Natural disasters and regional industrial production efficiency: evidence from pre-war Japan

ABSTRACT In this paper we investigate whether destruction due to natural disasters induces industries to increase their regional production efficiency using the case of pre-war Japan, a setting of frequent disasters and technological upgrading. To this end we compile a regional sectoral dataset of natural disaster destruction and production for machinery and textiles. We then employ a stochastic frontier analysis (SFA) approach to estimate the role of disaster events on changes in production efficiency. Our results show that earthquakes led to increases in efficiency for both machinery and textiles, although they were substantially greater for textiles due to the recovery persisting longer. Overall earthquakes contributed 6.8% of efficiency gains in textiles and 3.1% in machinery. However, allowing events to compound in their impact showed that such gains were dampened when there were damaging earthquakes in subsequent years. In contrast, for climate-related natural disaster events there is only weak, if any, evidence that these played a significant role in determining productive efficiency.


INTRODUCTION
Natural disasters can cause tremendous destruction in terms of both human and physical assets, and this can have considerable implications for the affected economies.While a priori the immediate direct effect on economies will be clearly negative, the direction of more mediumor long-term effects in terms of growth is unclear (Botzen et al., 2019).The underlying intuition in terms of a negative impact in the medium or long-term is straightforward: the destruction of human and/or physical capital, if sufficiently large, will lead to pushing economies downward from their growth path. 1 In contrast, a positive effect is feasible if the loss of capital results in the replacement of more productive and modern technologies or production methods that will push the economy to a newer, more efficient and higher growth path.This is known as the 'build back better' hypothesis. 2 Related to this, there could also be 'creative destruction' in the Schumpeterian sense, in that less efficient firms damaged by natural disasters are driven out of the market.
Since the seminal study by Albala-Bertrand (1993), numerous researchers have set out to quantify the economic impact of natural disasters, where a review of the results can be found in Cavallo and Noy (2011), Klomp and Valckx (2014), van Bergeijk and Lazzaroni (2015), Noy (2018) and Botzen et al. (2019).While these reviews emphasize the large differences in data and methods across existing studies, they also seem to reach the consensus that the literature points to a negative rather than a positive impact of natural disasters on economic wealth and/or growth, thus suggesting that natural disasters do not lead to any increase in production efficiency in response to the destroyed human or physical capital. 3However, it should be noted that there are two issues in most of the existing studies that may result in not being able to empirically identify a possible technological upgrading.First, most current papers look at the macroeconomic rather than regional effects of natural disasters.Because most natural disasters are very localized by nature, at least in terms of their damage, aggregate data may not be able to capture any technological upgrading that may occur.Second, almost all previous studies draw conclusions regarding an increase in efficiency through its translation into higher growth or income wealth, rather than examining changes in efficiency or technology directly. 4 In this paper we re-examine the issue of whether the destruction due to natural disasters can lead regional economies to become more productive by investigating how such events have affected regional production efficiency during the 1920s in Japan for two main industries at the time, namely textiles and machinery. 5This period provides a particularly suitable context within which to study this question in that (1) there was considerable development through technological progress, (2) there were many natural disasters and (3) there were only limited disaster relief policies in place and these did not specifically encourage infrastructural and technological upgrading, so that any relationship between the former two aspects is unlikely to be confounded by the latter.However, arguably not only is Japan in the 1920s an interesting historical case study of the potential role of natural disasters in leading to greater production efficiency, but the setting also resembles that of many developing regions and countries in the modern period in that these tend to be the most hit by natural disasters (Zorn, 2018), are industrializing (Haraguchi et al., 2019), and have little disaster management policies in place (Surianto et al., 2019).Our paper may thus serve to provide some insight into what role natural disasters may play in regional economies in the developing world regarding becoming more or less production efficient after being damaged by such negative shocks.This may be particularly important given that predictions regarding climate change suggest that the number of extreme events is likely to increase and that developing countries may be those most affected (Kharin et al., 2018).
The analysis in this paper crucially rests on two previously unused data sources.Firstly we construct regional level data on outputs and inputs in the textile and machinery industries from 1919 to 1928 for Japan from historical official statistical sources.Additionally, we also build a database on the number of deaths from earthquakes and climatic disasters in each region from historical documents.Using the industry input and output panel dataset, we first estimate technical inefficiency for regional textile and machinery production using a stochastic frontier analysis (SFA). 6We then combine our measures of regional inefficiency with the natural disaster data to investigate whether natural disasters during this period can explain changes in the efficiency of these regional industries.We additionally explore whether compounding events may play a role in our findings. 7 The remainder of the paper is organized as follows.Section 2 outlines the theoretical background.Section 3 describes our datasets.Section 4 provides a demonstrative example.Section 5 shows our methodology.Our econometric analysis are presented in Section 6.A discussion and concluding remarks are in Sections 7 and 8.

THEORETICAL BACKGROUND
2.1.Regional resilience to negative shocks Regional economic resilience has been often discussed in the current literature on regional studies (e.g., Simmie & Martin, 2010;Martin, 2012;Boschma, 2015;Martin et al., 2016;Eraydin, 2016;Bailey & Turok, 2016), where resilience can be described as a region's ability to respond to as well as a region's adaptive capability after a large negative shock (e.g., economic shocks and natural disasters).The notion of the ability to respond, as is often emphasized by many economic geographers, refers to the ability to recover from downturns by negative shocks.They presume the return to pre-existing stable equilibrium state after the shock (Fingleton et al., 2012).The ability largely depends on a region's vulnerabilities, which are the components that dampen the capability to respond and deter recovery.Vulnerabilities are crucial in understanding how large a downturn after a negative shock is and how long it takes for recovery.Previous studies have found several sources of vulnerability in this regard.For instance, specialization of specific industries within the region tends to lead to be fragile to sectorspecific idiosyncratic shocks (Feyrer et al., 2007), while geographical distance, openness of trade, large reliance on exports and imports are also likely to constitute vulnerabilities (Pickles & Smith, 2011).On the other hand, diversification of industries within the region, input-output linkage with other sectors, inter-sectoral learning across regions, and transportation linkage with other regions could diversify risk and accommodate sector specific shocks (Frenken et al., 2007;Nyström, 2018).They could reduce vulnerability and foster resilience.
In contrast, adaptive capability, as proposed by the literature of evolutionary economic geography, is responsible for the long-term evolution of regions, where adaptability is defined as the ability to restructure industries by adopting new technology as a new equilibrium state (Martin & Sunley, 2006;Simmie & Martin, 2010).For instance, after the destruction caused by a negative shock, new industries and firms are replaced, and new technology is adopted in existing industries.This process creates new jobs and new varieties of goods in the regional economy.Martin (2012) proposes four dimensions of regional resilience, namely resistance (the degree of sensitivity or depth of reaction to a shock), recovery (the speed and degree of recovery from a shock), re-orientation (the extent of re-orientation and adaption of regional economy in response to a shock), and renewal (the extent to which the regional economy renews its growth path: resumption of pre-existing path or hysteretic shift to new growth trend).Many studies have confirmed that the resilience of regions depends on technological environment, innovation system, and their industrial history.They find several factors which enable regions to adjust over time, such as a strong regional innovation system (Clark et al., 2010) and high-tech clusters (Bramwell et al., 2008), traditional cluster formations (Treado & Giarratani, 2008), good public and private capital infrastructure (Hill et al., 2012), a mass of skilled, innovative and entrepreneurial people (Christopherson, 2010) and knowledge networks (Crespo et al., 2014).All have led regions to develop new technology and adjust to new growth path.In addition, governance and institutional factors are also important (Briguglio et al. 2006;Davies, 2011).For Natural disasters and regional industrial production efficiency: evidence from pre-war Japan 2055

REGIONAL STUDIES
instance, Foster (2007) claims that the ability of governance for effective strategies on recovery from recessions is also important.So far there are many studies on resilience in negative shocks (mainly on economic recessions).However, as far as we know, there are no studies on natural disasters using historical data, that is, Japan in the 1920s.Thus our contribution is to uncover the regional resilience by natural disasters in historical perspective.This allows us to discuss resilience by real exogenous shocks in the long-run period with primitive pre-disaster policies and primitive technological environments.

Japan in the 1920s
Japan in the 1920s experienced an exceptionally high number of large-scale natural disasters.For instance, on 1 September 1923, an earthquake of magnitude 7.9 hit the Greater Tokyo area (the Great Kanto Earthquake).The total number of dead and missing was estimated to be over 100,000 and the destruction of assets is estimated to be over 35% of gross national product (GNP) (Imaizumi et al., 2016).In fact, the Great Kanto Earthquake caused the largest amount of earthquake damage in modern Japanese history.Other large earthquakes hit Hyogo prefecture in 1925 (430 dead and 1300 collapsed buildings) and in 1927 (3000 dead and 12,500 collapsed buildings), and Shizuoka prefecture in 1930 (420 dead and 1300 collapsed buildings).Turning to storms, large-scale typhoons hit Tokyo in 1917 (1300 dead and 43,000 collapsed buildings) and hit many regions in 1921 (700 dead and 7400 collapsed buildings), and in 1927 (440 dead and 2200 collapsed buildings).In terms of whether these events may have encouraged greater economic growth and technological upgrading, there is already some evidence.For instance, Imaizumi et al. (2016) reported that after the 1923 Great Kanto Earthquake there was a recovery process of manufacturing output growth in Tokyo City, which lasted around six to seven years.From a sample of manufacturing firms in Yokohama City after the 1923 Great Kanto Earthquake, Okazaki et al. (2019) found a substantial upgrading of machine technology and an increased survival of efficient firms.
As noted above, sources for vulnerabilities involve international trade and industrial specialization, and these aspects of resilience are also relevant for Japan in the 1920s More specifically, Japan had opened foreign trade since the 1850s and became highly specialized in a number of specific products.For example, textile products became one of the major export products at the time (Okubo, 2007).In the 1920s industrial specialization further increased at the regional level (Fukao & Settsu, 2017), thus likely increasing vulnerabilities and reducing resilience.On the other hand, the ability to respond to regional shocks was fortified by some other factors.In this regard, Japan already developed railways in the 1890s (Minami, 1965), which made a tight geographical connection among regions across Japan.This could moderate vulnerabilities.In the 1860s and 1870s, Japan introduced modern political, legal, military, and educational and banking systems from Europe in the Meiji Restoration.Then Japan established a central government system under constitutional monarchy and implemented several important new policies for the economy, industrial development and innovation.In this regard, the ability to respond is generally strongly related to disaster relief policies, and indeed Japan had several relevant disaster relief policies in place during the period of study.More specifically, with the law of 1899 each prefectural government was required to make disaster relief fund for food, clothes, barracks and medical treatments for injured and displaced persons after natural disasters.But, regarding recovery costs and technology, the government never directly subsidized specific firms and industries for technological and infrastructural upgrade after a natural disaster.Rather, local governments financed the recovery costs on natural disaster damage using tax incentives.To reduce this tax burden in damaged prefectures, the law 'Subsidies for prefectural spending on damaged infrastructure' was enacted in 1911 and the central government gave subsidies to local governments at a certain level of recovery costs in damaged infrastructure.In this respect the policy can be interpreted as fortifying the ability to respond to the shock and boost regional resilience.
Turning to adaptive capability, technological progress during this period is also worthwhile to mention.As noted above, the adaptive capability largely depends on technological development, the environment for technology access, and long-term industrial history.With regard to development through technological progress, Japan introduced modern political, legal, military, and educational and banking systems, as well as imported advanced technology from Europe in the 1860s and 1870s.In the 1880s industrial technology came as a result of the Industrial Revolution, while in the early 1900s some Zaibatsu groups formed industrial and financial conglomerates.They introduced mass manufacturing production with advanced technology and dominated the domestic market and exports in the 1910s and 1920s.One may want to note that these periods of technological upgrading coincided also with the second industrial revolutions of Germany and the United States.After the First World War, the Japanese economy managed to upgrade its industrial structure and technology in spite of harsh international competition and a prolonged depression.Previous studies highlight three main structural changes in this regard.First, this period saw substantial technological progress, which had led to the dramatic growth of heavy manufacturing industries and developed new industries. 8Furthermore, labour productivity increased in manufacturing sectors, and industries such as textiles, mining, construction, and heavy manufacturing (chemical and machinery) doubled their productivity in this period.Second, urbanization increased substantially with over 20% of the population living in urban areas by 1925 (Abe et al., 2017).Associated with technology progress, manufacturing and service industries also concentrated geographically and this high degree of concentration fostered agglomeration economies and increased income in Preeya Mohan et al.

REGIONAL STUDIES
cities (Fukao & Settsu, 2017).Finally, the Japanese economy experienced a motive power evolution from the steam engine to the electric/gas motor during this period (Minami, 1976), 9 so that by the 1920s, most small and medium-size factories and traditional sectors shifted to machines with electric motors, 10 which boosted their technology growth (Fukao & Settsu, 2017;Abe et al., 2017).All these aspects on technological environments may have led to regional economic resilience in Japan in response to negative shocks.Finally, it is notable that technological development and industrial history varied widely across sectors.Considering specifically technological development in the two sectors of focus here, textiles and machinery, one needs to point out that these were also the ones that experienced the most technological upgrading.In the cotton textile industry, one of the main export sectors, many factories in the long recession period after the First World War adopted new technologies and new machines and some of them established affiliates in China (Abe et al., 2017).Their technology subsequently matured and they were able to compete with UK firms in the international market.In the end, although labour productivity of the cotton textile sector was stagnant in the 1920s, it increased substantially in the 1930s (Nishikawa, 1987).In the machinery sector, foreign machine companies with advanced technology directed foreign direct investment to Japan and/or established joint ventures with Japanese domestic firms, which contributed greatly to technological progress (Udagawa, 1987;Fukao & Settsu, 2017).For example, in the mid-1920s, Ford and GM built new factories in Tokyo and Osaka and produced automobiles as new products using imported parts and components from the USA.Additionally, Tokyo Electric Co. (Tokyo Denki) and Nippon Electric Co. (Nippon Denki) entered into capital and technological cooperation with General Electric Co. and Western Electric Co., respectively, making electrical products such as electric bulbs and new products such as radios and telephones in response to the surge in demand for these new machines.Furthermore, a number of patents were registered (Nishimura, 2003).As mentioned in the theoretical background, all this evidence indicates that the adoption of advanced technology in foreign countries should have fostered Japan's regional resistance to natural disasters.

Industry data
Our unit of analysis are Japan's 47 prefectures, which are the highest administrative geographical units.Importantly, the spatial delineation of prefectures has not changed since 1890.For a map of the prefectures, see Figure A1 in Appendix A in the supplemental data online.
Our data for the machinery and textile industries at the prefecture level are taken from the annual Census of Manufacture, from the Ministry of Agriculture and Commerce.The data cover all manufacturing plants with more than five employees.The data include output, the number of employees, the number of factories and the total horsepower of employed machines.All manufacturing factory owners with more than five employees were legally mandated to answer the census under the government regulation.Thus, the coverage of manufacturing production is fairly high, although factories with fewer than four employees are not covered in the census. 11Overall, we were able to compile data for the two industries from 1919 to 1928. 12We note that horsepower data are only available at the sector-prefecture level for this period.Output data are deflated to 1920 yen values.
We provide summary statistics for the textile and machinery sectors in Tables A1 and A2 in Appendix A in the supplemental data online, respectively.Table A1 shows that the average total output in textiles is about 58 million yen, but with considerable variation across prefectures; the largest producer is Nagasaki prefecture, employing 128,361 workers on average, and the smallest is Yamanashi prefecture with a workforce of 53.On average, each prefecture has 708 factories and annually uses 5790 horsepower (HP) to produce their textile products.Table A2 shows that the total output value and number of persons employed in the machinery sector is about one-quarter that in the textile sector.However, the relative ratio of factories is only about one-third, and horsepower a little more than one-half in the machinery sector relative to the textile sector.This attests to the fact that production in the machinery sector is, unsurprisingly, relatively more capital-intensive than that in the textile sector.

Natural disaster data
We build an earthquake damage data index based on the number of persons killed per prefecture.To do so, we used Nichigai Associates (1993Associates ( , 2010)).Based on the information in these data sources, we additionally consulted newspaper articles for more detailed information.This allows us to cover all earthquakes with at least one person killed and thus construct the total number of persons killed per earthquake in each prefecture.The number of deaths was originally recorded by the police in each local district and reported to prefectural and then central governments.Our data sources are based on the government formal information.Using population data from the Population Census (Ministry of Internal Affairs), we then constructed an annual regional measure of the number of deaths per capita (where population is measured in thousands).
We also constructed a parallel index capturing deaths due to climatic disasters from the Nihon Teikoku Tokei Nenkan (Imperial Japan Statistical Yearbook), the annual publication by the Cabinet Office.Same as in earthquake damage, the number of deaths was originally recorded by the police in each local district and reported to prefectural and then central governments.Although climatic events are generally classified in these sources as being either due to (1) floods and water, (2) high tides, (3) typhoons or (4) heavy rain and storms, it was not always clear whether those due to (1), ( 2) or (4) were not simply also Natural disasters and regional industrial production efficiency: evidence from pre-war Japan 2057 part of a typhoon striking or some separate event.We thus simply combined all these together as a 'climatic' disaster group.
We constructed the earthquake and climate event indices for the period from 1914 to 1928.Summary statistics of the number of deaths per 1000 people by prefecture are shown in EQ and CLI in Tables A1 and A2 in Appendix A in the supplemental data online.As can be seen, annually the number of deaths due to climate-related disasters (CLI) is much smaller than that due to earthquakes (EQ), that is, about 5%.At face value, one would expect about five deaths per million for climatic events as opposed to 104 for earthquakes in each prefecture each year.The largest earthquake event (the Great Kanto Earthquake) killed up to 17,500 per million in one prefecture (Tokyo prefecture).
We also depict the distribution of the average per capita deaths for earthquakes and climatic events in Figures 1 and 2, respectively (see also Figure A1 in Appendix A in the supplemental data online).We note that the maps use average values over our sample period for each prefecture.In parallel to the previous discussion, the most serious earthquake was the Great Kanto Earthquake of 1923, which damaged Tokyo, Kanagawa, Chiba, Saitama, Yamanashi, Shizuoka and Yamanashi prefectures (see Table A3 in Appendix A in the supplemental data online).There were also some other earthquakes, mainly in West Japan, during this period, although only 11 prefectures were damaged.In contrast, the distribution of deaths by climatic events is much more evenly distributed across prefectures.In particular, prefectures in West and Central Japan facing the Pacific Ocean had serious damage, such as, for example, Kochi and Kumamoto prefectures.
We note that our paper uses the number of deaths per 1000 people rather than physical damage due to their high correlation, and the lack of comprehensive data on the latter.As mentioned below in the data section, our data source on the number of deaths is from the government reports based on the investigation by the Police at local district.As shown in some previous studies, the number of deaths is positively correlated with physical damage.For instance, according to Imaizumi et al. (2016) and Okazaki et al. (2019Okazaki et al. ( , 2022)), the Great Kanto Earthquake of 1923 saw high positive correlations between the number of deaths and the damaged buildings both at prefectural and city levels.Turning to typhoons, Okubo and Strobl (2021), which investigated Ise bay typhoon of Nagoya city in 1959, the largest typhoon in the Japanese history, showed a high correlation between building damage and human damage.One may want to also note that the Ise Bay typhoon was a milestone for disaster prevention policy.More specifically, today's disaster-risk reduction legislation was established by the central government in response to the extensive damaged suffered after Ise bay typhoon.This arguably has reduced the number of deaths by typhoon since then, probably undermining the likely high correlation between physical damage and human losses observed generally observed beforehand.
In terms of how disaster resilient buildings at the time were, one should note that brick buildings, introduced from Europe, were widespread after the big fire in central Tokyo City in 1872 and thought as earthquake-proof as well as fireproof (Kayanoki & Ito, 2008). 13However, the Natural disasters and regional industrial production efficiency: evidence from pre-war Japan 2059

REGIONAL STUDIES
Great Kanto Earthquake of 1923 destroyed many brick buildings or damaged them by fire in the aftermath of the earthquake.After the earthquake, reinforced cement buildings were adapted to be more earthquake resilient.

DEMONSTRATIVE EXAMPLES
To provide a greater understanding of the context of our study in terms of the impact of earthquakes on the textile and machinery sectors and its possible link to increases in productive efficiency at the time, we use anecdotal evidence on specific a typhoon and earthquake event during this period.

Typhoons
On 1 October 1917, a large typhoon struck Tokyo.The storm was one of the largest typhoons that Tokyo had experienced, with a minimum pressure of 952.7 hPa, the lowest pressure on record to hit Metropolitan Tokyo, and with a maximum wind speed of 43 m/s.The typhoon first made landfall in Shizuoka prefecture, then hit the Bay of Tokyo, Metropolitan Tokyo and went through Tohoku area.In total, more than 1324 people died or were missing and 2022 people were injured.Around 43,000 buildings collapsed.The damage extended to 29 prefectures over Japan.In Metropolitan Tokyo, the resulting storm surge at high tide flooded manufacturing and commercial areas in the Bay of Tokyo, such as Kyobashi, Fukagawa and Honjo wards.A total of 563 people died in Tokyo Metropolitan City.Artificial islands in the Bay of Tokyo, Tsukishima and Tsukuda-jima, were completely submerged, while Tsuki-shima island was isolated due to the collapse of the Aioi bridge which connected it to the mainland. 14 Many people drowned and many buildings and factories were flooded.Electricity and gas supply completely stopped in Honjo, Fukagawa, and Asakusa wards.Shinbashi-Shiodome station was a major terminal station for freight and cargo in Tokyo.In the large terminal station, more than 5000 tons of daily food and products (sake, vinegar, cotton, textile products, paper and pottery) were lost due to the flood. 15Since Tokyo was a manufacturing cluster area at the time, many manufacturing factories were seriously damaged.For example, a plant in Kyokuto Glass Co. in Kameido collapsed and then immediately ignited, while Nagasaki Celluloid Co. was also seriously damaged by the flood.Also, Mikado Match Factory in Fukagawa Ward was destroyed by the fire in the aftermath of typhoon. 16 After the typhoon, the government and many people realized that pre-disaster policies should be implemented in Tokyo City, as it was the seat of modernized manufacturing and commerce in Japan.For example, a leading article in Tokyo Asahi newspaper (2 October 1917) vehemently argued that one should think of taking largescale measures against natural disasters.Thus implementing pre-disaster policies became much more important to protect the national economy.For instance, the government needed to build an embankment with high technology and implement advanced telecommunication system in the event of a natural disaster. 17However, Inajiro Tajiri, Mayor of Tokyo City, was reluctant to spend pre-disaster investment due to financial problems.Rather the construction of railway and tram networks in Tokyo City was prioritized to deal with the high speed of population growth and urbanization of Tokyo.This, however, would have also needed a large amount of public expenditure, but the issuing of public bonds or making a public bank would have further created financial difficulties for the city.Thus, it was difficult to spend large-scale pre-disaster investments by the local government. 18

Earthquakes
Six years have passed.The Great Kanto Earthquake occurred on 1 September 1923 with a magnitude of 7.9.It is considered the most serious natural disaster that Japan has ever experienced.The total number of dead and injured was estimated at over 100,000.Table A3 in Appendix A in the supplemental data online shows that Tokyo and Kanagawa prefectures in particular experienced serious damage, where around 40% of buildings completely collapsed and 2% of the prefectural population was dead or missing.The destruction of Japan's capital counted for around 35% of GNP (Imaizumi et al., 2016).Moreover, the damage extended to several neighbouring prefectures.Just after the earthquake, the government established Teito Hukkou-in (Imperial Capital Reconstruction Department), which undertook new urban planning and large-scale infrastructure projects for the cities of Tokyo and Yokohama.By 1930, all planned recovery projects were completed and a celebration ceremony was held in March of that year in Tokyo's Imperial Palace Park (Tokyo City Government, 1932).Overall, recovery took several years.
Figure 3 shows the total output and number of employees in the manufacturing sector in Tokyo prefecture.Although there was clearly a sharp decline in the number of employees and output immediately after the earthquake, recovery was quite fast.The economy sharply turned down again by the Showa Depression of 1930.In parallel to these stylized facts, Imaizumi et al. (2016) econometrically estimated the impact of the recovery process on industrial workers in Tokyo prefecture (15 wards and five counties).It was found that while the earthquake caused mean downward shifts in the shares and numbers of industrial workers, these shifts disappeared by the early 1930s.Thus, Imaizumi et al. (2016) concluded that the recovery took around six to seven years.
Next, we decompose our data by manufacturing sector. Figure 4 plots output by sector in Tokyo prefecture.Output in the machinery sector drastically declined after the earthquake but recovered immediately.In 1925, the output level already exceeded the pre-earthquake level in 1922.According to Imaizumi (2014), the reasons for the quick recovery and growth of the machinery sector involved not only the demand for industrial recovery but also the large demand for new products such as automobiles and electronic devices.In contrast, the textile sector was stagnant after the earthquake.There are several reasons for this.First, textile sectors were labour intensive and employed a large number of young female workers under poor labour conditions in spite of there being more stringent labour regulations by enforcing the law of factory (Nakamura & Molteni, 1994). 19Thus, even if the factories (capital) damaged by earthquake recovered quickly, it was difficult to keep the employment of female workers as a source of cheap labour.Second, this was caused by the so-called dual structure between stagnant indigenous industries (e.g., food, textile) and grown heavy industries (e.g., machinery, steels, chemical), which emerged after the First World War (Nakamura & Feldman, 1983;Flath, 2005).But, as mentioned above, regardless of the stagnant 1920s, the cotton textile sectors reformed firm organization and adopted new high technology, which boosted labour productivity in the 1930s.
It is also insightful to examine some specific companies and how they were affected by the 1923 earthquake.We do this using Nippon Electric Co. (Nihon Denki Co.), a machine company, and Fuji Gas Spinning Co. (Fuji Gasu Boseki Co.), a textile company.Both were old, big and leading companies with advanced technology and large-scale production.Nippon Electric Co. was founded in 1899 as a joint venture company with Western Electric Co. in the United States and produced telecommunication devices with advanced technology.Fuji Gas Spinning Co., founded in 1896, was one of the largest spinning factories in the world as of 1920.It had over 6000 employees and several modern factories across regions.
Nippon Electric Co. was one of the largest electric machine companies at the time and was located at Mita, Shiba Ward in Tokyo.All its factory buildings collapsed as a result of the earthquake and 105 employees died.The total physical damage was estimated to be around 1,670,000 yen, of which 790,000 yen were for buildings, 390,000 yen for machines, and 240,000 yen for raw materials and parts.The total amount of damage accounted for 17% of total annual sales (Nippon Denki Co., 2001, p. 115).In spite of this serious damage, Nippon Electric Co. immediately drew up a recovery plan and rebuilt several new advanced factories at Mita, as, for instance, one specializing in earthquake-fireproof instruments.Just 34 days after the earthquake, production restarted and all business resumed after 15 weeks (p.116).It should be stressed that Nippon Electric Co. supplied many of the electrical products that were required for recovery.In addition, after the Great Kanto Earthquake, new high-technology electric machine products were developed, for example, radio and automatic telephone switchboards.Radio broadcasts started in Japan in 1925 while the automation of telephone exchange systems commenced in 1926.Thus, due to the huge public demand for telephone exchanges and radio stations, Nippon Electric Co. steadily increased sales (in particular  Natural disasters and regional industrial production efficiency: evidence from pre-war Japan 2061 telephone exchange machines) with the help of its US joint venture company (Western Electric Co.) and sought to catch up using the latest technology (pp.115-121).To put these developments into perspective, Figure 5 shows the company's total profits and return on assets (ROA).
In spite of the earthquake, profits increased, keeping up a high level of ROA.Fuji Gas Spinning Co. was one of Japan's largest cotton spinning companies in the 1920s.It had several modern factories in Shizuoka, Kanagawa and Tokyo prefectures, all of which were seriously damaged areas.Three factories were completely burnt down.One of them, Oshiage factory in Tokyo City, was finally sold without reconstruction in 1924.According to records, the Great Kanto Earthquake resulted in the death of 770 employees and many machines (174,000 spindles, i.e., 30% of all spindles) were also destroyed (Fuji Boseki Co., 1947, pp. 210-211).Indeed, the damage experienced by Fuji Gas Spinning Co. was the most serious in the whole cotton spinning industry 20 and was estimated at more than 13 million yen (Yakura, 1997).As shown in Figure 6, the company's ROA and profits decreased after the earthquake over time without recovery.

METHODOLOGY
To investigate the impact of natural disasters on regional technical efficiency in Japan for the machinery and textile sectors, we adopt a two-stage approach, following Otsuka et al. (2010). 21See also Appendix A in the supplemental data online for more details.
First, we derive a measure of technical efficiency using a SFA model as follows for machinery and textiles separately: where OUTPUT is the value of output produced deflated to 1920 prices (yen), EMPLOYMENT is the number of persons employed in production, FACTORY is the number of factories, and HORSEPOWER is the power for production machines in factories measured by horsepower in prefecture i in year t; μ and λ are prefecture fixed effects and time dummies, respectively, while IE is a non-negative random variable accounting for technical inefficiency in the production function, and V is the usual error term where both are independently distributed for all production units (i ¼ 1, 2, … , N). Importantly, IE it stands for time-varying technical inefficiency scores in prefecture i in year t.If IE it ¼ 0, then prefecture i in year t is defined as being totally technically efficient and is at its maximum output level given the inputs used and technology available.If IE it > 0, then prefecture i in year t is defined as being technically inefficient.In essence, the IE measures the distance to the production possibilities frontier where a greater distance implies greater inefficiency, measured in terms of logged output.Next, we regress the estimated technical inefficiency scores on the number of deaths per capita caused by earthquakes and climatic disasters: where EQ is the number of deaths per 1000 people caused by earthquakes, CLI is the number of deaths per 1000 people due to climate-based natural disasters (high tide, floods and typhoons), and ρ and τ represent prefecture and year dummies.Given that we use the negative value of IE the estimated coefficients can be more intuitively interpreted as impacts on technological efficiency, rather than inefficiency.

Production function
We estimate equation ( 1) for the textile and machinery sectors separately, and the results for these are shown in Table 1, where the estimated coefficients can be straightforwardly interpreted as elasticities.For both sectors, all three inputs significantly predict output.In the machinery sector (column 1), the elasticity with respect to employment is highest while that due to horsepower is lowest.In the production of textiles (column 2), while labour elasticity is also highest in the column 1, regional output in contrast reacts more elastically to the amount of horsepower than to the number of factories.Comparing coefficients between two sectors, the elasticity with respect to employment in the machinery sector is 43% higher.Theoretically, since a capital-intensive sector such as machinery uses less labour, the marginal product of labour (the elasticity in employment) in the capital-intensive sector (machinery sector) is higher than that of a labour intensive sector (textile sector). 22Regional machinery production is also more susceptible (38%) to changes in the number of factories.In contrast, the elasticity with respect to horsepower is 7.7 percentage points higher in the textile sector.The textile sector observed more factories, while textiles used relatively fewer machines than the machinery sector. 23Thus, the marginal product of factories, that is, the elasticity with respect to the number of factories, is lower, while that of horsepower, that is, the elasticity with respect to horsepower, is higher in the textile sector.
Finally, one may want to note that when we excluded the number of factories, horsepower was not a significant input into production for either sector, thus suggesting that it alone is not sufficient to capture capital as an input.

Distribution of inefficiency scores
After estimating equation ( 1), we obtain the technical inefficiency scores from the production frontier.Tables A1 and A2 in Appendix A in the supplemental data online contain summary statistics on basic regional technical inefficiency scores for the machinery and textile industries, respectively.The mean inefficiency score is higher in the textile than in the machinery sector, but less variant.We also depict the distribution of the mean regional scores over time for both sectors in Figures 7 and 8 (see Figure A1 in Appendix A for prefecture names).Inefficiency scores are positively correlated with disaster damages shown in Figures 1 and 2. For the machinery sector (Figure 7), there are some notable patterns of concentration in terms of inefficiency.Specifically, there appears to be a geographical centralization of inefficiency in the industry.Indeed, eight of the 10 most inefficient machinery producing prefectures are in core with big machine production (Tokyo, Osaka, Hyogo, Aichi, Kanagawa, Kyoto, Fukuoka, and Nagasaki prefectures), where Tokyo is the most inefficient and had the biggest machine production.The Great Kanto Earthquake seriously damaged Tokyo and Kanagawa.For the textile sector (Figure 8), while the Osaka is the most inefficient region and had the biggest textile production, 24 there are also some inefficient textile prefectures as for instance, Tokyo, Nagano, Hyogo, Kyoto, Gifu and Gumma.All are big textile producers, and many of them were seriously damaged by earthquakes or climate events.On the other hand, the most efficient producers in both sectors are located at the most southern part of Japan, namely Okinawa south islands, where earthquakes are much less likely to happen and typhoons hit but there is generally small damage due to their high resilience.In both sectors, producers are on average more efficient in the north-east and south-west of Japan, as indicated in Figures 7 and 8 by the greater concentration of yellow or orange shading in that part of the country.These regions have less damage of natural disasters and less manufacturing production.Therefore, prefectures which have many producers and were seriously damaged by earthquakes and/or climate events, tend to have high inefficient scores.Also natural disasters seriously damaged buildings and infrastructure in big cluster regions (Greater Tokyo and Greater Osaka), which is consistent with our anecdotal evidence on the Great Kanto Earthquake, as mentioned above.Natural disasters and regional industrial production efficiency: evidence from pre-war Japan 2063

Production efficiency and natural disasters
Our results of regressing the prefecture inefficiency scores on our natural disaster indicators as in equation ( 2) are shown for the textiles and machineries in Tables 2 and  3, respectively, where we allow for up to five lags of the two natural disaster indices in both. 25For the full machinery sector sample (first column, Table 3) one finds that climatic disasters have no contemporaneous or lagged effects on efficiency in the industry.In contrast, earthquake damage, as measured by the number of deaths per capita in a prefecture, increases inefficiency significantly, albeit as a one-time sudden shock.Taken at the mean nonzero value of EQ (2.52), the estimated coefficient suggests that an average damage-causing earthquake induces an efficiency gain of 0.3% relative to the mean (14.16), whereas the largest observed value (21.47) induced a gain in efficiency in machinery production of 2.8%.As productivity efficiency increased annually by an average 0.0601 logged output and the fact that the average annual value of EQ was 0.102 in our sample, our estimated coefficient on EQ implies that about 3.1% of efficiency gains in the sector was due to damaging earthquakes.
As with the machinery sector, climate-related natural disasters appear to have had no effect on the technical efficiency of textile-producing firms in Japan during the 1920s (first column, Table 2).In contrast, while earthquake damage resulted in a similar rise in efficiency in the industry, the effect was much more persistent.Specifically, the significant positive effect of EQ on -IE in the textile sectors lasts up to at least five years after the event.Additionally, the estimated coefficient only drops marginally as  Preeya Mohan et al.

REGIONAL STUDIES
time since the event passes.If we take into account the cumulative effect over our five-year window, then the estimated coefficients imply that the efficiency gain is at least 0.9% for the average observed damage and 7.7% for the largest observed value of deaths per capita due to earthquakes relative to the mean (17.10) in textiles.Since over our sample period on average production efficiency rose by 0.034 logged output units annually.Thus, using the sum of the coefficients on EQ and its lags and multiplying this by the average annual mean value of EQ (0.037), suggests that annually earthquakes contributed about 6.8% of the productivity gains in textiles.

Robustness checks
Japan is a highly centralized nation, where Tokyo is the centre of government administration, economy, and politics.Additionally, our period saw the above-mentioned drastic urbanization by population inflow, technology growth, and the development of public services, commerce, telecommunication and transportation infrastructure in Tokyo.For this reason, we exclude Tokyo to determine whether this prefecture may be driving our results.
More specifically, we excluded all observations on Tokyo from the total sample and re-ran our efficiency specification in (2).Results of this are provided in the second column for the textiles and machinery of Tables 2 and 3, respectively.Accordingly, for the machinery sector there is still no impact of climatic disasters on efficiency, while for earthquakes the contemporaneous impact remains, although slightly larger.Additionally, one also finds an efficiency boost at t -2.For textiles, the results remain Natural disasters and regional industrial production efficiency: evidence from pre-war Japan 2065 Table 2. Impact of natural disasters on technical efficiency in textiles.
( qualitatively the same as for the total sample, except that the persistent efficiency effect of earthquakes is somewhat larger when the Tokyo prefecture is excluded.
We also experimented with excluding those prefectures where machinery and textile production was least important.To this end, we for each industry, ranked prefectures by mean employment levels in the sector over our sample period, excluded the bottom ranked five prefectures, and then re-estimated the efficiency specification.The results of this exercise are depicted in the third column of the Tables 2 and 3 for textiles and machinery, respectively.For machinery, while the contemporaneous productivity enhancing impact of earthquakes remains as in the total sample, albeit somewhat smaller, one now finds that climatic disasters do boost productivity a year after the event.For textiles, in contrast, all results for the reduced sample are qualitatively the same and quantitatively similar as for the total sample.
Our base specification assumes that the impact of sequential events are independent of each other.Feasibly, however, having a damaging natural disaster in the previous period could affect its impact in the current period, commonly known as the impact of compound events (Zscheischler et al., 2020).See Appendix C in the supplemental data online for more detail.

DISCUSSION
The above results are consistent with the anecdotal evidence as well as stylized facts reported above.As confirmed by our anecdotal evidence and some previous studies (e.g., Imaizumi et al., 2016), earthquakes caused serious damage with a longer recovery time compared with typhoons, because of more complete destruction with the collapse of buildings and social infrastructure.As shown in Figure 4, earthquakes decreased output in all industries but the recovery process differed across industries.While recovery was rapid in the machinery sector, it was sluggish in the textile sector.In parallel, as suggested by our anecdotal evidence of Nippon Electric, a major machine company, the quick recovery in the machinery sector was caused not only by the demand for the recovery of the economy, but also by the large domestic demand for new technology and new products from the Industrial Revolution (Imaizumi, 2014;Abe et al., 2017).This boosted technology growth of the machinery sector, which can thus be seen as a kind of creative destruction.On the other hand, the textile sector was stagnant in the 1920s.The export of silk products, the major export product in the early modernization period, was already in decline in the international market during the First World War.On the other hand, the bubble of the First World War boom broke in 1920 and seriously hit the cotton spinning sector.In the 1920s, the cotton spinning sector faced rationalization of firm organization and stringent labour regulations introduced by revision of the Factory Acts (Nishikawa, 1974).This prolonged the recovery from natural disasters.Abe et al. (2017) notes that during the long recession after the First World War, the cotton textile sector was stagnant in labour productivity but had prepared for economic growth owing to the adoption of new technology and vertical integration from cotton yarn companies to trading companies.Exports dramatically increased as a result of yen depreciation, trade policy reforms and colonial trade in the empire system in this period (Okubo, 2007).In the 1930s, Japan became the largest cotton textile producer in the world.This could explain why the textile sector took longer to recover.
In sum, it is clear that natural disasters seriously damaged technology, but the duration of recovery and technological progress after a disaster are heterogeneous across industries, and is likely in part to depend on whether these events are compound.Technological progress and market conditions in each industry mattered.Some industries such as the machinery sector saw quick recovery and technology growth with the help of substantial technological progress, the market entry of advanced foreign technology, and the demand for new products.On the other hand, for other industries such as the textile sector, recovery was lengthy due to the decline in exports with strong international market competition.Arguably then serious physical damage resulting from natural disasters could create some space for the renewal of technology given the collapse of the old facilities and replacement by new onesalthough this depended on market conditions.
The creative destruction with a time lag and sectoral heterogeneity can be observed in previous studies on natural disasters.In Great Kanto Earthquake of 1923, Okazaki et al. (2019) found that seriously damaged large manufacturing firms were likely to adopt new high technology machines at least two years later.In the Ise bay typhoon of 1959, the largest typhoon in the Japanese disaster history, as shown in Okubo and Strobl (2021), some specific sectors partially observed creative destruction in the flooded areas in Nagoya city, one to four years later.Natural disasters and regional industrial production efficiency: evidence from pre-war Japan 2067 Table 3. Impact of natural disasters on technical efficiency in machinery. (

CONCLUSIONS
In this study we used the case study of pre-war Japan to investigate whether natural disasters could lead to greater production efficiency.We constructed a panel dataset of regional inputs and outputs, as well as indicators of natural disaster damage, for the textile and machinery industries in the 1920s.Employing an SFA approach, we then estimated whether these natural events led to Japanese regions becoming more efficient in production in these sectors.
Our results show that this was indeed the case, at least for earthquake-induced destruction, although the effect was substantially larger due to recovery persisting longer for the textile sector.In contrast we find weak, if any evidence, for climatic disasters, although this may be because we were unable to distinguish between the various climatic types (typhoons, floods, etc.), given the nature of our data.
Our findings confirm the possibility that natural disasters can drive industries to become more productive as they change production technologies in response to damaged or lost physical and human capital.It might be suspected that this is a particular feature of the uniqueness of the historical Japanese setting that we examined.However, using firm level data after the 1995 Great Hanshin Earthquake, Cole et al. (2019) found that, for the manufacturing sector, the least productive firms were more likely to exit if they were damaged, survivors tended to experience a productivity boost after being damaged, and more firms were created in damaged areas around Kobe.Thus, despite generally overwhelming evidence at the aggregate level to the contrary, some industries may actually benefit in the long-run from natural disasters.
Although our results relate to a specific historical context for Japan, they arguably can be used to draw implications for some current settings.In particular, many developing countries are currently in a period of rapid modernization by importing technology from abroad (Galeeva & Zinurova, 2016), as Japan was during pre-war times.At the same time many of these countries are also exposed to a disproportionate share of natural disasters.While such natural disasters will of course cause unwanted destruction, our results here also show that they can present an opportunity for some industries to become more efficient, through either creative destruction or building back better, depending on the industry and the type of natural disaster.Worryingly, however, we also provide evidence that natural disaster compounding may undermine such productivity enhancing effect.This potentially provides a role for government intervention, perhaps through foreign aid, when natural disasters compound over a short period of time.Additionally, more extreme compound natural disasters may be what the future holds with climate change (Zscheischler et al., 2020).

NOTES
1.The uncertainty regarding such a negative impact is whether this will only manifest itself over the short term, that is, whether economies will recover quickly back to their equilibrium growth trajectory or if recovery will only be gradual.2. From a theoretical modeling perspective, the predictions in this regard rest in part crucially on the choice of economic growth model.Models based on neoclassical growth theory are generally only able to predict a negative impact because they assume rather than try to model technical change; in contrast, endogenous growth models can facilitate higher technology growth after natural disasters.3. Pelli and Tschopp (2017) found that countries switch to exporting industries in which they have a comparative advantage after natural disasters.4. One exception is Okazaki et al. (2019), who studied the upgrading of machine horsepower after an earthquake. 5.According to the Census of Manufacture, shares of output in all manufacturing sectors is 42% for textiles (the largest) and 16% for the machine sector (the second largest) as of 1920.Natural disasters and regional industrial production efficiency: evidence from pre-war Japan 2069 REGIONAL STUDIES 6. SFA was previously employed to study the technical inefficiency of Japanese regional industries by Otsuka et al. (2010) and Otsuka and Goto (2015).7. A similar approach was taken for Caribbean countries with regard to hurricanes and their impact on country technical efficiency by Mohan et al. (2019), who found that there was a short efficiency boost for these islands.
Arguably, this approach is more suitable to the context here, since we are explicitly comparing efficiency across regions for the same sector within the same country where it is much more likely that technologies and inputs are similar.In contrast, comparisons at the national level, as in Mohan et al. (2019), must inherently assume that feasible countries can achieve the same technological frontiers regardless of differences in resources and sectoral structures.8. Abe et al. (2017) show that the share of manufacturing was less than 30% in 1900, but this had risen to 44% by 1925.9. Using Japan's Census of Manufacture data, Minami (1976) illustrated the evolution of motive power and emphasized how this evolution contributed to industrialization.10.After the First World War, the indigenous industries stagnated for a time, which led to a 'dual structure' of large productive firms and small unproductive firms (Nakamura, 1971).11.See Appendix B in the supplemental data online.12. Unfortunately, information on the number of factories with machines and the amount of horsepower is unavailable for a few prefectures in 1922 due to technical reasons at the statistical office.The Great Kanto Earthquake in September 1923 destroyed the statistical office building and the government lost the data for 1922 in the process of compiling the data.13.After the big fire, the Order of Fire Defence ('Bouka Rei') was legislated in Tokyo City in 1881.In the 22 main streets in central area of Tokyo City, all buildings were requested to replace the architecture by brick or stone until a due date.In the central Tokyo wards, such as Nihonbashi, Kyobashi, Kanda and Kojimachi Wards, buildings were requested to use fireproof materials (e.g., steel and copper) for their roofs, doors and windows.14.  20. Tokyo Nichi Nichi newspaper (25 September 1923).21.A similar approach was taken by Otsuka et al. (2010) who examined whether agglomeration economies, market access and public fiscal transfers affected Japanese regional industries.They found that while agglomeration economies and greater market access increased efficiency, public fiscal transfers had a negative effect.
22.This is due to the law of diminishing marginal returns.
23.The 1920s were characterized as the period of 'engine revolution' and many manufacturing factories introduced machines with electric motors, involving high-horsepower engines (Minami, 1976).Abe et al. (2017) found that the machinery sector saw a higher percentage of factories with machines as well as the use of machines with electric motors compared with the textile sector as of 1919.
24.The raw correlation between mean regional inefficiency across the two sectors is 0.57.

Figure 4 .
Figure 4. Sectoral output in Tokyo Prefecture.Source: Census of Manufacture.

Figure 3 .
Figure 3. Manufacturing output and employment in Tokyo Prefecture.Source: Census of Manufacture.

Figure 8 .
Figure 8. Inefficiency scores for the textiles sector.
Figure 5. Nihon Electric Co. Source: Nippon Electric Co. Business Report, various issues.
Figure6.Fuji Gas Spinning Co. Source: Fuji Gas Spinning Co. Business Report, various issues.2062PreeyaMohan et al.
Note: Robust standard errors are shown in parentheses.** and * indicate 1% and 5% significance levels, respectively.Year and prefecture dummies are included but not reported.

Table 3 .
Continued.: Robust standard errors are shown in parentheses.** and * indicate 1% and 5% significance levels, respectively.Year and prefecture dummies are included but not reported. Note Tokyo Nichinichi newspaper (2 October 1917).15.Tokyo Asahi newspaper (2 October 1917).16.Tokyo Asahi newspaper (2 October 1917).17.Tokyo Asahi newspaper (2 October 1917).18. Tokyo Asahi newspaper (6 April 1918).19.The Law of Factories came into force in 1916.It prohibited child labour and long working hours of young female workers.However, many workers in textile sectors continued to work for long hours.