock. Journal of Urban Economics, 61(1), 152-169.
• Brakman, S., & van Marrewijk, C. (2015). Factor prices and geographical economics. In Handbook of Research Methods and Applications in Economic Geography. Edward Elgar Publishing.
• Brakman, S., Garretsen, H., & Schramm, M. (2004). The strategic bombing of German cities during World War II and its impact on city growth. Journal of Economic Geography, 4(2), 201-218.
• Brakman, S., Garretsen, H., & van Marrewijk, C. (2009). The new introduction to geographical economics. Cambridge University Press.
• Brakman, S., Garretsen, H., & van Marrewijk, C. (2015). Regional resilience across Europe: On urbanisation and the initial impact of the Great Recession. Cambridge Journal of Regions, Economy and Society, 8(2), 225-240.
• Brakman, S., Garretsen, H., & van Marrewijk, C. (2019). An Introduction to Geographical and Urban Economics: A Spiky World. Cambridge University Press.
• Briozzo, A., & Cardone-Riportella, C. (2016). Spanish SMEs’ subsidized and guaranteed credit during economic crisis: a regional perspective. Regional Studies, 50(3), 496-512.
• Bristow, G., & Healy, A. (2018). Innovation and regional economic resilience: an exploratory analysis. The Annals of Regional Science, 60(2), 265-284.
• Bristow, G., & Healy, A. (2020). Regional resilience: an agency perspective. In Handbook on Regional Economic Resilience. Edward Elgar Publishing.
• Bristow, G., & Healy, A. (Eds.). (2018). Economic crisis and the resilience of regions: A European study. Edward Elgar Publishing.
• Bronzini, R., & Piselli, P. (2016). The impact of R&D subsidies on firm innovation. Research Policy, 45(2), 442-457.
• Brueckner, J. K. (2003). Strategic interaction among governments: An overview of empirical studies. International Regional Science Review, 26(2), 175-188.
• Cainelli, G., & Ganau, R. (2019). Related variety and firm heterogeneity. What really matters for short-run firm growth?. Entrepreneurship & Regional Development, 31(9-10), 768-784.
• Callaway, B., & Sant'Anna, P. H. (2019). Difference-in-differences with multiple time periods. Available at SSRN 3148250.
178
• Calomiris, C. W., & Himmelberg, C. P. (1993). Directed credit programs for agriculture and industry: arguments from theory and fact. The World Bank Economic Review, 7(suppl_1), 113-138.
• Cannone, G., & Ughetto, E. (2014). Funding innovation at regional level: an analysis of a public policy intervention in the Piedmont region. Regional Studies, 48(2), 270-283.
• Cantoni, D., & Yuchtman, N. (2020). Historical Natural Experiments: Bridging Economics and Economic History (No. w26754). National Bureau of Economic Research.
• CAO. (2011). Annual Report on the Japanese Economy and Public Finance 2011. Available at http://www5.cao.go.jp/keizai3/2011/0722wp-keizai/summary.html. Last accessed: October 29, 2018.
• CAO. (2015). White Paper on Disaster Management in Japan 2015. Available in http://www.bousai.go.jp/kaigirep/hakusho/pdf/WP2015_DM_Full_Version.pdf. Last accessed: October 13 2018.
• CAO. (2016). Thinking about How to Reduce Disaster Risks at the National and Community Levels. Available in https://www.cas.go.jp/jp/seisaku/kokudo_kyoujinka/en/. Last accessed: October 7 2018.
• CAO. (2018). Fundamental Plan for National Resilience: For Building a Strong and Flexible Country. Available at https://www.cas.go.jp/jp/seisaku/kokudo_kyoujinka/en/fundamental_plan.pdf. Last accessed: October 1, 2019.
• Carbó Valverde, S., López del Paso, R., & Rodríguez Fernández, F. (2007). Financial innovations in banking: Impact on regional growth. Regional Studies, 41(3), 311-326.
• Carpenter, R. E., & Petersen, B. C. (2002). Is the growth of small firms constrained by internal finance?. Review of Economics and statistics, 84(2), 298-309.
• Carvalho, V., Nirei, M., Saito, Y., & Tahbaz-Salehi, A. (2016). Supply Chain Disruptions: Evidence from the Great East Japan Earthquake. Faculty of Economics, University of Cambridge.
• Cavallo, E., & Noy, I. (2011). Natural disasters and the economy—a survey. International Review of Environmental and Resource Economics, 5(1), 63-102.
• Chang, H. H., van Marrewijk, C., & Schramm, M. (2015). Empirical studies in geographical economics. In Handbook of Research Methods and Applications in Economic Geography. Edward Elgar Publishing.
• Chapple, K., & Lester, T. W. (2010). The resilient regional labour market? The US case. Cambridge journal of regions, economy and society, 3(1), 85-104.
• Christensen, T., Lægreid, P., & Rykkja, L. H. (2016). Organizing for crisis management: building governance capacity and legitimacy. Public Administration Review, 76(6), 887-897.
• Christopher, M. (2016). Logistics & supply chain management. Pearson UK. • Clarysse, B., Wright, M., & Muster, P. (2009). Behavioural additionality of public
subsidies. A learning perspective. Research Policy, 38(10), 1517-1533. • Cliff, A. D., & Ord, J. K. (1981). Spatial processes: models & applications. Taylor &
Francis. • Col, J. M. (2007). Managing disasters: The role of local government. Public
Administration Review, 67, 114-124. • Cole, M. A., Elliott, R. J., Okubo, T., & Strobl, E. (2017). Pre-disaster planning and post-
disaster aid: Examining the impact of the great East Japan Earthquake. International journal of disaster risk reduction, 21, 291-302.
179
• Cole, M. A., Elliott, R. J., Okubo, T., & Strobl, E. (2019). Natural disasters and spatial heterogeneity in damages: the birth, life and death of manufacturing plants. Journal of Economic Geography, 19(2), 373-408.
• Compensation Agency. (1948) Compensation Plants List[賠償指定工場名簿 ].[in Japanese]
• Content, J., & Frenken, K. (2016). Related variety and economic development: a literature review. European Planning Studies, 24(12), 2097-2112.
• Cook, S. J., An, S. H., & Favero, N. (2019). Beyond policy diffusion: Spatial econometric models of public administration. Journal of Public Administration Research and Theory, 29(4), 591-608.
• Crespy, C., Heraud, J. A., & Perry, B. (2007). Multi-level governance, regions and science in France: between competition and equality. Regional Studies, 41(8), 1069-1084.
• Criscuolo, C., Martin, R., Overman, H. G., & Van Reenen, J. (2019). Some causal effects of an industrial policy. American Economic Review, 109(1), 48-85.
• Czarnitzki, D., & Hussinger, K. (2018). Input and output additionality. Applied Economics, 50(12), 1324-1341.
• Czarnitzki, D., & Lopes-Bento, C. (2013). Value for money? New microeconometric evidence on public R&D grants in Flanders. Research Policy, 42(1), 76-89.
• Davis, D. R., & Weinstein, D. E. (2002). Bones, bombs, and break points: the geography of economic activity. American Economic Review, 92(5), 1269-1289.
• Davis, D. R., & Weinstein, D. E. (2008). A search for multiple equilibria in urban industrial structure. Journal of Regional Science, 48(1), 29-65.
• Davlasheridze, M., & Geylani, P. C. (2017). Small Business vulnerability to floods and the effects of disaster loans. Small Business Economics, 49(4), 865-888.
• Dawley, S., Pike, A., & Tomaney, J. (2010). Towards the resilient region?. Local Economy, 25(8), 650-667.
• De Groot, H. L., Poot, J., & Smit, M. J. (2016). Which agglomeration externalities matter most and why?. Journal of Economic Surveys, 30(4), 756-782.
• Deng, G., Gan, L., & Hernandez, M. A. (2015). Do natural disasters cause an excessive fear of heights? Evidence from the Wenchuan earthquake. Journal of Urban Economics, 90, 79-89.
• Di Caro, P. (2018). To be (or not to be) resilient over time: facts and causes. The Annals of Regional Science, 60(2), 375-392.
• Di Caro, P., & Fratesi, U. (2018). Regional determinants of economic resilience. The Annals of Regional Science, 60(2), 235-240.
• Doh, S., & Kim, B. (2014). Government support for SME innovations in the regional industries: The case of government financial support program in South Korea. Research Policy, 43(9), 1557-1569.
• Duranton, G., & Puga, D. (2004). Micro-foundations of urban agglomeration economies. In Handbook of regional and urban economics (Vol. 4, pp. 2063-2117). Elsevier.
• Dzigbede, K., Gehl, S. B., & Willoughby, K. (2020). Disaster resiliency of US local governments: Insights to strengthen local response and recovery from the COVID-19 pandemic. Public Administration Review, 80(4), 634-643.
• Entress, R., Tyler, J., & Sadiq, A. A. (2020). Managing mass fatalities during COVID‐19: Lessons for promoting community resilience during global pandemics. Public Administration Review, 80(5), 856-861.
180
• Eriksen, C., & Prior, T. (2013). Defining the importance of mental preparedness for risk communication and residents well-prepared for wildfire. International Journal of Disaster Risk Reduction, 6, 87-97.
• ESPON, E. (2014). ECR2–Economic crisis: Resilience of regions. Scientific Report Luxembourg: ESPON.
• Evenhuis, E. (2017). New directions in researching regional economic resilience and adaptation. Geography Compass, 11(11), 1-15.
• Faggian, A., Gemmiti, R., Jaquet, T., & Santini, I. (2018). Regional economic resilience: the experience of the Italian local labor systems. The Annals of Regional Science, 60(2), 393-410.
• Falck, O., Heblich, S., & Kipar, S. (2010). Industrial innovation: direct evidence from a cluster-oriented policy. Regional Science and Urban Economics, 40(6), 574-582.
• FDMA. (2015a). White Paper on Disaster Management 2015. Available at http://www.bousai.go.jp/kyoiku/panf/pdf/WP2015_DM_Full_Version.pdf. Last accessed: October 23, 2019.
• FDMA. (2015b). The Guide to Develop Business Continuity Planning for Municipalities. Available at http://www.bousai.go.jp/taisaku/chihogyoumukeizoku/pdf/H27bcpguide.pdf. Last accessed: October 23, 2019. [in Japanese]
• Feldman, M. P., & Audretsch, D. B. (1999). Innovation in cities: science-based diversity, specialization and localized competition. European Economic Review, 43(2), 409-429.
• Fernández-Ribas, A. (2009). Public support to private innovation in multi-level governance systems: an empirical investigation. Science and Public Policy, 36(6), 457-467.
• Fiksel, J., Polyviou, M., Croxton, K. L., & Pettit, T. J. (2015). From risk to resilience: Learning to deal with disruption. MIT Sloan management review, 56(2), 79-86.
• Fingleton, B., Garretsen, H., & Martin, R. (2012). Recessionary shocks and regional employment: evidence on the resilience of UK regions. Journal of regional science, 52(1), 109-133.
• Fingleton, B., Garretsen, H., & Martin, R. (2015). Shocking aspects of monetary union: the vulnerability of regions in Euroland. Journal of Economic Geography, 15(5), 907-934.
• Flanagan, K., Uyarra, E., & Laranja, M. (2011). The ‘policy mix’ for innovation: rethinking innovation policy in a multi-level, multi-actor context. Research Policy, 40(5), 702-713.
• Forslid, R., & Ottaviano, G. I. (2003). An analytically solvable core‐periphery model. Journal of Economic Geography, 3(3), 229-240.
• Franzese Jr, R. J., & Hays, J. C. (2008). Interdependence in comparative politics: Substance, theory, empirics, substance. Comparative Political Studies, 41(4-5), 742-780.
• Fratesi, U., & Perucca, G. (2018). Territorial capital and the resilience of European regions. The Annals of Regional Science, 60(2), 241-264.
• Freel, M., Liu, R., & Rammer, C. (2019). The export additionality of innovation policy. Industrial and Corporate Change, 28(5), 1257-1277.
• Frenken, K., Van Oort, F., & Verburg, T. (2007). Related variety, unrelated variety and regional economic growth. Regional studies, 41(5), 685-697.
• Fritsch, M., & Kublina, S. (2018). Related variety, unrelated variety and regional growth: the role of absorptive capacity and entrepreneurship. Regional Studies, 52(10), 1360-1371.
• Fritsch, M., & Storey, D. J. (2014). Entrepreneurship in a Regional Context: Historical Roots, Recent Developments and Future Challenges. Regional Studies, 48(6), 939-954.
181
• Fujita, M., & Thisse, J. F. (2013). Economics of agglomeration: cities, industrial location, and globalization. Cambridge University Press, Cambridge.
• Fujita, M., Hamaguchi, N., and Kameyama, Y. (2018). Spatial Economics in the Age of Declining Population. Nikkei Publishing. [in Japanese]
• Fujita, M., Krugman, P. R., & Venables, A. (1999). The spatial economy: Cities, regions, and international trade. MIT press.
• Ganotakis, P., & Love, J. H. (2012). Export propensity, export intensity and firm performance: The role of the entrepreneurial founding team. Journal of International Business Studies, 43(8), 693-718.
• Gervais, A. (2018). Uncertainty, risk aversion and international trade. Journal of International Economics, 115, 145-158.
• Glaeser, E. L., Kallal, H. D., Scheinkman, J. A., & Shleifer, A. (1992). Growth in cities. Journal of political economy, 100(6), 1126-1152.
• Golovko, E., & Valentini, G. (2011). Exploring the complementarity between innovation and export for SMEs’ growth. Journal of International Business Studies, 42(3), 362-380.
• Goodman-Bacon, A. (2018). Difference-in-differences with variation in treatment timing (No. w25018). National Bureau of Economic Research.
• Grabher, G. (1993). The weakness of strong ties; the lock-in of regional development in Ruhr area. The embedded firm; on the socioeconomics of industrial networks, 255-277.
• Grant, R. M. (2016). Contemporary strategy analysis: Text and cases edition. John Wiley & Sons.
• Greenstone, M., Hornbeck, R., & Moretti, E. (2010). Identifying agglomeration spillovers: Evidence from winners and losers of large plant openings. Journal of Political Economy, 118(3), 536-598.
• Guiso, L., Sapienza, P., & Zingales, L. (2004). Does local financial development matter?. The Quarterly Journal of Economics, 119(3), 929-969.
• Harner, S. (2012). BTW, Get Ready for a 34 Meter Tsunami. Forbes. Available in https://www.forbes.com/sites/stephenharner/2012/04/02/btw-get-ready-for-a-34-meter-tsunami. Last accessed: October 7 2018."
• Hassink, R. (2010). Regional resilience: a promising concept to explain differences in regional economic adaptability?. Cambridge journal of regions, economy and society, 3(1), 45-58.
• Head, K., & Mayer, T. (2004). The empirics of agglomeration and trade. In Handbook of regional and urban economics (Vol. 4, pp. 2609-2669). Elsevier.
• Helpman, E. (1998). The size of regions. In Topics in public economics: Theoretical and applied analysis (pp. 33-54). Cambridge University Press.
• Henstra, D. (2010). Evaluating local government emergency management programs: What framework should public managers adopt?. Public Administration Review, 70(2), 236-246. https://doi.org/10.1111/j.1540-6210.2010.02130.x
• Holling, C. (1985). Resilience of ecosystems: local surprise and global change (No. 5, pp. 228-269). Cambridge University Press.
• Holling, C. S. (1973). Resilience and stability of ecological systems. Annual review of ecology and systematics, 4(1), 1-23.
• Hopp, W. J., Iravani, S. M., & Liu, Z. (2012). Mitigating the impact of disruptions in supply chains. In Supply chain disruptions (pp. 21-49). Springer, London.
• Hornbeck, R., & Keniston, D. (2017). Creative destruction: Barriers to urban growth and the Great Boston Fire of 1872. American Economic Review, 107(6), 1365-98.
182
• Høyvarde Clausen, T. (2013). Firm heterogeneity within industries: how important is ‘industry’to innovation?. Technology Analysis & Strategic Management, 25(5), 527-542.
• Hsu, P. H., Lee, H. H., Peng, S. C., & Yi, L. (2018). Natural disasters, technology diversity, and operating performance. Review of Economics and Statistics, 100(4), 619-630.
• Imai, K., & Van Dyk, D. A. (2004). Causal inference with general treatment regimes: Generalizing the propensity score. Journal of the American Statistical Association, 99(467), 854-866.
• Imai, K., King, G., & Stuart, E. A. (2008). Misunderstandings between experimentalists and observationalists about causal inference. Journal of the Royal Statistical Society: Series A (statistics in society), 171(2), 481-502.
• Isouchi, C. (2017). District continuity plans for large-scale disaster coordination: Case study in Kagawa District. Journal of Disaster Research, 12(4), 733-740.
• Jacobs, J. (1969). The economy of cities. Random House, New York. • Kakderi, C., & Tasopoulou, A. (2017). Regional economic resilience: the role of national
and regional policies. European Planning Studies, 25(8), 1435-1453. • Kamal, F., & Sundaram, A. (2019). Do institutions determine economic Geography?
Evidence from the concentration of foreign suppliers. Journal of Urban Economics, 110, 89-101.
• Karlan, D., & Morduch, J. (2010). Access to finance. In Handbook of development economics (Vol. 5, pp. 4703-4784). Elsevier.
• Kashiwagi, Y. (2019). Postdisaster subsidies for small and medium firms: Insights for effective targeting. Asian Development Bank Economics Working Paper Series, (597).
• Keim, M. E. (2008). Building human resilience: the role of public health preparedness and response as an adaptation to climate change. American Journal of Preventive Medicine, 35(5), 508-516.
• Kendall, J. (2012). Local financial development and growth. Journal of Banking & Finance, 36(5), 1548-1562.
• Kinki Region Cooperating Industry Council & Osaka Cooperating Industry Union. (1943). Cooperating Plants List in Kinki Region[近畿地區發註工塲竝ニ協力工塲名簿].[in Japanese]
• Kitagawa, F. (2007). The regionalization of science and innovation governance in Japan? Regional Studies, 41(8), 1099-1114.
• Kitsos, A., & Bishop, P. (2018). Economic resilience in Great Britain: the crisis impact and its determining factors for local authority districts. The Annals of Regional Science, 60(2), 329-347.
• Kiyota, K., & Okazaki, T. (2005). Foreign technology acquisition policy and firm performance in Japan, 1957–1970: Micro-aspects of industrial policy. International Journal of Industrial Organization, 23(7-8), 563-586.
• Kleinknecht, A. (1987). Measuring R & D in small firms: How much are we missing? The Journal of Industrial Economics, 36(2), 253-256.
• Koschatzky, K., & Kroll, H. (2007). Which side of the coin? The regional governance of science and innovation. Regional Studies, 41(8), 1115-1127.
• Kousky, C. (2014). Informing climate adaptation: A review of the economic costs of natural disasters. Energy Economics, 46, 576-592.
• Krugman, P. (1991). Increasing returns and economic geography. Journal of political economy, 99(3), 483-499.
• Lanahan, L. (2016). Multilevel public funding for small business innovation: a review of US state SBIR match programs. Journal of Technology Transfer, 41(2), 220-249.
183
• Laranja, M., Uyarra, E., & Flanagan, K. (2008). Policies for science, technology and innovation: translating rationales into regional policies in a multi-level setting. Research Policy, 37(5), 823-835.
• Lee, Y. S. (2018). Government guaranteed small business loans and regional growth. Journal of Business Venturing, 33(1), 70-83.
• Leite, W. (2016). Practical propensity score methods using R. Sage Publications. • MacKinnon, D., & Derickson, K. D. (2013). From resilience to resourcefulness: A critique
of resilience policy and activism. Progress in human geography, 37(2), 253-270. • MAFF (2011) The Primary Data of the Municipalities and the Agricultural Settlements
Damaged by the Great East Japan Earthquake. Available in http://www.maff.go.jp/j/tokei/census/afc/2010/saigai.html. Last accessed: October 13 2018. [in Japanese]
• Magurran, A. E., & McGill, B. J. (Eds.). (2011). Biological diversity: frontiers in measurement and assessment. Oxford University Press.
• Malmendier, U., & Nagel, S. (2016). Learning from inflation experiences. The Quarterly Journal of Economics, 131(1), 53-87.
• Malmendier, U., Pouzo, D., & Vanasco, V. (2020). Investor experiences and international capital flows?. Journal of International Economics, forthcoming.
• Martín-García, R., & Santor, J. M. (2019). Public guarantees: a countercyclical instrument for SME growth. Evidence from the Spanish Region of Madrid. Small Business Economics, 1-23.
• Martin, P., Mayer, T., & Mayneris, F. (2011). Public support to clusters: a firm level study of French “Local Productive Systems”. Regional Science and Urban Economics, 41(2), 108-123.
• Martin, R. (2012). Regional economic resilience, hysteresis and recessionary shocks. Journal of economic geography, 12(1), 1-32.
• Martin, R. L. (2018). Shocking aspects of regional development: Towards an economic geography of resilience. In The new Oxford handbook of economic geography.
• Martin, R., & Simmie, J. (2008). Path dependence and local innovation systems in city-regions. Innovation, 10(2-3), 183-196.
• Martin, R., & Sunley, P. (2006). Path dependence and regional economic evolution. Journal of economic geography, 6(4), 395-437.
• Martin, R., & Sunley, P. (2015). On the notion of regional economic resilience: conceptualization and explanation. Journal of Economic Geography, 15(1), 1-42.
• Matsushita, T. (2012). The analysis of Relations between Industrial Agglomerations of Osaka and the Arsenal of Osaka : A case of Technologies "Iron and steel, Aluminum, Machinery-metalworking"[大阪砲兵工廠と大阪産業集積との関係性: 鉄鋼, アルミニウム, 機械金属加工技術から考察]. Sankaiken Ronshu, (24), 7-19.[in Japanese]
• Matyas, D., & Pelling, M. (2015). Positioning resilience for 2015: the role of resistance, incremental adjustment and transformation in disaster risk management policy. Disasters, 39(s1), s1-s18.
• Mazzola, F., Cascio, I. L., Epifanio, R., & Di Giacomo, G. (2018). Territorial capital and growth over the Great Recession: a local analysis for Italy. The Annals of Regional Science, 60(2), 411-441.
• McCann, P., & Ortega-Argilés, R. (2013). Redesigning and reforming European regional policy: The reasons, the logic, and the outcomes. International Regional Science Review, 36(3), 424-445.
184
• McCann, P., & Van Oort, F. (2019). Theories of agglomeration and regional economic growth: a historical review. In Handbook of regional growth and development theories. Edward Elgar Publishing.
• McMillen, D. P. (1992). Probit with spatial autocorrelation. Journal of Regional Science, 32(3), 335-348.
• Meerow, S., Newell, J. P., & Stults, M. (2016). Defining urban resilience: A review. Landscape and urban planning, 147, 38-49.
• Mehiriz, K., & Gosselin, P. (2016). Municipalities' preparedness for weather hazards and response to weather warnings. PloS One, 11(9), e0163390.
• Metcalfe, J. S., Foster, J., & Ramlogan, R. (2006). Adaptive economic growth. Cambridge Journal of Economics, 30(1), 7-32.
• Miguel, E., & Roland, G. (2011). The long-run impact of bombing Vietnam. Journal of development Economics, 96(1), 1-15.
• MITI, ed. (1963). The official history of Japanese commercial and industrial policy[商工政策史] (Vol. 12). Shoko Seisaku-shi Kenkyukai.[in Japanese]
• Miyake, K. (1993). The Study of the Osaka Arsenal[大阪砲兵工廠の研究 ]. Shibunkaku.[in Japanese]
• Nachum, L. (2004). Geographic and industrial diversification of developing country firms. Journal of Management Studies, 41(2), 273-294.
• Nakamura, S., Akiya, S., Kiyonari, T., Yamazaki, M., & Bando, T. (1981). Modern small business economic history[現代中小企業史]. Nikkei Inc.[in Japanese]
• Nakamura, T., & Ohashi, H. (2012). Effects of re-invention on industry growth and productivity: evidence from steel refining technology in Japan, 1957–1968. Economics of Innovation and New Technology, 21(4), 411-426.
• Naoi, M., Seko, M., & Sumita, K. (2009). Earthquake risk and housing prices in Japan: Evidence before and after massive earthquakes. Regional Science and Urban Economics, 39(6), 658-669.
• Nathan, M., & Overman, H. (2013). Agglomeration, clusters, and industrial policy. Oxford Review of Economic Policy, 29(2), 383-404.
• Neffke, F., & Henning, M. (2013). Skill relatedness and firm diversification. Strategic Management Journal, 34(3), 297-316.
• Neumark, D., & Simpson, H. (2015). Place-based policies. In Handbook of regional and urban economics (Vol. 5, pp. 1197-1287). Elsevier.
• Nishimura, J., & Okamuro, H. (2011a). R&D productivity and the organization of cluster policy: an empirical evaluation of the Industrial Cluster Project in Japan. Journal of Technology Transfer, 36(2), 117-144.
• Nishimura, J., & Okamuro, H. (2011b). Subsidy and networking: the effects of direct and indirect support programs of the cluster program. Research Policy, 40(5), 714-727.
• Nunn, N. (2014). Historical development. In Handbook of economic growth (Vol. 2, pp. 347-402). Elsevier.
• O'Toole Jr, L. J., & Meier, K. J. (2014). Public management, context, and performance: In quest of a more general theory. Journal of Public Administration Research and Theory, 25(1), 237-256. https://doi.org/10.1093/jopart/muu011
• Oates, W. E. (1972). Fiscal federalism. Harcourt Brace Jovanovich, New York. • OCCI. (1950). Yearbook of Osaka Economies[大阪経済年鑑].[in Japanese] • OECD. (2020). The Territorial Impact of COVID-19: Managing the Crisis across Levels
of Government.
185
• Okamuro, H., & Nishimura, J. (2015a). Local management of national cluster policies: comparative case studies of Japanese, German, and French biotechnology clusters. Administrative Sciences, 5(4), 213-239.
• Okamuro, H., & Nishimura, J. (2015b). Not just financial support? Another role of public subsidy in university-industry research collaborations. Economics of Innovation and New Technology, 24(7), 633-659.
• Okamuro, H., & Nishimura, J. (2020a). Effects of multilevel policy mix of public R&D subsidies: empirical evidence from Japanese local SMEs. CCES Discussion Paper Series, No. 70, Hitotsubashi University Graduate School of Economics, Tokyo, Japan.
• Okamuro, H., & Nishimura, J. (2020b). What shapes local innovation policies? Empirical evidence from Japanese cities. Administrative Sciences, 10 (11).
• Okamuro, H., Nishimura, J., & Kitagawa, F. (2019). Multilevel policy governance and territorial adaptability: evidence from Japanese SME innovation programmes. Regional Studies, 53(6), 803-814.
• Okubo, T., Okazaki, T., & Tomiura, E. (2016). Industrial Cluster Policy and Transaction Networks: evidence from firm-level data in Japan. RIETI Discussion Paper Series, 16-E-071.
• Okura, M., Hashimoto, A., & Arai, H. (2019). Community and municipal organizational characteristics impacting the completion of disaster plans by local public entities in Japan. International Journal of Disaster Risk Reduction, 36, 101087.
• Onuma, H., Shin, K. J., & Managi, S. (2017). Reduction of future disaster damages by learning from disaster experiences. Natural Hazards, 87(3), 1435-1452.
• ORCIE. (1953). Subcontract Industry in Munition Manufacturing[兵器産業における下請工業]. Keiken, (56).[in Japanese]
• Osaka City. (1953). History of Osaka City in Showa[昭和大阪市史].[in Japanese] • Osaka Prefecture. (1951). The Program for Compensation of Loss in SME Modernization
Fund[中小企業近代化資金融資に対する損失補償制度]. [in Japanese] • Osaka Prefecture. (1952). Prefectural Administration Pictorial.[in Japanese] • Osaka Prefecture. (1954). Borrowers' List of the SME Modernization Fund[中小企業設備近代化融資工場名簿 -中小企業近代化資金融資に対する損失補償制度に基く貸付状況-].[in Japanese]
• Osaka Prefecture. (1968). 100-year History of Osaka[大阪百年史].[in Japanese] • Oyun, G. (2017). Interstate spillovers, fiscal decentralization, and public spending on
medicaid home‐and community‐based services. Public Administration Review, 77(4), 566-578.
• Palich, L. E., Cardinal, L. B., & Miller, C. C. (2000). Curvilinearity in the diversification–performance linkage: an examination of over three decades of research. Strategic management journal, 21(2), 155-174.
• Pendall, R., Foster, K. A., & Cowell, M. (2010). Resilience and regions: building understanding of the metaphor. Cambridge Journal of Regions, Economy and Society, 3(1), 71-84.
• Perry, B. (2007). The multi-level governance of science policy in England. Regional Studies, 41(8), 1051-1067.
• Pettit, T. J., Fiksel, J., & Croxton, K. L. (2010). Ensuring supply chain resilience: development of a conceptual framework. Journal of business logistics, 31(1), 1-21.
• Pike, A., Dawley, S., & Tomaney, J. (2010). Resilience, adaptation and adaptability. Cambridge journal of regions, economy and society, 3(1), 59-70.
186
• Pinkse, J., & Slade, M. E. (1998). Contracting in space: An application of spatial statistics to discrete-choice models. Journal of Econometrics, 85(1), 125-154.
• Ponomarov, S. Y., & Holcomb, M. C. (2009). Understanding the concept of supply chain resilience. International Journal of Logistics Management, 20(1).
• Powell, W. W. (1998). Learning from collaboration: Knowledge and networks in the biotechnology and pharmaceutical industries. California Management Review, 40(3), 228-240.
• Puga, D. (1999). The rise and fall of regional inequalities. European economic review, 43(2), 303-334.
• Qian, G., Li, L., Li, J., & Qian, Z. (2008). Regional diversification and firm performance. Journal of International Business Studies, 39(2), 197-214.
• Raadschelders, J. C., & Lee, K. H. (2011). Trends in the study of public administration: Empirical and qualitative observations from Public Administration Review, 2000–2009. Public Administration Review, 71(1), 19-33.
• Radovanovic, N., & Benner, M. (2019). Smart Specialisation and the Wider Innovation Policy Context in the Western Balkans (No. JRC118199). Joint Research Centre (Seville site).
• Ramezani, J., & Camarinha-Matos, L. M. (2020). Approaches for resilience and antifragility in collaborative business ecosystems. Technological Forecasting and Social Change, 151.
• Ranghieri, F., & Ishiwatari, M. (Eds.). (2014). Learning from megadisasters: lessons from the Great East Japan Earthquake. The World Bank.
• Redding, S. J. (2020). Trade and Geography. National Bureau of Economic Research Working Paper Series, (w27821).
• Redding, S. J., & Rossi-Hansberg, E. (2017). Quantitative spatial economics. Annual Review of Economics, 9, 21-58.
• Redding, S. J., Sturm, D. M., & Wolf, N. (2011). History and industry location: evidence from German airports. Review of Economics and Statistics, 93(3), 814-831.
• Richey, R. G., Natarajarathinam, M., Capar, I., & Narayanan, A. (2009). Managing supply chains in times of crisis: a review of literature and insights. International Journal of Physical Distribution & Logistics Management. 39 (7), 535-573.
• Rizzi, P., Graziano, P., & Dallara, A. (2018). A capacity approach to territorial resilience: The case of European regions. The Annals of Regional Science, 60(2), 285-328.
• Rose, A. (2007). Economic resilience to natural and man-made disasters: Multidisciplinary origins and contextual dimensions. Environmental Hazards, 7(4), 383-398.
• Rosenbaum, P. R., & Rubin, D. B. (1985). Constructing a control group using multivariate matched sampling methods that incorporate the propensity score. The American Statistician, 39(1), 33-38.
• Rugman, A. M. (1979). International diversification and the multinational enterprise. Lexington Books, Lanham.
• Rupasingha, A., Crown, D., & Pender, J. (2019). Rural business programs and business performance: The impact of the USDA's Business and Industry (B&I) Guaranteed Loan Program. Journal of Regional Science, 59(4), 701-722.
• Salazar, M., & Holbrook, A. (2007). Canadian science, technology and innovation policy: the product of regional networking? Regional Studies, 41(8), 1129-1141.
• Sawai, M. (2017). Industrial rehabilitation policies of Osaka in the late 1940s[1940年代後半期における大阪の産業復興政策]. Nanzan Management Review, 32(2), 169-189.[in Japanese]
187
• Sawai, M. (2018). From munition industry to defense industry: case in Osaka Prefecture[特需生産から防衛生産へ 大阪府の場合]. Academia: social science, (14), 41-61.[in Japanese]
• Scholten, K., Scott, P. S., & Fynes, B. (2019). Building routines for non-routine events: supply chain resilience learning mechanisms and their antecedents. Supply Chain Management: An International Journal, forthcoming.
• Sharpe, W. F. (1970). Portfolio theory and capital markets. New York: McGraw-Hill. • Shimada, G. (2016). Inside the Black Box of Japan’s Institution for Industrial Policy.
Efficiency, Finance, and Varieties of Industrial Policy: Guiding Resources, Learning, and Technology for Sustained Growth, 156.
• Shimada, H. (2011). Postwar small business policies chronology[戦後中小企業政策年表]. Nihon Tosho Center.[in Japanese]
• Shipan, C. R., & Volden, C. (2008). The mechanisms of policy diffusion. American Journal of Political Science, 52(4), 840-857.
• Simmie, J., & Martin, R. (2010). The economic resilience of regions: towards an evolutionary approach. Cambridge journal of regions, economy and society, 3(1), 27-43.
• Somers, S., & Svara, J. H. (2009). Assessing and managing environmental risk: Connecting local government management with crisis management. Public Administration Review, 69(2), 181-193.
• Sotarauta, M., & Kautonen, M. (2007). Co-evolution of the Finnish national and local innovation and science arenas: towards a dynamic understanding of multi-level governance. Regional Studies, 41(8), 1085-1098.
• Spadavecchia, A. (2005). State subsidies and the sources of company finance in Italian industrial districts, 1951–1991. Enterprise & Society, 6(4), 571-580.
• Spencer, C., & Temple, P. (2016). Standards, learning, and growth in B ritain, 1901–2009. The Economic History Review, 69(2), 627-652.
• Stuart, E. A. (2010). Matching methods for causal inference: A review and a look forward. Statistical science: a review journal of the Institute of Mathematical Statistics, 25(1), 1.
• Stuart, E. A., King, G., Imai, K., & Ho, D. (2011). MatchIt: nonparametric preprocessing for parametric causal inference. Journal of Statistical Software, 42(8).
• Stuart, E. A., Lee, B. K., & Leacy, F. P. (2013). Prognostic score–based balance measures can be a useful diagnostic for propensity score methods in comparative effectiveness research. Journal of Clinical Epidemiology, 66(8), S84-S90.
• Takebe, Y. (1982). History of Osaka Industry[大阪産業史]. Yuhikaku.[in Japanese] • Tassey, G. (2000). Standardization in technology-based markets. Research policy, 29(4-
5), 587-602. • Terry, L. D., & Stivers, C. (2002). Democratic governance in the aftermath of September
11, 2001. Public Administration Review, 62(S1), 16. • Tingvall, P. G., & Videnord, J. (2020). Regional differences in effects of publicly
sponsored R&D grants on SME performance. Small Business Economics, 54(4), 951-969. • Todo, Y., Nakajima, K., & Matous, P. (2015). How do supply chain networks affect the
resilience of firms to natural disasters? Evidence from the Great East Japan Earthquake. Journal of Regional Science, 55(2), 209-229.
• Tödtling, F., & Trippl, M. (2005). One size fits all?: towards a differentiated regional innovation policy approach. Research Policy, 34(8), 1203-1219.
• Topa, G., & Zenou, Y. (2015). Neighborhood and network effects. In Handbook of regional and urban economics (Vol. 5, pp. 561-624). Elsevier.
188
• Treado, C. D. (2010). Pittsburgh's evolving steel legacy and the steel technology cluster. Cambridge Journal of Regions, Economy and Society, 3(1), 105-120.
• Ueda, H. (2004). Subcontract Industry during Wartime Japan: SME and "Subcontract=Cooperating Policies"[戦時期日本の下請工業: 中小企業と「下請=協力工業政策」]. Minerva Shobo.[in Japanese]
• Ueno, H., Murakoso, T., & Hirai, T. (2006). Supplier system and innovation policy in Japan. In Small Firms and Innovation Policy in Japan. Routledge.
• Ughetto, E., Cowling, M., & Lee, N. (2019). Regional and spatial issues in the financing of small and medium-sized enterprises and new ventures. Regional Studies, 53(5), 617-619.
• United States Strategic Bombing Survey. (1947). Effects of Air Attack on Japanese Urban Economy. Washington, D.C.: Urban Areas Division.
• Vittas, D., & Cho, J. Y. (1999). Credit policies: lessons from East Asia. The World Bank. • Wang, C. C. (2015). Geography of knowledge sourcing, search breadth and depth patterns,
and innovative performance: a firm heterogeneity perspective. Environment and Planning A, 47(3), 744-761.
• Wang, C. Y., & Kuo, M. F. (2017). Strategic styles and organizational capability in crisis response in local government. Administration & Society, 49(6), 798-826.
• Weitzen, S., Lapane, K. L., Toledano, A. Y., Hume, A. L., & Mor, V. (2004). Principles for modeling propensity scores in medical research: a systematic literature review. Pharmacoepidemiology and drug safety, 13(12), 841-853.
• Wolman, H., Wial, H., Clair, T. S., & Hill, E. (2017). Coping with adversity: Regional economic resilience and public policy. Cornell University Press.
• Zhu, H., Deng, Y., Zhu, R., & He, X. (2016). Fear of nuclear power? Evidence from Fukushima nuclear accident and land markets in China. Regional Science and Urban Economics, 60, 139-154.
• Zhu, L., Ito, K., & Tomiura, E. (2016). Global sourcing in the wake of disaster: Evidence from the Great East Japan Earthquake (No. 89-E). RIETI Discussion Paper Series.
• Zolli, A., & Healy, A. M. (2012). Resilience: Why things bounce back. Hachette UK.
189
Appendix
190
Appendix for Chapter 3
A: Different specification of spatial weight matrix
Table 3.A.1: Estimated spatial parameter"based on alternative definitions and cut-offs of spatial weight matrix
#!"#$ = %1'!"%
()'!" < 50[km]012ℎ45#(64
#!"& = %1'!"
()'!" < ℎ[km]012ℎ45#(64
#!"'( = 71()'!" < ℎ[km]012ℎ45#(64
Notes: Plot ■ represents that estimated spatial parameter and is statistically significant at least 10% level, □ represents “not significant”. Vertical arrows represent the 95% confidence interval of the estimated spatial parameter.
191
Appendix for Chapter 4
A: Covariate balance in PSM
This appendix outlines the detailed procedures and the results of the propensity score
matching. Firstly, we explain the first stage estimation, the estimation of propensity scores
using a logit model. As also described in the main text, the independent variables (covariates)
include basic firm characteristics representing their capability in terms of R&D activities such
as sales and the number of employees in natural logarithm (lnSALES and lnEMP), and firm
age (AGE). We also include industry dummies to distinguish between low- and high-tech
industries. To improve the fitting of the logit model, we further introduce square terms of each
quantitative variable and interaction terms between each variable. To avoid imprecise and
biased prediction of propensity score (Weitzen, Lapane, Toledano, and Mor, 2004), we exclude
covariates that cause serious multicollinearity (appearing as an unrealistically large estimated
coefficient or standard error). In addition, we predict propensity score avoiding complete
separation. To obtain the PS prediction model with higher generalization performance (i.e., to
avoid the over-fitting of the prediction model), we conduct the stepwise variable selection
based on AIC. We report the estimation results for each prefecture in Tables 4.A.1.1, 4.A.1.2,
and 4.A.1.3.
In the second stage, based on the predicted propensity score, we matched a subset of
untreated firms to treated firms based on 10 nearest neighbor matching using Stuart, King, Imai,
and Ho’s (2011) algorithm. The matching results are reported in Tables 4.A1.4, 4.A1.5, and
4.A.1.6. One of the representative indices for evaluating the balance of covariates between
treatment and control group is standardized bias. We may generally judge that the covariate
balance can be achieved if the absolute value of standardized bias corresponding to each
variable is below certain thresholds such as 0.1 or 0.25 (Stuart, Lee, & Leacy, 2013). Since the
192
absolute value of standardized bias is smaller than 0.1 for most covariates in all prefectures,
there is no convincing evidence for remarkable differences between the treatment and control
groups within the limits of observable covariates. Although the literature on the matching
methods urges caution in using measures that conflate sample size and balance, such as
hypothesis tests (e.g., Imai, King, & Stuart, 2008), we may also evaluate the balance of
covariates between using the z-test. Indeed, we cannot observe any covariate that is statistically
different between the treatment and control groups31.
31 We calculate the values of mean and standardized bias based on two different definitions. One is the naïve mean and standardized bias for matched treatment and control group (used in the z-test), and the other is weighted mean and standardized bias (used in balance check based on a 0.1/0.25 threshold). The weights take 0 if units are unmatched, take 1 if units are matched treated, and take the value proportional to the number of treatment units to which it was matched if units are matched control, respectively. See Stuart et al. (2011) for more detail.
193
A.1 Estimation results of logistic regression to predict propensity scores
Table 4.A.1.1 Estimation result about Prefecture A beta z val AGE −0.045 −1.721 * lnEMP 3.400 2.747 *** lnSALES 1.865 1.556
(lnEMP)^2 −0.234 −1.579
(lnSALES)^2 −0.154 −1.833 * D_GENERAL_MACH 3.137 2.224 ** D_TEXTILE 6.119 2.892 *** D_FOOD −1.798 −0.462
D_CERAMIC 6.674 2.909 *** D_CHEMICAL 1.378 2.811 *** D_PAPER 11.790 1.840 * D_METALLIC 7.419 3.134 *** AGE×lnSALES 0.005 1.500
AGE×D_FOOD 0.018 2.383 ** AGE×D_CERAMIC 0.027 1.663 * lnEMP×D_FOOD −1.785 −1.969 ** lnEMP×D_TEXTILE −1.133 −2.024 ** lnEMP×D_CERAMIC −2.147 −2.885 *** lnEMP×D_METALLIC −2.268 −3.146 *** lnEMP×D_GENERAL_MACH −0.534 −1.604
lnSALES×D_FOOD 1.041 1.429
lnSALES×D_PAPER −1.827 −1.707 * (Intercept) −18.667 −4.622 *** PseudoR-sq 0.195 N 6066
Notes: Significance: ***1%, **5%, *10%. Dependent variable is "#$%('#()*! = 1) . '#()*! takes 1 if."#$%_"#$", 0 = 0,1,2, dummy variable indicating the duration after first adoption, takes one at some point. Covariates are selected based on the forward-backward stepwise method with AIC.
194
Table 4.A.1.2 Estimation result about Prefecture B beta z val lnSALES 6.980 1.590
AGE −0.025 −1.668 * lnEMP −6.002 −1.320
(lnSALES)^2 −1.321 −1.733 * (lnEMP)^2 −1.489 −1.475
D_ELECTRICAL_MACH 0.832 1.684 * D_GENERAL_MACH 0.135 0.152
lnSALES×lnEMP 2.743 1.616
AGE×D_GENERAL_MACH 0.032 1.751 * (Intercept) −15.836 −2.149 ** PseudoR-sq 0.151 N 1644
Notes: Significance: ***1%, **5%, *10%. Dependent variable is "#$%('#()*! = 1) . '#()*! takes 1 if."#$%_"#$", 0 = 0,1,2, dummy variable indicating the duration after first adoption, takes one at some point. Covariates are selected based on the forward-backward stepwise method with AIC.
195
Table 4.A.1.3 Estimation result about Prefecture C beta z val AGE 0.015 1.238 lnEMP 2.019 1.751 * lnSALES 0.968 1.337
D_GENERAL_MACH 4.260 3.097 *** D_FOOD −16.157 −2.442 ** D_TRANSPORTATION −6.298 −1.609
D_ELECTRICAL_MACH −0.758 −0.301
D_METALLIC −5.253 −1.381
D_MISCELLANEOUS −12.931 −1.806 * AGE×D_GENERAL_MACH −0.133 −2.510 ** AGE×D_ELECTRICAL_MACH −0.053 −1.602
lnEMP×lnSALES −0.332 −1.845 * lnEMP×D_METALLIC 1.580 1.602
lnEMP×D_ELECTRICAL_MACH 1.335 1.909 * lnEMP×D_TRANSPORTATION 2.018 2.128 ** lnSALES×D_FOOD 2.172 2.642 *** lnSALES×D_MISCELLANEOUS 1.943 2.047 ** (Intercept) −11.506 −2.745 *** PseudoR-sq 0.169 N 2624
Notes: Significance: ***1%, **5%, *10%. Dependent variable is "#$%('#()*! = 1) . '#()*! takes 1 if."#$%_"#$", 0 = 0,1,2, dummy variable indicating the duration after first adoption, takes one at some point. Covariates are selected based on the forward-backward stepwise method with AIC.
196
A.2
Cov
aria
te v
alan
ce b
efor
e/af
ter P
SM
Ta
ble
4.A
.1.4
Mea
n va
lue
of e
ach
cova
riate
bef
ore/
afte
r PSM
abo
ut P
refe
ctur
e A
Be
fore
PSM
A
fter P
SM (W
eigh
ted)
A
fter P
SM (U
nwei
ghte
d)
M
eans
Tre
at M
eans
Con
trol S
td. B
ias M
eans
Tre
at M
eans
Con
trol S
td. B
ias
Std.
Bia
s z
val
AG
E 58
.966
47
.244
0.
266
52.0
36
52.2
52
−0.0
05
0.01
1 0.
097
lnEM
P 4.
029
2.7
1.39
3.
983
3.97
8 0.
006
0.08
8 0.
769
lnSA
LES
7.13
7 5.
818
1.12
9 7.
056
7.06
−0
.004
0.
054
0.46
6 (ln
EMP)
^2
17.1
37
8.72
4 1.
176
16.7
59
16.8
34
−0.0
11
0.08
2 0.
71
(lnSA
LES)
^2
52.2
87
35.7
1.
03
51.0
34
51.2
08
−0.0
11
0.04
7 0.
411
D_G
ENER
AL_
MA
CH
0.34
1 0.
23
0.23
2 0.
357
0.35
0.
014
0.00
9 0.
075
D_C
ERA
MIC
0.
08
0.02
8 0.
188
0.08
3 0.
075
0.03
1 0.
039
0.33
8 D
_FO
OD
0.
08
0.07
2 0.
027
0.04
8 0.
052
−0.0
16
−0.0
33
−0.2
91
D_C
ERA
MIC
0.
045
0.04
8 −0
.011
0.
048
0.05
1 −0
.017
−0
.033
−0
.291
D
_CH
EMIC
AL
0.06
8 0.
019
0.19
2 0.
06
0.05
8 0.
005
0.14
1.
223
D_P
APE
R 0.
011
0.03
1 −0
.182
0.
012
0.01
3 −0
.011
−0
.019
−0
.165
D
_MET
ALL
IC
0.04
5 0.
136
−0.4
31
0.04
8 0.
043
0.02
3 0.
007
0.05
8 A
GE×
lnSA
LES
434.
479
280.
317
0.39
8 36
9.82
5 37
5.33
4 −0
.014
0.
01
0.08
8 ln
EMP×
D_G
ENER
AL_
MA
CH
1.34
0.
594
0.38
8 1.
404
1.37
1 0.
017
0.02
3 0.
197
lnEM
P×D
_TEX
TILE
0.
279
0.07
3 0.
211
0.29
2 0.
266
0.02
6 0.
039
0.33
8 A
GE×
D_F
OO
D
11.8
3 4.
465
0.14
7 3.
512
3.88
8 −0
.008
−0
.026
−0
.224
ln
EMP×
D_F
OO
D
0.33
5 0.
205
0.11
2 0.
178
0.20
3 −0
.021
−0
.041
−0
.356
ln
SALE
S×D
_FO
OD
0.
635
0.45
4 0.
082
0.35
2 0.
39
−0.0
17
−0.0
36
−0.3
1 A
GE×
D_C
ERA
MIC
3.
159
2.61
4 0.
034
3.31
2.
788
0.03
3 0.
021
0.18
7 ln
EMP×
D_C
ERA
MIC
0.
138
0.13
0.
012
0.14
4 0.
144
0.00
1 −0
.016
−0
.141
ln
SALE
S×D
_PA
PER
0.06
6 0.
181
−0.1
86
0.06
9 0.
076
−0.0
12
−0.0
19
−0.1
65
lnEM
P×D
_MET
ALL
IC
0.12
8 0.
373
−0.3
53
0.13
4 0.
129
0.00
8 −0
.007
−0
.064
N
otes
: The
met
hod
of P
SM is
10
near
est-n
eigh
bor m
atch
ing.
197
Tabl
e 4.
A.1
.5 M
ean
valu
e of
eac
h co
varia
te b
efor
e/af
ter P
SM a
bout
Pre
fect
ure
B
Befo
re P
SM
Afte
r PSM
(Wei
ghte
d)
Afte
r PSM
(Unw
eigh
ted)
Mea
ns T
reat
M
eans
Con
trol
Std.
Bia
s M
eans
Tre
at
Mea
ns C
ontro
l St
d. B
ias
Std.
Bia
s z
val
lnSA
LES
6.81
5.
869
0.79
2 6.
752
6.73
0.
018
0.11
9 0.
63
AG
E 41
.25
41.3
22
−0.0
04
40.8
06
38.2
78
0.12
3 0.
269
1.41
7 ln
EMP
3.88
1 2.
883
0.97
2 3.
831
3.78
8 0.
042
0.15
4 0.
814
(lnSA
LES)
^2
47.7
51
36.0
99
0.71
1 46
.886
46
.375
0.
031
0.13
7 0.
722
(lnEM
P)^2
16
.085
9.
666
0.81
2 15
.65
15.1
77
0.06
0.
18
0.95
3 D
_ELE
CTRI
CAL_
MA
CH
0.25
0.
15
0.22
8 0.
258
0.20
1 0.
13
0.12
6 0.
667
D_G
ENER
AL_
MA
CH
0.40
6 0.
179
0.45
6 0.
387
0.42
8 −0
.082
0.
006
0.03
ln
SALE
S×ln
EMP
27.5
48
18.2
22
0.81
26
.921
26
.371
0.
048
0.16
4 0.
868
AG
E×D
_GEN
ERA
L_M
ACH
20
.062
6.
244
0.48
1 18
.935
17
.927
0.
035
0.21
6 1.
141
Not
es: T
he m
etho
d of
PSM
is 1
0 ne
ares
t-nei
ghbo
r mat
chin
g.
198
Tabl
e 4.
A.1
.6 M
ean
valu
e of
eac
h co
varia
te b
efor
e/af
ter P
SM a
bout
Pre
fect
ure
C
Befo
re P
SM
Afte
r PSM
(Wei
ghte
d)
Afte
r PSM
(Unw
eigh
ted)
Mea
ns T
reat
M
eans
Con
trol
Std.
Bia
s M
eans
Tre
at
Mea
ns C
ontro
l St
d. B
ias
Std.
Bia
s z
val
AG
E 40
.308
41
.962
−0
.089
40
.308
41
.909
−0
.086
−0
.113
−0
.55
lnEM
P 3.
687
2.96
5 0.
715
3.68
7 3.
66
0.02
7 0.
101
0.49
ln
SALE
S 6.
894
6.11
7 0.
595
6.89
4 6.
892
0.00
2 0.
069
0.33
6 D
_GEN
ERA
L_M
ACH
0.
154
0.19
6 −0
.113
0.
154
0.16
1 −0
.018
−0
.045
−0
.22
D_F
OO
D
0.07
7 0.
133
−0.2
07
0.07
7 0.
064
0.04
6 0.
052
0.25
D
_TRA
NSP
ORT
ATI
ON
0.
154
0.07
0.
229
0.15
4 0.
17
−0.0
45
−0.0
12
−0.0
57
D_E
LECT
RICA
L_M
ACH
0.
269
0.07
0.
44
0.26
9 0.
245
0.05
3 0.
145
0.70
3 D
_MET
ALL
IC
0.11
5 0.
156
−0.1
24
0.11
5 0.
127
−0.0
35
−0.0
74
−0.3
58
D_M
ISCE
LLA
NEO
US
0.07
7 0.
094
−0.0
61
0.07
7 0.
055
0.07
9 0.
071
0.34
3 ln
EMP×
lnSA
LES
26.5
72
19.3
17
0.63
5 26
.572
26
.199
0.
033
0.11
1 0.
536
AG
E×D
_GEN
ERA
L_M
ACH
2.
885
7.96
−0
.669
2.
885
3.02
2 −0
.018
−0
.054
−0
.261
ln
SALE
S×D
_FO
OD
0.
654
0.85
6 −0
.087
0.
654
0.52
8 0.
054
0.06
9 0.
336
lnEM
P×D
_TRA
NSP
ORT
ATI
ON
0.
705
0.22
9 0.
279
0.70
5 0.
784
−0.0
46
−0.0
05
−0.0
24
AG
E×D
_ELE
CTRI
CAL_
MA
CH
7.69
2 2.
216
0.35
1 7.
692
7.79
6 −0
.007
0.
045
0.22
ln
EMP×
SD_E
LECT
RICA
L_M
ACH
1.
037
0.21
4 0.
45
1.03
7 0.
945
0.05
0.
157
0.76
ln
EMP×
D_M
ETA
LLIC
0.
461
0.46
5 −0
.004
0.
461
0.52
4 −0
.048
−0
.084
−0
.408
ln
SALE
S×D
_MIS
CELL
AN
EOU
S 0.
598
0.56
0.
018
0.59
8 0.
435
0.07
7 0.
069
0.33
5 N
otes
: The
met
hod
of P
SM is
10
near
est-n
eigh
bor m
atch
ing.
199
B: Policy effects on total factor productivity
B.1 Methodology
From a comparative perspective of diversification indices, we also employ the TFP as a
conventional outcome. We estimate TFP based on the method proposed by Levinsohn & Petrin
(2003). Their method (LP) is advantageous in the sense that it can address the endogeneity
problem in estimating TFP that emerges due to simultaneity between productivity and capital.
To estimate TFP, we utilize the financial database by TDB. With the database, we can
capture almost all items included in the Financial Statement. In 2015, the database included
213,013 firms, so the coverage is limited compared to the inter-firm transaction data. However,
this dataset is very advantageous since it covers financial information of numerous unlisted
firms, which is of the focus of our investigation. The estimation result of TFP is shown in Table
4.B.1.
B.2 Policy Effects on TFP
This section describes the estimation results on the effects of R&D subsidy on subsidized
firms’ TFP, comparing between the three prefectures. Due to the restriction of sample size, we
do not use PSM on panel data for evaluating the effect on TFP. We show sample size of panel
data in Table 4.B.2.
Table 4.B.1 Estimation result of TFP
beta SE
ln(EMP) 0.1803 0.00084
ln(CAPITAL) 0.0727 0.00073
N 3305734
Notes: Dependent variable is the logarithm of sales due to limited availability of value-added data. As independent
variables, we use the number of employees (labor input as a free variable), and the amount of tangible fixed asset
(capital input as a state variable). As a proxy variable, we use the cost of sales. Dataset utilized to estimate TFP is
unbalanced panel data from 2000 to 2015 including firms throughout Japan and industrial sectors based on the
financial database.
200
Table 4.B.2 Sample size of manufacturing SME panel data including TFP
Prefecture A B C
!!_# = 1 69 39 31
!$_% = 1
25 13
!&_' = 1
7 2
n 11010 3716 8883
Notes: !#()!_#()#, % = 0,1,2is a dummy variable that takes 1 if 3% + 1, 3% + 2, 3% + 3 years have passed
since a firm was first adopted. The rows whose name is!#()!_#()#show the number of records corresponding to
each duration.
Table 4.B.3 Estimation results of FE (Dependent variable: TFP)
Prefecture A Prefecture B Prefecture C
(1) (2) (3)
Beta t val beta t val beta t val
D_1_3 0.055 0.843
0.057 0.676
−0.029 −0.341
D_4_9
−0.047 −0.255
D_4_6
−0.140 −1.285
D_7_9
−0.082 −0.442
MULTI 0.113 2.421 ** −0.004 −0.042
INDEC −0.009 −0.351
−0.032 −0.564
ADD −0.103 −1.280 −0.168 −1.606
Firm FE YES YES YES
Year FE YES YES YES
2-digit FE YES YES YES
2-digit×Year FE YES YES YES
n 11010 6041 8883
Notes: Significance: ***1%, **5%, *10%. Standard errors are clustered at the firm level. Main treatment variable
!#()!_#()#, % = 0,1,2is a dummy variable that takes 1 if 3% + 1, 3% + 2, 3% + 3 years have passed since a
firm was first adopted. For Prefecture C, we substitute!$_%and!&_' with !$_'. in the unbalance panel data since
the number of observations with!&_' = 1can be too small to obtain reliable and representative estimation results.
We show the results of the FE models in Table 4.B.3. Whereas the coefficients of the
dummy variables for the duration after the first adoption are not statistically significant for
Prefecture A [Column (1)], we observe a positive effect of"#$%&on firms’ TFP. This result
indicates that the TFP of firms that received a multiple-year subsidy significantly increased,
yet without lagged effects. Contrarily, for Prefecture B [Column (2)], we cannot find any
statistically significant effects on subsidized firms’ TFP. For Prefecture C [Column (3)] too,
we find neither statistically significant effects on subsidized firms’ TFP. In sum, we find
statistically significant and positive effects of local subsidy on recipient firms’ TFP only in
Prefecture A.
201
C: P
olic
y ef
fect
s on
the
num
ber o
f cus
tom
ers i
n th
e G
reat
er T
okyo
Dist
rict
Fo
r the
com
paris
on w
ith O
kubo
et a
l. (2
016)
, and
as a
rela
ted
outc
ome
to IN
TRA
REG
, our
esti
mat
ion
incl
udes
the
num
ber o
f cus
tom
ers i
n
Gre
ater
Tok
yo D
istric
t, th
e la
rges
t are
a of
con
sum
ptio
n an
d bu
sines
s in
Jap
an, T
YO
CUS.
We
show
the
resu
lts o
f pol
icy
effe
ct o
n"#$%&'
in
Tabl
e 4.
C.1.
Fro
m th
e re
sults
show
n in
Col
umns
(3) a
nd (4
) we
obse
rve
the
posit
ive
and
statis
tical
ly si
gnifi
cant
effe
cts o
f sub
sidy
with
in a
t lea
st
thre
e ye
ars,
whi
ch is
con
siste
nt w
ith O
kubo
et a
l. (2
016)
. Thu
s, w
e fin
d th
at th
e nu
mbe
r of c
usto
mer
s of s
ubsid
ized
firm
s in
Toky
o M
etro
polit
an
Are
a be
cam
e sig
nific
antly
larg
er in
Pre
fect
ure
B, b
ut n
ot in
Pre
fect
ures
A a
nd C
.
T
able
4.C
.1 E
stim
atio
n re
sult
s of
FE
(D
epen
dent
var
iabl
e: T
YO
CU
S)
Pre
fect
ure
A
Pre
fect
ure
B
Pre
fect
ure
C
(1
) (2
) (3
) (4
) (5
) (6
)
be
ta
t va
l
be
ta
t va
l
be
ta
t va
l
be
ta
t va
l
be
ta
t va
l
be
ta
t va
l
D_1
_3
0.00
6 0.
045
0.
104
0.59
4
0.67
1 2.
781
***
0.51
7 1.
834
* 0.
140
0.90
5
0.17
3 0.
977
D
_4_6
−
0.15
7 −
0.83
5
0.07
4 0.
261
0.
392
1.47
8
0.53
5 2.
052
**
0.38
8 1.
384
0.
512
1.42
7
D_7
_9
0.23
3 1.
118
0.
336
1.56
3
0.19
8 0.
863
0.
409
1.54
1
MU
LT
I −
0.83
5 −
4.40
6 **
* −
0.60
3 −
2.49
6 **
−
0.16
4 −
0.62
2
−0.
329
−1.
344
IN
DE
C
−0.
013
−0.
226
0.
027
0.40
2
0.21
7 1.
076
0.
274
1.34
8
AD
D
0.18
6 1.
322
0.
093
0.47
0
−0.
598
−1.
557
0.
027
0.07
5
Fir
m F
E
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
Yea
r F
E
YE
S
YE
S
YE
S
YE
S
YE
S
YE
S
2-di
git
FE
Y
ES
Y
ES
Y
ES
Y
ES
Y
ES
Y
ES
2-
digi
t×Y
ear
FE
Y
ES
Y
ES
Y
ES
Y
ES
Y
ES
Y
ES
PS
M
NO
Y
ES
N
O
YE
S
NO
Y
ES
n 57
614
6041
32
875
3744
52
049
3120
N
otes
: S
igni
fica
nce:
***
1%, *
*5%
, *10
%. S
tand
ard
erro
rs a
re c
lust
ered
at
the
firm
lev
el. M
ain
trea
tmen
t va
riab
le !
!"#$_!"#
!,#=0,1,2
is a
dum
my
vari
able
tha
t ta
kes
1 if
3#
+1, 3#
+2, 3#
+3
year
s ha
ve p
asse
d si
nce
a fi
rm w
as f
irst
ado
pted
. The
met
hod
of P
SM
is
10 n
eare
st-n
eigh
bor
mat
chin
g.
202
D: Descriptive statistics of the panel data
Table 4.D.1 Descriptive statistics of the unbalanced panel data for examining effect on TFP
Pref Variables n Mean SD Min Max
A
TFP 11010 11.996 1.077 5.872 16.226
D_1_3 11010 0.006 0.079 0 1
CUM 11010 0.013 0.139 0 3
MULTI 11010 0 0.01 0 1
INDEC 11010 0.002 0.081 −1 1
ADD 11010 0.002 0.049 0 1
B
TFP 3716 11.728 1.072 8.045 15.328
D_1_3 3716 0.01 0.102 0 1
D_4_6 3716 0.007 0.082 0 1
D_7_9 3716 0.002 0.043 0 1
CUM 3716 0.03 0.209 0 3
MULTI 3716 0.005 0.067 0 1
INDEC 3716 0.001 0.085 −1 1
ADD 3716 0.006 0.078 0 1
C
TFP 8883 12.002 1.058 6.812 15.293
D_1_3 8883 0.003 0.059 0 1
D_4_9 8883 0.002 0.041 0 1
203
Table 4.D.2 Descriptive statistics of the unbalanced panel data for examining effect on outcomes except for TFP
Pref Variables n Mean SD Min Max
A
INTERIND 57614 1.032 0.764 0 3.865
INTERREG 57614 0.836 0.83 0 4.871
INTRAIND 57614 0.37 0.471 0 3.122
INTRAREG 57614 0.232 0.381 0 2.322
TYOCUS 57614 0.939 2.707 0 190
D_1_3 57614 0.003 0.057 0 1
D_4_6 57614 0 0.019 0 1
MULTI 57614 0 0.007 0 1
INDEC 57614 0 0.054 −1 1
ADD 57614 0.001 0.038 0 1
B
INTERIND 32875 0.966 0.755 0 3.775
INTERREG 32875 0.878 0.81 0 4.139
INTRAIND 32875 0.413 0.505 0 2.948
INTRAREG 32875 0.229 0.38 0 2.156
TYOCUS 32875 1.409 1.994 0 47
D_1_3 32875 0.003 0.051 0 1
D_4_6 32875 0.002 0.041 0 1
D_7_9 32875 0.001 0.026 0 1
MULTI 32875 0.001 0.034 0 1
INDEC 32875 0 0.043 −1 1
ADD 32875 0.002 0.044 0 1
C
INTERIND 52049 0.93 0.752 0 3.24
INTERREG 52049 0.87 0.817 0 5.365
INTRAIND 52049 0.356 0.474 0 2.948
INTRAREG 52049 0.229 0.38 0 2.45
TYOCUS 52049 1.506 2.466 0 85
D_1_3 52049 0.002 0.04 0 1
D_4_6 52049 0.001 0.028 0 1
D_7_9 52049 0 0.014 0 1
204
Table 4.D.3 Descriptive statistics of the balanced panel data for examining effect on outcomes except for TFP
Pref Variables n Mean SD Min Max
A
INTERIND 6041 1.334 0.718 0 3.865
INTERREG 6041 1.304 0.947 0 4.615
INTRAIND 6041 0.548 0.491 0 2.608
INTRAREG 6041 0.373 0.425 0 2
TYOCUS 6041 1.752 2.378 0 27
D_1_3 6041 0.027 0.162 0 1
D_4_6 6041 0.003 0.055 0 1
MULTI 6041 0.001 0.026 0 1
INDEC 6041 0.003 0.151 −1 1
ADD 6041 0.011 0.105 0 1
B
INTERIND 3744 1.347 0.724 0 3.328
INTERREG 3744 1.323 0.782 0 3.889
INTRAIND 3744 0.597 0.527 0 2.948
INTRAREG 3744 0.361 0.404 0 1.961
TYOCUS 3744 2.233 2.062 0 18
D_1_3 3744 0.02 0.14 0 1
D_4_6 3744 0.011 0.105 0 1
D_7_9 3744 0.005 0.069 0 1
MULTI 3744 0.009 0.092 0 1
INDEC 3744 0.001 0.121 −1 1
ADD 3744 0.012 0.108 0 1
C
INTERIND 3120 1.263 0.728 0 3.181
INTERREG 3120 1.223 0.848 0 4.844
INTRAIND 3120 0.578 0.541 0 2.322
INTRAREG 3120 0.357 0.425 0 2.176
TYOCUS 3120 2.227 3.597 0 53
D_1_3 3120 0.021 0.144 0 1
D_4_6 3120 0.009 0.094 0 1
D_7_9 3120 0.002 0.04 0 1
205
E: Goodman-Bacon decomposition
Recent studies (Goodman-Bacon, 2018; Callaway and Sant’Anna, 2019) point out
potential problems in the interpretation of two-way fixed effects estimators with variation in
the treatment timing. For instance, if one uses a simple two-way fixed effects model in a DID
design ignoring the difference of treatment timing, the estimated DID parameter captures not
only the before-after difference between treated and untreated observations but also that
between early and later treated observations. Goodman-Bacon (2018) developed the
decomposition method to distinguish the first difference from the second one and to examine
the contribution of each difference for the balanced panel data. He showed that the DID (with
staggered treatment timing) estimated by a simple two-way fixed effects model is a weighted
average of different treatment effects described above under the following specification,
where"!"takes one if an individual#receives treatment after period$.
%!" = '" + )! + *##"!" + +!" . (A1)
Utilizing the PS-matched balanced panel data, for each prefecture, we firstly examine the
contribution of each difference represented with the weights in Table 4.E.1. We can confirm
that it was the before-after difference between treated and untreated firms that mostly
contributed to the estimated DID parameters in our analysis. Because the weights are merely a
function of the size of the subsample, the relative size of treatment and control units, and the
timing of treatment in the subsample, the value of weights is common for any outcomes in the
same panel data for each prefecture. Also, we show the plots of the decomposition results of
treatment variables based on these weights for each outcome in Figure 4.E.1, 4.E.2, and 4.E.3.
Table 4.E.1: Contribution of each difference on a simple DID estimator
Type of difference Prefecture A Prefecture B Prefecture C
Earlier vs Later Treated 0.014 0.026 0.027
Later vs Earlier Treated 0.007 0.014 0.010
Treated vs Untreated 0.979 0.960 0.963
206
INTERIND
INTRAIND
INTERREG
INTRAREG
Figure 4.E.1 Goodman-Bacon decomposition of a simple DID estimator (Prefecture A)
207
INTERIND
INTRAIND
INTERREG
INTRAREG
Figure 4.E.2 Goodman-Bacon decomposition of a simple DID estimator (Prefecture B)
208
INTERIND
INTRAIND
INTERREG
INTRAREG
Figure 4.E.3 Goodman-Bacon decomposition of a simple DID estimator (Prefecture C)
209
Appendix for Chapter 5
A: Descriptive Statistics
Table 5.A.1: Descriptive statistics of key variables in two-wave panel data
n mean sd median min max
ln(emp) 1196 4.031 0.989 4.078 1.099 7.772
ln(sales) 1196 5.382 1.348 5.247 1.609 9.928
ln(sales per capita) 1196 1.351 1.302 1.204 −2.608 5.123
treat 1196 0.075 0.264 0.000 0.000 1.000
210
Table 5.A.2: Correlation matrix of key variables in two-wave panel data
ln(emp) ln(sales) ln(sales per capita) treat
ln(emp) 1
ln(sales) 0.412 1
ln(sales per capita) −0.332 0.722 1
Treat 0.138 −0.09 −0.197 1
211
B: Empirical Analysis Using No. of Employees as Outcome
Table 5.B.1: DD table
ln(emp)
Before After
Control 3.828 4.156
Treatment 4.218 4.798
DD
0.252
212
Table 5.B.2: DDD table
ln(emp)
Treatment Kosho Before After DD
0 0 3.806 4.125
1 0 4.327 4.867 0.221
0 1 4.019 4.43
1 1 3.882 4.586 0.292
DDD 0.071
Notes: Rows with Treat=1 show the results in the treated firms, and those with Kosho=1 show the results in the
firms in wards neighboring the arsenal in 1951.
213
Table 5.B.3: Estimation result of DDD (former arsenal)
ln(emp)
beta tval
after×kosho −0.051 −0.409
treat×after 0.166 1.613
treat×after×kosho 0.181 0.565
Treated firms 45
n 1196
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model.
Standard errors are clustered at the firm level. All models include time-sector and time-city fixed effects.
214
Table 5.B.4: Estimation result (industrial heterogeneity)
ln(emp)
beta tval
treat×after×chemical 0.024 0.12
treat×after×wood 0.308 1.09
treat×after×machinery 0.424 1.952 *
treat×after×textile 0.375 1.588
treat×after×metal −0.059 −0.178
treat×after×other 0.112 0.7
Treated firms 45
n 1196
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model.
Standard errors are clustered at the firm level. All models include time-sector and time-city fixed effects.
215
C: Propensity Score Matching
The premise of the empirical analysis with DD and DDD is that the control group of
nonborrowers can precisely represent the counterfactual case had the borrowers not received
funding. However, the precise generation of the counterfactual suitable for DD and DDD
becomes difficult if the likelihood of receiving a loan is different between the treatment and
control groups. For example, factors related to each firm’s capability, such as the stability of
business conditions, can be covariates affecting this likelihood. Additionally, the importance
and urgency of a fund may be different across industrial sectors. If whether a firm received
funding was selectively determined by these covariates, the firm characteristics between the
treatment and control groups could be naturally different. A difference between the treatment
and control groups like that described above, as a confounding factor, would bias the DD and
DDD estimates.
To tackle this problem, we utilize propensity score matching (PSM). With PSM, we can
match treated firms to a subset of untreated firms with similar attributes (a similar propensity
score, in other words). One of the advantages of PSM over ordinary least squares (OLS)
multiple regression is that we can exclude observations that do not satisfy the common support
condition. In other words, PSM allows us to implement a comparison discarding inadequate
observations such as those absolutely treated or untreated.
We illustrate the procedures for PSM. First, we predict the propensity score based on the
following specification with logistic regression.
-./0($.23$! = 1|6!) = 6′!9 + :! , (A1) The dependent variable<.23$is a dummy variable taking one if firm#was a borrower and
zero otherwise. The independent variables6!, common to all outcomes in DD and DDD, are
the logarithm of capital (million yen), industrial classification dummies to control the
difference in the likelihood between sectors, and city dummies to control that between cities.
216
We add the logarithm of annual sales in the PS prediction corresponding to ln(emp) and add
the logarithm of the number of employees in that corresponding to ln(sales) and ln(sales per
capita) as a covariate. Considering the nonlinear association between PS and the covariates and
interaction effects between the covariates, we also add the squared value of the quantitative
covariates and interaction terms between the quantitative and dummy variables. To obtain the
PS prediction model with a more generalizable performance (to avoid overfitting of the
prediction model, in other words), we carry out stepwise variable selection based on the Akaike
information criterion (AIC). Table 5.C.1 shows the PS prediction result for ln(emp), and 5.C.2
shows that for ln(sales) and ln(sales per capita).
After the prediction, we implement the matching between the treatment and control
groups. In this paper, we use 5-nearest-neighbor matching. We may generally judge that
covariate balance has been achieved if the absolute value of standardized bias corresponding
to each variable is below certain thresholds such as 0.1 or 0.25 (Stuart, Lee, & Leacy, 2013).
We show the covariate balance for ln(emp) in Table 5.C.3 and that for ln(sales) and ln(sales
per capita) in Table 5.C.4. Since the absolute value of standardized bias is smaller than 0.1 for
most covariates, there is no convincing evidence for significant differences between the
treatment and control groups within the limits of the observable covariates.
Table 5.C.5 shows the PSM-DD estimation results corresponding to the baseline results
in Section 5.5.1. The DD estimate using ln(emp) as an outcome is smaller than that without
PSM and is statistically insignificant even at the 10% level under the alternative hypothesis of
=$ ≠ 0. However, we cannot completely say that the effect on ln(emp) is insignificant because
the DD estimate is statistically significant at the 10% level under the alternative hypothesis
of=$ > 0. The DD estimate for ln(sales) is robustly significant, and the value of the estimate
is not so different from that in the regression result without PSM. Additionally, the result is not
different for ln(sales per capita). In Table 5.C.6, we show the PSM-DDD estimation results
217
corresponding to the DDD estimation focusing on the Osaka Arsenal in Section 5.5.2. Although
there is no remarkable difference from the results without PSM, the magnitude of the DDD
estimate for ln(sales) becomes marginally larger. We show the PSM-DD estimation results
corresponding to the DD examining the cross-sectoral effect heterogeneity in Section 5.5.3 in
Table 5.C.7. Although there are several differences from the results without PSM, we can
robustly observe the positive effects on the metal and machinery industry.
In addition, we conduct 5-nearest-neighbor matching with a caliper restriction, which
limits the number of standard deviations of the propensity distance measure within which to
draw control units. Following the previous literature (e.g., Austin, 2011), we match treated
units with the subset of untreated units using calipers of width equal to 0.2 of the standard
deviation of the logit of the propensity score. We show the covariate balance for ln(emp) in
Table 5.C.8 and that for ln(sales) and ln(sales per capita) in Table 5.C.9. Since the absolute
value of standardized bias is smaller than 0.1 for most covariates in PSM for ln(emp), and for
all covariates in PSM for ln(sales) and ln(sales per capita), there is no convincing evidence for
significant differences between the treatment and control groups within the limits of the
observable covariates. From Table 5.C.10, 5.C.11, and 5.C.11, we show the estimation results
of the PSM-DD(D) corresponding to the baseline in Section 5.5.1, the DDD estimation
focusing on the Osaka Arsenal in Section 5.5.2, and the DD examining the cross-sectoral effect
heterogeneity in Section 5.5.3, respectively. We can observe that the qualitative implications
of these results are consistent with those without a caliper restriction.
218
Table 5.C.1: PS prediction result for ln(emp)
beta z value
(Intercept) −6.078 −1.193
ln(sales) 2.55 1.654
ln(sales)^2 −0.392 −2.126
ln(capital) 0.054 0.186
chemical −6.128 −2.134
Wood −10.203 −1.988
Textile −4.898 −1.498
Metal −17.036 −1.96
Higashi −1.95 −2.839
Minato 2.507 2.31
other_cities −0.877 −2.419
ln(sales)×chemical 1.367 2.144
ln(sales)×wood 2.715 2.28
ln(sales)×textile 1.148 1.622
ln(capital)×metal 1.158 1.966
Pseudo R2 0.171
n 597
219
Table 5.C.2: PS prediction result for ln(sales) and ln(sales per capita)
beta z value
(Intercept) 13.765 1.718
ln(emp) 0.924 2.786
ln(capital) −1.36 −2.373
Chemical −17.107 −1.722
Wood 1.933 2.251
machinery −10.02 −0.992
Textile −4.35 −1.143
Metal −16.468 −1.588
Higashi −1.981 −2.917
Minato 2.275 2.127
other_cities −0.939 −2.581
ln(emp)×machinery −1.269 −2.412
ln(emp)×textile 1.083 1.307
ln(capital)×chemical 1.197 1.756
ln(capital)×machinery 1.094 1.488
ln(capital)×metal 1.131 1.591
Pseudo R2 0.177
n 597
220
Tab
le 5
.C.3
: Cov
aria
te b
alan
ce f
or ln
(em
p)
Bef
ore
mat
chin
g A
fter
mat
chin
g
Mea
ns T
reat
ed
Mea
ns C
ontr
ol
Std.
Bia
s M
eans
Tre
ated
M
eans
Con
trol
St
d. B
ias
ln(s
ales
) 4.
341
4.95
6 −0
.753
4.
303
4.25
1 0.
064
ln(s
ales
)^2
19.4
95
26.1
75
−0.9
66
19.1
6 18
.697
0.
067
ln(c
apita
l)
14.5
92
14.6
76
−0.1
1 14
.591
14
.538
0.
069
Che
mic
al
0.17
8 0.
223
−0.1
17
0.18
6 0.
163
0.06
Woo
d 0.
111
0.02
2 0.
281
0.07
0.
042
0.08
8
Tex
tile
0.13
3 0.
25
−0.3
39
0.14
0.
172
−0.0
95
Met
al
0.17
8 0.
199
−0.0
56
0.18
6 0.
167
0.04
8
Hig
ashi
0.
067
0.30
1 −0
.928
0.
07
0.08
4 −0
.055
Min
ato
0.04
4 0.
004
0.19
6 0.
047
0.00
9 0.
179
othe
r_ci
ties
0.35
6 0.
455
−0.2
05
0.37
2 0.
433
−0.1
25
ln(s
ales
)×ch
emic
al
0.82
4 1.
085
−0.1
43
0.86
2 0.
772
0.05
ln(s
ales
)×w
ood
0.50
4 0.
084
0.28
8 0.
288
0.17
0.
081
ln(s
ales
)×te
xtile
0.
639
1.36
4 −0
.44
0.66
8 0.
835
−0.1
01
ln(c
apita
l)×m
etal
2.
656
2.91
7 −0
.045
2.
78
2.46
0.
055
221
Tab
le 5
.C.4
: Cov
aria
te b
alan
ce f
or ln
(sal
es)
and
ln(s
ales
per
cap
ita)
Bef
ore
mat
chin
g A
fter
mat
chin
g
Mea
ns T
reat
ed
Mea
ns C
ontr
ol
Std.
Bia
s M
eans
Tre
ated
M
eans
Con
trol
St
d. B
ias
ln(e
mp)
4.
218
3.82
7 0.
514
4.17
5 4.
253
−0.1
03
ln(c
apita
l)
14.5
92
14.6
76
−0.1
1 14
.564
14
.654
−0
.117
Che
mic
al
0.17
8 0.
223
−0.1
17
0.2
0.20
5 −0
.013
Woo
d 0.
111
0.02
2 0.
281
0.05
0.
05
0
Mac
hine
ry
0.33
3 0.
199
0.28
1 0.
35
0.32
0.
063
Tex
tile
0.13
3 0.
25
−0.3
39
0.15
0.
12
0.08
7
Met
al
0.17
8 0.
199
−0.0
56
0.17
5 0.
215
−0.1
03
Hig
ashi
0.
067
0.30
1 −0
.928
0.
075
0.06
5 0.
04
Min
ato
0.04
4 0.
004
0.19
6 0
0.01
−0
.048
othe
r_ci
ties
0.35
6 0.
455
−0.2
05
0.4
0.41
−0
.021
ln(e
mp)
×mac
hine
ry
1.32
6 0.
821
0.25
8 1.
379
1.32
7 0.
027
ln(e
mp)
×tex
tile
0.61
7 0.
915
−0.1
87
0.69
4 0.
557
0.08
5
ln(c
apita
l)×c
hem
ical
2.
653
3.27
1 −0
.107
2.
985
3.05
4 −0
.012
ln(c
apita
l)×m
achi
nery
4.
806
2.92
0.
274
5.02
1 4.
666
0.05
2
ln(c
apita
l)×m
etal
2.
656
2.91
7 −0
.045
2.
598
3.18
−0
.101
222
Table 5.C.5: PSM-DD estimation results (baseline) (1) (2) (3) ln(emp) ln(sales) ln(sales per capita)
beta tval beta tval beta tval treat×after 0.156 1.306 0.439 2.423 ** 0.195 1.005
Treated firms 43 40 40 n 516 480 480
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector and time-city fixed effects. The results are based on 5-nearest-neighbor matching.
223
Table 5.C.6: PSM-DDD estimation results (former arsenal)
(1) (2) (3) ln(emp) ln(sales) ln(sales per capita)
beta tval beta tval beta tval after×kosho −0.064 −0.389 −0.284 −0.908 −0.359 −1.218 treat×after 0.072 0.675 0.165 0.979 −0.02 −0.139 treat×after×kosho 0.287 0.874 0.873 1.947 * 0.684 1.28 Treated firms 43 40 40 n 516 480 480
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector and time-city fixed effects. The results are based on 5-nearest-neighbor matching.
224
Table 5.C.7: PSM-DD estimation results (industrial heterogeneity)
(1) (2) (3)
ln(emp) ln(sales) ln(sales per capita)
beta tval beta tval beta tval treat×after×chemical 0.065 0.261 −0.322 −1.003 −0.225 −1.14 treat×after×wood 0.249 0.615 0.031 0.069 −0.525 −1.131 treat×after×machinery 0.431 2.111 ** 0.678 2.342 ** 0.219 0.613 treat×after×textile 0.158 0.479 0.239 0.793 −0.875 −3.263 ***
treat×after×metal −0.159 −0.471 1.11 2.11 ** 1.125 2.458 **
treat×after×other −0.194 −0.811 0.074 0.298 −0.101 −0.318
Treated firms 43 40 40
n 516 480 480 Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector and time-city fixed effects. The results are based on 5-nearest-neighbor matching.
225
Table 5.C.8: Covariate balance for ln(emp) with caliper restriction Before matching After matching
Means Treated Means Control Std. Bias Means Treated Means Control Std. Bias ln(sales) 4.341 4.956 −0.753 4.254 4.267 −0.016 ln(sales)^2 19.495 26.175 −0.966 18.702 18.866 −0.024 ln(capital) 14.592 14.676 −0.11 14.558 14.532 0.034 Chemical 0.178 0.223 −0.118 0.195 0.141 0.14 Wood 0.111 0.022 0.284 0.049 0.065 −0.05 Textile 0.133 0.25 −0.343 0.146 0.161 −0.043 Metal 0.178 0.199 −0.056 0.171 0.196 −0.067 Higashi 0.067 0.301 −0.938 0.073 0.088 −0.059 Minato 0.044 0.004 0.198 0.024 0.024 0 other_cities 0.356 0.455 −0.207 0.366 0.43 −0.135
226
Table 5.C.9: Covariate balance for ln(sales) and ln(sales per capita) with caliper restriction Before Matching After matching
Means Treated Means Control Std. Bias Means Treated Means Control Std. Bias ln(emp) 4.218 3.827 0.514 4.175 4.242 −0.088 ln(capital) 14.592 14.676 −0.11 14.564 14.637 −0.095 Chemical 0.178 0.223 −0.118 0.2 0.208 −0.022 Wood 0.111 0.022 0.284 0.05 0.069 −0.061 Machinery 0.333 0.199 0.284 0.35 0.308 0.09 Textile 0.133 0.25 −0.343 0.15 0.121 0.086 Metal 0.178 0.199 −0.056 0.175 0.2 −0.065 Higashi 0.067 0.301 −0.938 0.075 0.068 0.027 Minato 0.044 0.004 0.198 0 0 0 other_cities 0.356 0.455 −0.207 0.4 0.396 0.009
227
Table 5.C.10: PSM-DD estimation results (baseline) with caliper restriction (1) (2) (3) ln(emp) ln(sales) ln(sales per capita)
beta tval beta tval beta tval treat×after 0.164 1.293 0.391 2.181 ** 0.141 0.732 Treated firms 41 40 40 n 454 426 426
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector and time-city fixed effects. The results are based on 5-nearest-neighbor matching. The calipers of width equal to 0.2 of the standard deviation of the logit of the propensity score is used.
228
Table 5.C.11: PSM-DDD estimation results (former arsenal) with caliper restriction (1) (2) (3) ln(emp) ln(sales) ln(sales per capita)
beta tval beta tval beta tval after×kosho −0.06 −0.325 −0.312 −0.965 −0.362 −1.2 treat×after 0.08 0.698 0.12 0.738 −0.071 −0.505 treat×after×kosho 0.274 0.801 0.867 1.927 * 0.682 1.269 Treated firms 41 40 40 n 454 426 426
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector and time-city fixed effects. The results are based on 5-nearest-neighbor matching. The calipers of width equal to 0.2 of the standard deviation of the logit of the propensity score is used.
229
Table 5.C.12: PSM-DD estimation results (industrial heterogeneity) with caliper restriction (1) (2) (3) ln(emp) ln(sales) ln(sales per capita)
beta tval beta tval beta tval treat×after×chemical 0.109 0.411 −0.284 −0.878 −0.21 −1.063 treat×after×wood 0.269 0.453 0.12 0.184 −0.611 −0.967 treat×after×machinery 0.472 2.276 ** 0.567 1.936 * 0.115 0.323 treat×after×textile 0.117 0.33 0.287 0.971 −0.798 −2.99 *** treat×after×metal −0.237 −0.612 1.07 2.01 ** 1.138 2.507 ** treat×after×other −0.192 −0.733 0.021 0.078 −0.263 −0.772 Treated firms 41 40 40 n 454 426 426
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector and time-city fixed effects. The results are based on 5-nearest-neighbor matching. The calipers of width equal to 0.2 of the standard deviation of the logit of the propensity score is used.
230
D: Alternative Specification for the Parallel Trend
Another way to address the violation of the parallel trend assumption in DD is to include
the interaction terms between the time-invariant observables and time fixed effects as control
variables (Angrist & Pischke, 2008)32. Specifically, we employ an alternative specification
including the interaction with variables explaining each firm’s basic capability like the
logarithm of the number of employees and capital observed in the baseline period, as is the
case of PSM. Due to the possibility of a non-linear association, we also include the quadratic
terms for these control variables.
Table 5.D.1 shows the estimation results corresponding to the baseline results in Section
5.5.1. The DD estimate for ln(sales) is robustly significant, and the value of the estimate is not
so different from that in the regression result without control attributes. Additionally, the result
is not different for ln(sales per capita). In Table 5.D.2, we show DDD estimation results
corresponding to the DDD estimation focusing on the Osaka Arsenal in Section 5.5.2. Although
there is no remarkable difference from the results without control attributes, the magnitude of
the DDD estimate for ln(sales) becomes marginally smaller. We show the DD estimation
results corresponding to the DD examining the cross-sectoral effect heterogeneity in Section
5.5.3 in Table 5.D.3. Although there are several differences from the results without control
attributes, we can robustly observe the positive effects on the metal and machinery industry.
32 As an example of empirical literature employing this strategy, Duflo (2001) that evaluated the impact of a national school construction program on educational attainment and wages relied on the DD specifications that control for the interaction of time fixed effects with baseline enrollment and the intensity of a national water and sanitation program.
231
Table 5.D.1: DD estimation results with additional control (baseline) (1) (2) ln(sales) ln(sales per capita)
beta tval beta tval treat×after 0.43 2.869 *** 0.152 0.968 Treated firms 45 45 n 1194 1194
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector fixed effects, time-city fixed effects, interaction terms between time fixed effects and baseline corporate attributes (the logarithm of sales and capital), and squared interaction terms.
232
Table 5.D.2: DDD estimation results with additional control (former arsenal) (1) (2) ln(sales) ln(sales per capita)
beta tval beta tval after×kosho −0.187 −0.946 −0.284 −1.735 * treat×after 0.249 1.906 * −0.009 −0.082 treat×after×kosho 0.651 1.672 * 0.581 1.284 Treated firms 45 45 n 1194 1194
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector fixed effects, time-city fixed effects, interaction terms between time fixed effects and baseline corporate attributes (the logarithm of sales and capital), and squared interaction terms.
233
Table 5.D.3: DD estimation results with additional control (industrial heterogeneity) (1) (2) ln(sales) ln(sales per capita)
beta tval beta tval treat×after×chemical −0.004 −0.015 −0.101 −0.548 treat×after×wood 0.274 0.653 −0.213 −0.566 treat×after×machinery 0.442 1.657 * 0.075 0.254 treat×after×textile 0.525 1.883 * −0.2 −0.517 treat×after×metal 0.991 2.296 ** 0.979 2.31 ** treat×after×other 0.284 1.502 0.058 0.234 Treated firms 45 45 n 1194 1194
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector fixed effects, time-city fixed effects, interaction terms between time fixed effects and baseline corporate attributes (the logarithm of sales and capital), and squared interaction terms.
234
E: Effect on Standardization
In this section, we examine whether the modernization fund was associated with the
standardization of the recipients’ products33. To examine this association, we focus on the
acquisition of Japanese Industrial Standards (JIS) as an outcome. JIS is a Japanese certification
scheme established based on Industrial Standardization Act in 1949, which aims to verify that
products satisfy the prescribed standards and that quality control is adequately managed. In this
regard, the certification with JIS might have been an outcome to capture one of the
achievements brought by the modernization of firms’ production system. In addition, by the
beginning of the 1960s, the certification with JIS became essential to participate in the market
competition for firms in certain industrial sectors (Agency of Industrial Science and
Technology; AIST, 1964).
The source of data about the JIS certification of firms is AIST (1964). The dataset
summarizes the information about Japanese firms certified by 1964 (name, address, and date
of certification) by product category. We manually match this database with the dataset used
in our main text. We define the outcome as a dummy variable taking 1 if a firm was certified
with JIS in the period, D_JIS, and estimate the DD(D) specified in Equations (1)-(3). One
limitation of using JIS certification as the outcome is the incompleteness of the scheme in the
1950s. Specifically, the original scheme designed in 1949 covered only specific products and
thus did not completely cover all the industrial products that existed then. Although the
coverage was extended in the 1950s, the progress was quite gradual.
Table 5.E.1 shows the estimation results corresponding to the baseline results in Section
5.5.1 based on the before-after comparison between 1951 and 1957 (Column (1)), 1951 and
33 Tassey (2000) and Blind (2019), for example, comprehensively summarized the relationship between innovation and standardization. As a historical analysis, Spencer & Temple (2016) showed a positive association between the number of standards and country-level labor productivity utilizing a century data in the UK.
235
1960 (Column (2)), and 1951 and 1964 (Column 3), respectively. Unlike the result of ln(sales),
we cannot observe a statistically significant DD estimator in Column (1). Although the DD
estimator in Column (2) remains insignificant even at the 10% level under the alternative
hypothesis of !! ≠ 0, we cannot completely say that the effect on D_JIS is insignificant
because the DD estimate is statistically significant at the 10% level under the alternative
hypothesis of!! > 0. In Column (3), we can observe the positive association statistically
significant at the 10% level. These results imply the lagged positive effect of the modernization
fund on the improvement of product quality.
In Table 5.E.2, we show DDD estimation results corresponding to the DDD estimation
focusing on the Osaka Arsenal in Section 5.5.2. In Column (2) and (3), we can observe strongly
positive DDD estimators. As with the baseline results, the convergence of the estimators can
be observed. These results imply that the improvement of product quality progressed larger
among the recipients geographically close to the former arsenal, which might explain most
parts of the positive association observed in the baseline results. The DD estimation results
corresponding to the DD examining the cross-sectoral effect heterogeneity in Section 5.5.3 are
shown in Table 5.E.3. From Column (2) and (3), we can observe positive and significant DD
estimators only about machinery industry, and there is little difference in the value of the
estimators between Column (2) and (3). In summary, the effect of the modernization fund on
the recipients’ improvement of product quality measured with the certification was lagged and
geographically and industrially limited. However, we should be cautious about the
interpretation of these results. That is, these results might merely be due to the circumstance
that the products produced by the recipients were not sufficient covered in JIS in 1957
considering the incompleteness of the JIS scheme in the 1950s, which can be reflected in the
convergence of the value of DD estimator.
236
Table 5.E.1: Estimation results of DD (baseline): JIS certification (1) (2) (3) D_JIS (1951-1957) D_JIS (1951-1960) D_JIS (1951-1964)
beta tval beta tval beta tval treat×after 0.066 1.114 0.112 1.557 0.131 1.75 * Treated firms 45 45 45 n 1196 1196 1196
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector and time-city fixed effects.
237
Table 5.E.2: Estimation results of DDD (former arsenal): JIS certification (1) (2) (3) D_JIS(1951-1957) D_JIS(1951-1960) D_JIS(1951-1964)
beta tval beta tval beta tval after×kosho 0.129 2.049 ** 0.084 1.166 0.109 1.446 treat×after 0.028 0.438 −0.016 −0.248 0.019 0.252 treat×after×kosho 0.139 0.913 0.462 2.637 *** 0.407 2.257 ** Treated firms 45 45 45 n 1196 1196 1196
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector and time-city fixed effects.
238
Table 5.E.3: Estimation results of DD (industrial heterogeneity): JIS certification (1) (2) (3) D_JIS(1951-1957) D_JIS(1951-1960) D_JIS(1951-1964)
beta tval beta tval beta tval treat×after×chemical 0.165 1.015 0.129 0.761 0.119 0.686 treat×after×wood 0.014 0.066 −0.048 −0.243 −0.043 −0.216 treat×after×machinery 0.103 0.886 0.275 2.007 ** 0.263 1.943 * treat×after×textile 0.01 1.742 * −0.044 −0.65 −0.045 −0.669 treat×after×metal 0.029 0.262 0.048 0.289 0.189 0.929 treat×after×other −0.115 −2.256 ** −0.127 −1.905 * −0.13 −1.846 * Treated firms 45 45 45 n 1196 1196 1196
Notes: Statistical significance at *** 1%, ** 5%, and * 10%. Estimation results of a two-way fixed effects model. Standard errors are clustered at the firm level. All models include time-sector and time-city fixed effects.
239
F: Matching between the IDBC and Borrower List
Table 5.F.1 shows the result of matching the IDBC and the borrower list from Osaka
Prefecture (1954). We attempt to match 326 borrowers with capital of 1 million yen or more.
The borrowers included in the panel data used in our analysis were limited to 45 firms
whose attributes used in generating the outcomes were observed in both 1951 and 1957 (we
exclude one borrower whose number of employees was more than 300). In terms of survival
bias, it might be better to check the difference in the attributes between the successfully
matched borrower group and the unmatched group. For example, we compare the attributes of
matched borrowers and unmatched borrowers (whose attributes were observed in 1951 but not
in 1957). Table 5.F.2 shows the difference in key variables between the matched (46) and
unmatched (31) firms in 1951. For any variable, we cannot necessarily find a statistically
significant difference on average. However, due to data limitations, we cannot check the
difference in the attributes of matched borrowers and other kinds of unmatched borrowers
(whose attributes could be observed neither in 1951 nor 1957 or whose attributes were observed
in 1957 only).
240
Table 5.F.1: Matching result based on 1951 IDBC Matched 102
No. of employees, sales & sector are identified
Yes No
77
25 No. of employees & sales were also observed in 1957
Yes No
46 31
Unmatched 224
241
Table 5.F.2: Difference in key attributes between the matched (46) and unmatched (31) firms Unobserved Observed SD z val
ln (sales) 4.324 4.382 0.2 0.287
ln (emp) 4.425 4.263 0.182 −0.891
ln (sales per capita) −0.101 0.119 0.177 1.243
242
Doctoral Thesis
Empirical analysis toward resilient and adaptive local economy: Evidence from Japan
Doctoral Program in Policy and Planning Sciences 201830120
Keisuke Takano
Supervisor Professor Morito Tsutsumi
Vice-supervisors
Professor Nobuyuki Harada & Professor Mitsuru Ota
September 2021
