POLITICAL AND ECONOMIC MOTIVATIONS FOR JAPANESE RAILWAYS

For the Japanese government, better transport infrastructure (railroads) meant centralization of political control, more responsive national defense, and the ability to spread economic growth to outlying areas (Aoki et al. Reference Eiichi, Imashiro, Kato and Wakuda2000, p. 15; Crawcour Reference Sydney and Yamamura1997, pp. 58–61). Up to the 1860s Japan in effect was a collection of semi-autonomous feudal domains ruled by powerful local families who were vassals of the Tokugawa shogun based in Tokyo (then known as Edo). Lower-ranked samurai in the western territories of Choshu and Satsuma helped to reinstall the emperor as head of government in 1868, who in turn sought advice from a small group of industrialists and technocrats to administer a modernization program. To consolidate its authority, the new Meiji government redrew the political and administrative map of Japan and formed the modern system of 47 prefectures. It also focused on connecting local population centers with the main cities of Tokyo and Osaka via telegraphs and railroads. Improved communications and transport allowed for speedier movement of the new conscript army to pacify restless areas and to prepare for possible attack from western colonial powers. For the private sector, railways were viewed as a profitable venture in no small part because of the rate of return guarantees offered by the government.Footnote ⁹ As railroad investments required larger sums than could be provided by any single private investor, these enterprises made active use of domestic bond markets and the newly established equity exchanges. To ensure that with the private sector delivered the transport network it wanted, the government issued a number of laws in the 1880s and 1890s that standardized railway construction. It also identified which government-planned lines could be bid on by private investors. Localities could also petition for amendments or extensions to the railroad plan after 1892, when the Railway Construction Law was passed. Public-private collaboration continued until the nationalizations of 1906 and 1907. These put the trunk lines in the public sector leaving only urban tramlines and ancillary extensions in the private sector.

While railroads are seen as major engines of growth for the Japanese economy, many regions were already linked commercially before the Meiji period via coastal shipping and some land routes (Yamamoto Reference Hirofumi and Yamamoto1993, p. 5; Lockwood Reference Lockwood1954, p. 105). Coastal freight, the principal cargoes being rice, cotton, cloth, and other dry bulk goods, relied on preindustrial technology and most ships had flat bottoms and single sails; for example, to sail the thousand kilometers between Tokyo and the western island of Kyushu required at least a month's passage.Footnote ¹⁰ Land transport was much more limited, with few roads connecting different regions, and even slower.Footnote ¹¹ Prior to the introduction of the railroad, which reduced the one-day journey to less than an hour, overland travel conditions were described as highly variable and virtually unchanged for a millenium (Free Reference Free2008, p. 11).Footnote ¹² In the decades that followed, rail freight quickly dominated other forms of domestic transport even as the latter experienced rapid growth along with the rest of the economy (Lockwood Reference Lockwood1954, pp. 107–08).Footnote ¹³

Railroad construction in Japan began in 1872 with a 29 kilometer stretch between Tokyo and its nearest deep sea port Yokohama. Two years later, in 1874, a similar length of track was laid between Osaka and the port city Kobe, which were approximately 500 kilometers to the west of Tokyo. The government, which provided funding for these projects, had anticipated that the two major cities of Tokyo and Osaka would be connected by rail quickly. Yet, financial obstacles meant that the Tokaido route between them was only completed in 1889. Given the costs of modernization, deteriorating balance of payments, pacifying rebellious samurai, and controlling inflation, the central government found it difficult to continue investing in railroad infrastructure (Crawcour Reference Sydney and Yamamura1997, p. 59; Yamamoto Reference Hirofumi and Yamamoto1993, p. 12; Aoki et al. Reference Eiichi, Imashiro, Kato and Wakuda2000, p. 11).Footnote ¹⁴ The railroad network's expansion began in earnest in the 1880s when the government allowed the private sector to enter the industry and lay its own tracks in local areas and the parts of the country not served by the public network (see Table 1).Footnote ¹⁵ On the main island of Honshu, both public and private railroads expanded by radiating away from Tokyo and Osaka toward the other major cities. On the islands of Hokkaido, Shikoku, and Kyushu, privately owned and resource extraction-oriented short tracks dominated.Footnote ¹⁶

Table 1

LENGTH OF JAPANESE RAIL NETWORK, KILOMETERS

^aincludes long-distance and local rail and tram networks.

Source:

Japan Statistical Association (2007), series 8-4 and 8-14.

It is easy to find examples of localities and sectors that benefitted from railway access. For instance, after the Shin'etsu railroad was completed in 1893, highly perishable raw silk from the central Nagano area could be carried to the closest train station 40 kilometers away instead of five times that distance to the capital. When the line was extended to the Suwa district itself, “production…increased dramatically and the cocoon collection area expanded” to cover most of sericultural area in central Japan (Aoki et al. Reference Eiichi, Imashiro, Kato and Wakuda2000, pp. 21–22; Ericson Reference Ericson1996, pp. 42–48). Elsewhere consumption of coal, which largely relied on railroad or mixed rail-sea transport for distribution, grew as the network expanded. Countering these examples, however, is the iron and steel sector, which despite obvious industrial linkages to rail construction and transport arguably did not benefit much because of an initial reliance on imported materials. Only after the 1907 railway nationalization that guaranteed demand for domestically produced iron and steel did this sector enjoy any spillovers (Ericson Reference Ericson1996, pp. 32, 38).

Given the debate and largely anecdotal accounts about the impact of railroads on Japanese economic development; ongoing scholarship based on other countries' railroad experience; and industrial and prefectural differences, it seems appropriate to reexamine whether the relationship between Japan's expanding rail system and economic activity was coincident or causal.Footnote ¹⁷ Scholars of Japanese railroads have typically eschewed statistical or economic estimates the railway's impact on the economy.Footnote ¹⁸ In contrast, there is an extensive literature for other countries. It suggests that the railroad improved economic welfare as measured through cost savings, consumer surplus, urbanization, agglomeration, productivity, and market access.Footnote ¹⁹ Many of these studies, especially those using the social savings approach, rely on static counterfactuals to support the claim of welfare gains, which can pose an epistemological problem in attributing causality.Footnote ²⁰

Alternative methods that use time-series data and do not rely on static comparisons have also demonstrated a link between railroads and growth. David Donaldson (forthcoming) uses a general equilibrium trade model and highly detailed price data to estimate the direct impact of an expanding railway system on the colonial Indian economy. This allows the author to calculate reductions in trade costs and regional price differences as well as gains in income, which he finds amounting to a 16 percent increase due to rail access, much higher than Fogel's (Reference Fogel1964) estimate of 4.7 percent for the United States in 1890. Donaldson and Richard Hornbeck (Reference David and Hornbeck2013) examine county-level land values and transport accessibility to assess the impact of railways, and estimate a decrease in value of nearly two-thirds without them. Jeremy Atack et al. (Reference Jeremy, Bateman, Haines and Margo2010) focus on population growth and urbanization as possible consequences of railroad development, and find evidence that railroads “caused” urbanization in the American Midwest.

Looking specifically at the relationship between railroads and firm scale, Atack, Michael Haines, and Robert Margo (Reference Jeremy, Haines and Margo2008) find that American establishments were more likely to be factories, employing sixteen or more workers, in counties that gained access to rail between 1850 and 1870, when the United States was itself undergoing a “transport revolution” via rail.Footnote ²¹ The authors attribute this effect to productivity increases from increased labor division found in larger establishments, and use a difference-in-differences estimator to identify the causal impact. In terms of research design, the current article is most similar to the approach taken by the latter study, with the data and methodology described below.

RESEARCH DESIGN

This article uses prefectural data compiled by the imperial statistical agency attached to the Japanese Cabinet Office (Japan Statistical Association 1962) to analyze the impact of Japanese railways on industries and across regions. Starting with 1883 the agency produced annual volumes of prefecture-level statistics, with coverage of land, population, industry, and health across the country (later including its colonies). The data, which are also disaggregated by major industry groups, include the total number of enterprises in each prefecture and their total invested capital by prefecture. Table 2 presents population and geographic data for Japan and its major islands in the year 1882, immediately prior to the nationwide rollout of the railway.

I constructed an annual data set for each prefecture from 1883 to 1912 covering all major industry groups along with prefecture population. I added coastline length, surface water area, and land area culled from the Historical Statistics of Japan.Footnote ²² A third source of data is the Ekimei Jiten handbook (Chuo Shoin Reference Chuo1995), which provides a comprehensive list of all rail stations established in the country, their founding dates, locations by city and prefecture, and other identifying information. I use the earliest year a station was opened in each prefecture to determine access, and verified these dates with secondary sources.Footnote ²³

Table 2

PREFECTURE STATISTICS, 1882

Source:

Japan Statistical Association (1962, 2007).

I use a difference-in-differences methodology, comparing the annual measures of firm activity in each prefecture before and after it gains railroad access to those with no change to their rail access over the same period.Footnote ²⁴ Following the economic geography and transaction cost literatures, which predict that railroad access increases factor and goods mobility through lower transportation costs and wider market access, this methodology examines whether one observes an increase in firm activity after the introduction of the rail. In other words, one can test the hypothesis that the number of firms and levels of invested capital would increase when areas become connected to the network. Moreover, capitalization levels should differ by sector: because of scale economies, manufacturing would be more affected by rail access than the primary sector. Similarly, one expects increased total capitalization in the services sector, which includes the rapidly expanding financial industry during this period, to facilitate local industrial investment.Footnote ²⁵ Finally, changes in firm activity will depend on pre-access market size, leading to agglomeration in larger markets and dispersion in smaller ones. The reduced form OLS model follows, with separate regressions for each major industry group:

• y_it = β₀ + β₁·x_1it + β₂·x_2i + β₃·x_3t + β₄·x_4it + ε_it , where

• y_it = firm activity for prefecture i in year t

• x_1it = rail access dummy variable for prefecture i in year t

• x_2i = prefecture-fixed effect

• x_3t = year-fixed effect

• x_4it = interaction of rail access and control variables for prefecture i in year t

• ε_it = error term

To measure the effect on firm activity, the dependent variable y_it is one of three measures: the total annual number of firms in each of the prefectures, indexed by i; the total capitalization of firms by prefecture; and the average capital per firm in each prefecture. These three variables capture extensive (firm number, total industry capital) and intensive (average firm capital) firm activity over time. The dependent variables are available for the economy as a whole and for each of the major industry groups, with the results from the latter allowing one to determine their relative contribution to the national economy.

The main control variables of interest are the term x_1it , which takes the value of zero for all years prior to rail access in prefecture i and the value of one in post-access years, and the interaction terms x_4it that account for initial conditions at the prefecture level that are exogenous to the arrival of the railroad.Footnote ²⁶ These variables include the prefectural population in 1882, prefectural coastline length, surface water area, and land area. Prefectural population in 1882 can be interpreted as a crude proxy for local market demand and agglomeration potential. Similarly, coastline length and surface water are used to proxy for access to coastal and inland water transport, which may act as substitutes or complements to rail transport. Land area is used to control for differences in the time needed for rail construction and market density, given the inclusion of initial population. Finally, fixed effects for prefecture and year are also included in the model to allow for local or temporal variation.

This article uses the years between 1883 and 1893 to identify a treatment period, and thus separates the country's prefectures into two groups: a treatment group of 18 prefectures that gained access during this period, and a control group of 18 prefectures that gained access after 1893.Footnote ²⁷ The starting year was selected based on the earliest available data while the end year was selected for three reasons: first, it precedes the Sino-Japanese War (1894/95), which contributed to a dramatic expansion of the public investment in armaments and public works and may have had knock-on effects to firm activity that differed substantively from previous years. Second, the 1893 Commercial Code led to a reclassification of enterprise ownership, which makes the numbers reported not directly comparable to earlier years. Lastly, this year also marks the promulgation of the previous year's Railway Construction Act, which increased the scope of private investment in railways. In terms of historical significance, this period also coincides with the first wave of railway expansion and the onset of industrialization.

For a causal interpretation of estimates based on the difference- in-differences methodology, two conditions should be met. The first is in the choice of control and treatment groups, which must be similar in aspects aside from the treatment (i.e., rail access). This seems reasonable given the government's aim to connect all prefectures to the national system, which was largely accomplished by the end of the Meiji period.Footnote ²⁸ While commercial activity and population size were considerations of the government in deciding the placement of railways, they were not determining factors.Footnote ²⁹ Thus, separating control and treatment prefectures becomes effectively an issue of timing, and prefectural differences were largely due to geography. In particular, official documents indicate that terrain difficulty was the primary impediment to railway construction to certain prefectures, with those gaining earlier access located along the low-lying plains and coast.Footnote ³⁰ This natural variation allows for division of the prefectures into control and treatment groups based on year of access and other time-invariant features. Table 3 compares the control and treatment prefecture groups with a breakdown of various characteristics.

To increase the comparability of prefecture groups in the analysis, estimates from specifications that exclude prefecture outliers are reported along with those including all prefectures. Outliers are identified as prefectures in either the top or bottom 5 percent of any of the following variables: population in 1882, coastline length, surface water area, and land area. An additional robustness check uses only prefectures located on the main island of Honshu, which had railways that did not require other forms of transport to reach the major cities of Tokyo and Osaka and their nearby ports, and is the location of the major pre-Meiji land routes.

The second condition for a causal interpretation is that the decision to extend the railway network was not itself determined by the measured outcomes of firm activity. In other words, entrepreneurs intending to establish firms or expand operations should not have influenced the decision to introduce rail access to a particular location. As mentioned above, geographic considerations figured prominently in the timing and selection of prefecture access, with considerable delays occurring due to land gradient and government preoccupation with possible invasion along the eastern coast of Honshu (facing the Pacific Ocean). Prior to the 1892 Railway Construction Act, private rail companies needed to submit applications to construct and manage railways planned by the railway ministry, with no guarantee of approval. Official reports from the railroad ministry indicate that the main priorities of railway construction were to facilitate transportation between regions, to increase national security, and to promote industry.Footnote ³¹ That is, the government's objective to establish a railroad network at the start of the Meiji period was not explicitly for commercial development.Footnote ³²

Table 3

PREFECTURAL AVERAGE COMPARISON, 1883–1893

Source:

Japan Statistical Association (1962, 2007).

In practical terms, extensions of the system north and west of Honshu connected population centers usually along the most geographically and cost-efficient paths and were meant to (and did) facilitate troop movements and securing the Pacific coastline, which was considered more vulnerable to foreign invasion.Footnote ³³ Moreover, before the creation of the national parliament in 1890 and the passage of the 1892 Railway Construction law, there was no explicit local and commercial lobbying for the placement of railroad track.Footnote ³⁴ To address concerns about endogeneity, the analysis uses data up to the year 1893 and includes specifications that only use as a control group prefectures that did not gain rail access until after the period of analysis. There is also a robustness check that includes only those prefectures whose capital cities gained access the same year as the prefecture in general.