The role of the state in fostering industrial development has been heavily contested. Studies of early industrializing countries, in particular Britain, underline how industrial production grew as a result of the unplanned interaction between entrepreneurs, inventors, and financiers (Mokyr Reference Mokyr2010). Economists studying late industrializing countries, on the other hand, frequently emphasize the importance of the state in channeling finance to industrial enterprises. In classic work, Gerschenkron (Reference Gerschenkron1962, pp. 123–26) argued that countries such as Germany or Russia were too scarce in capital to develop through a free interplay of market forces. He proposed that especially in Russia, the state had substituted for “missing” markets by directing investments through state institutions (the most important of which was the public banking system).Footnote 1 Murphy, Shleifer, and Vishny (Reference Murphy, Shleifer and Vishny1989) famously generalized Gerschenkronian arguments in a formal model, which stressed the need for coordination between the investment decisions of individual enterprises through a government-initiated “Big Push.” The importance of state-sponsored finance for industrial growth has since been used to explain the rapid development of East Asian economies such as South Korea after WWII (Woo Reference Woo1991; Allen Reference Allen2011; Lane Reference Lane2025). More recent policy debates on the role of national development banks in the industrial policy of emerging economies such as Brazil tread similar ground (Musacchio et al. Reference Musacchio, Lazzarini, Makhoul and Simmons2017).
Empirically, however, research into the role of states in financing industrialization often suffers from two limitations (Juhász and Steinwender Reference Juhász and Steinwender2024; Juhász, Lane, and Rodrik Reference Juhász, Lane and Rodrik2024). On the one hand, it is hard to separate the causal impact of industrial policy from other country-specific factors (such as endowments, culture, or technology). We circumvent this limitation in this paper by investigating the role of industrial policy in a single country: the Russian Empire. This choice is auspicious because the Russian Empire underwent an unprecedented industrial boom in the final decade of the nineteenth century. As shown in Figure 1a, industrial production grew by 8–9 percent annually between 1890 and 1900, out-performing other prominent industrializing countries of the time, including Germany and Japan. Moreover, Russian policy in the 1890s promoted direct lending by the state to industrial enterprises. Figure 1b demonstrates that output in provinces receiving more credit from the state banking system indeed grew at a faster rate. Based on such patterns, older generations of economic historians have followed Gerschenkron in attributing Russia’s industrialization to the provision of public funds (Crisp Reference Crisp1976; Garvy Reference Garvy1972). However, as more recent contributions have pointed out, there are limitations to drawing causal inferences from highly aggregated data: Russian economic growth may have occurred despite, rather than because of, intervention by the Tsarist state.Footnote 2 This underscores the second challenge in the literature: country-specific studies of industrial policy frequently do not observe the counterfactual, that is, the development of industry in the absence of government intervention.
Figure 1 INDUSTRIAL OUTPUT AND CREDIT DURING LATE INDUSTRIALIZATION
Sources: (a) Bénétrix, O’Rourke, and Williamson (Reference Bénétrix, O’Rourke and Williamson2015); (b) authors’ calculations based on archival material.
This paper brings new micro-data to the debate on the financing of late industrialization. We focus on the ambitious policy by the reformist Russian Finance Minister Sergei Witte between 1892 and 1903 that used the State Bank of the Russian Empire to extend credit to private industry. Before Witte’s appointment, the ability of the State Bank to lend directly to private enterprises had been restricted. During Witte’s tenure, these restrictions were rescinded, and the State Bank lent heavily to industrial enterprises across the country. After Witte was removed from office, his credit policy was dismantled. Witte’s credit expansion therefore presents a clearly delimited policy intervention. We evaluate the impact of this policy on revenue, labor productivity, and machine use at the factory level, leveraging newly geocoded data on manufacturing establishments in the Russian Empire between 1890 (before the start of the policy) and 1908 (after the policy had ended).
This data allows us to exploit the geographical distance from each individual factory to the local bank branch as an exogenous determinant of its access to public credit. We can treat distance as exogenous in our context because the location of bank branches and factories was determined before the start of Witte’s credit policy. Additionally, we show evidence suggesting that neither factories nor bank branches sorted geographically before the start of the policy in a way that affects later factory-level outcomes.Footnote 3 Accordingly, distance to province capitals housing a State Bank branch is a statistically significant predictor of a factory’s outcomes in 1908, after Witte’s credit policy had ended, but not in 1890, before the start of the policy. Moreover, as we use factory data over two periods, we can control for factory characteristics at baseline. Finally, our use of plant-level data allows us to control for fixed effects at the level of the bank branch. In other words, our empirical strategy relies on variation between factories with differing levels of access to the same branch of the public banking system. Unlike much of the literature on the banking-growth nexus (King and Levine Reference King and Levine1993; Jayaratne and Strahan Reference Jayaratne and Strahan1996; Levine, Loayza, and Beck Reference Levine, Loayza and Beck2000; Guiso, Sapienza, and Zingales Reference Guiso, Sapienza and Zingales2004; Burgess and Pande Reference Burgess and Pande2005; Berkowitz, Hoekstra, and Schoors Reference Berkowitz, Hoekstra and Schoors2012; Pascali Reference Pascali2016), our principal estimates do not rely on comparisons between bank branches in different regions. Two recent papers from the literature on banking and growth in historical settings (Heblich and Trew Reference Heblich and Trew2019; Lehmann-Hasemeyer and Wahl Reference Lehmann-Hasemeyer and Wahl2021) rely on finding exogenous variation in regional financial development, either through an instrumental variable strategy or through local policy shocks. In contrast, our study primarily exploits variation at the sub-regional level.
Our empirical results suggest that access to a State Bank branch did indeed lead to a higher growth rate of factory-level revenue, productivity, and machinery use. This offers an explanation of the astounding pace of industrial change in Tsarist Russia. In a second step, we examine how the effect of access to public credit depended on factory and region characteristics. In line with Gerschenkron’s argument, we find that the effect of the State Bank was more important for factories without access to private sources of finance and for factories located in areas where input and output markets were weakly developed.Footnote 4 This suggests the state substituted for private capital or weak fundamentals.
Our research therefore sheds new light on the determinants of Russian industrial growth before the Revolution. Markevich and Nafziger (Reference Markevich and Nafziger2017) and Zhuravskaya, Guriev, and Markevich (Reference Zhuravskaya, Guriev and Markevich2024) highlight the heterogeneity of institutional developments in Imperial Russia. Gregg (Reference Gregg2020), using some of the same manufacturing censuses as this paper, shows how incorporation helped industrial enterprises to grow, despite the complicated concession system involved. Whereas Gregg (Reference Gregg2020) focuses on incorporation as a way for the largest firms to secure equity finance, our study focuses on the mass of industrial plants that relied on external credit. The role of the State Bank in this process has not been quantitatively tested, despite the centrality of the bank to Imperial economic policies.Footnote 5
It is important to note, however, that we do not argue that lending by the government was an optimal allocation of resources in Russia. We show that access to public banking spurred industrial revenue and productivity, especially in poorer regions. We cannot formally assess the full welfare implications of state banking, as we observe neither the opportunity cost of public funds nor the deadweight loss incurred in raising them. It is likely that Witte’s policy redistributed income from the bottom to the top in a society where incomes at the bottom were already meager. This is because Witte’s policies involved a redistribution from taxpayers to recipients of industrial loans. Taxes were largely indirect and thus regressive (Ananich Reference Ananich and Lieven2006; Lindert and Nafziger Reference Lindert and Nafziger2014).Footnote 6 Witte’s policy pursued one goal—industrialization to maintain Russian geopolitical pre-eminence—at steep trade-offs.
Moreover, the Russian state was no impartial social planner. Political patronage networks among industrialists, bankers, and bureaucrats significantly influenced the allocation of credit across the provinces (Lychakov Reference Lychakov2018). These networks facilitated preferential access to financial resources for those within close proximity to key decision-makers. The Russian case therefore contributes to a broader literature emphasizing the role of social networks in entrepreneurship and credit allocation (Lamoreaux Reference Lamoreaux1996; Gupta Reference Gupta2014; Colvin, Henderson, and Turner Reference Colvin, Henderson and Turner2020). We capture personal distance to state bankers using geographical distance between factory and bank as a proxy for loan access, thus exploiting the tendency of business networks to decrease with distance.Footnote 7
The rest of the paper proceeds as follows. The next sections provide the historical background of reforms in Witte’s Russia, describe the data collected, and present the identification strategy. We then estimate the effect of the State Bank on factory-level outcomes, before examining the importance of regional characteristics. The last section concludes.
HISTORICAL BACKGROUND
Establishment of State Bank Branches
The State Bank of the Russian Empire was founded in 1860 as part of a reform package sponsored by Tsar Alexander II. Anxious to retain Russia’s vaunted status as a great power after its defeat in the Crimean War, the Tsar sought to modernize the Russian economy, including its financial system. Upon its founding, the State Bank was effectively incorporated as an agent of the Ministry of Finance. The Bank’s Charter placed tight limits on its ability to issue credit to commercial enterprises, although the Bank did sometimes advance short-term working capital on the basis of discounting promissory notes in the ensuing decades (see Figure 2c). Nonetheless, the Bank’s principal functions during the first decades of its existence revolved around the coordination of public finances, in particular the placing of government bonds, managing the Imperial gold reserve, and collecting and transferring tax payments. This last function mandated the buildup of an Empire-wide system of deposit accounts and a payment mechanism, which led to the establishment of branches outside of the Empire’s principal cities of St. Petersburg and Moscow (Gindin Reference Gindin1960; Bugrov Reference Bugrov2012; Garvy Reference Garvy1972; Ananich Reference Ananich and Lieven2006). Between 1860 and 1866, 33 branches were set up (see Figure 2a). As the objective was to maximize the collection of deposits from the regions, Tsarist authorities allocated branches to the commercially most important towns, that is, those with a high density of taxpayers and savers. Through these branches, the State Bank effectively acted as a giant “pump” funneling resources from Russia’s provinces to St. Petersburg for use by the Imperial government (Bugrov Reference Bugrov2012, pp. 180, 183; Frenkel Reference Frenkel2017).
Figure 2 EVOLUTION OF STATE BANK OF RUSSIAN EMPIRE, 1860–1913. VERTICAL (RED) LINES INDICATE TENURE OF SERGEI WITTE AS MINISTER OF FINANCE
Sources: Crisp (Reference Crisp1976), Bugrov (Reference Bugrov2012), Salomatina (2018), and authors’ calculations based on Russian State Archives.
Further expansion of the branch network proceeded slowly for about a decade. This was due to the difficulty in attracting skilled staff to remote locations and the low level of capitalization of the Bank itself. Deposits also grew slowly in the first decades (Figure 2b). Many branches had difficulty breaking even, partly because the high level of central control written into the Bank’s Charter limited the ability of branches to adjust their assets flexibly to local conditions. Central control, in turn, was perceived as necessary given the low levels of human capital of banking staff employed in the regional branches. This further reduced the appetite for expansion, until resources for a second wave of expansion were again available in the 1880s (see Figure 2a). In expanding, authorities followed the rule “every town a bank,” meaning that a new branch was allocated to the administrative capital of a province. The principal reason behind this decision was that the State Bank was a bureaucratic institution, which followed the general hierarchy of the Empire’s administrative divisions.Footnote 8 By the start of Witte’s tenure in 1892, most provincial capitals had received a local branch (Bugrov Reference Bugrov2012; Frenkel Reference Frenkel2017, p. 184).
The Bank’s early history has important implications for our empirical setup. Firstly, the timing of branch establishment was clearly endogenous to local economic conditions. This precludes a straightforward comparison between branches. Instead, our empirical strategy exploits within-branch variation provided by distance. Secondly, while the timing of branch placement was endogenous, the location of placement was determined by administrative criteria (namely, the location of the provincial capital). Thirdly, the purpose of the early public banking system was not to support local industry. Quite reversely, it was used to “pump” taxable surplus out of the regions. Being geographically close to a branch was therefore not necessarily advantageous for industrialists before the start of Witte’s policy.Footnote 9
Witte’s Policy Experiment
The role of the State Bank changed dramatically after the reformer Sergei Witte took over the Ministry of Finance and hence authority over the State Bank in 1892. Witte was a follower of Friedrich List, the German economist whose writings on development strategies for late industrializing countries dissented from the tenets of classical economics. From List’s writings, Witte distilled two policy recommendations. Firstly, protective tariffs were to insulate Russian industries from more advanced Western competition (Suesse Reference Suesse2023). Secondly, the government was to dispense credit to domestic industry in order to aid its expansion and technological upgrading. This would make Russian factories productive enough to export, first to less competitive markets in Asia, and eventually to Europe. In order to mobilize the public funds for this ambitious scheme, Witte did not only rely on domestic tax revenues, but also increased the issuance of Russian government bonds abroad. Finally, to enhance Russia’s attractiveness for foreign investors, Witte formalized the ruble’s link to gold in 1897 (Drummond Reference Drummond1976; Ananich Reference Ananich and Lieven2006; Wcislo Reference Wcislo2011).Footnote 10
The State Bank was a key institution for the implementation of Witte’s new policy framework. Macroeconomically, it received the right to issue currency backed by gold and would act as Russia’s guarantor of gold convertibility. Most importantly for our purposes, the State Bank would support the provision of industrial credit. In order to carry out these new functions, Witte rescinded the Bank’s restrictive rulebook by sponsoring a new Charter in 1894. This provided Witte with several policy levers. Firstly, the new Charter abolished previous restrictions on commercial lending and additionally made provisions for granting loans to smaller borrowers, which could include individually owned factories or small craft workshops. It also extended the term structure of existing financial instruments, making it possible for entrepreneurs to finance a broader range of capital needs on the basis of promissory notes. Secondly, Witte created a new set of financial instruments specifically designated as “industrial credit” for longer-term investment purposes, which were attractive for capital-investment projects. As Figure 2c shows, the volume of credit under this heading expanded massively once the new Charter took effect. Thirdly, the State Bank not only expanded the volume and structure of lending, but also made credit cheaper (Figure 2d). Interest rates on State Bank loans during Witte’s tenure were lower than in other periods and were typically lower than those demanded by commercial banks (Ishkinina Reference Ishkinina2010; Von Laue 1968; Crisp Reference Crisp1976).Footnote 11 Fourthly, Witte delegated the authority to approve loans to local branches.
During Witte’s tenure, the amount of funds allocated to each regional branch was often decided centrally. This gave precedence to poorer regions, as the center’s aim was “to fill the gap left unattended by other credit institutions” (Crisp Reference Crisp1976, pp. 134, 155). Conditional on these regional allocations, however, the decision on which borrower was to receive credit came ultimately down to the management of local State Bank branches and was not fixed in the Charter. This is crucial for our analysis: local branches enjoyed a large degree of discretion in determining the individual recipient of credit. In utilizing this discretion in loan allocation, they tended to rely on local elite networks, which were often centered around the province capital (Gindin Reference Gindin1960). There were several reasons for this. First, webs of kinship, social relations, or religious affiliation were well-established ways of dispensing credit in Russia, even for commercial banks (Crisp Reference Crisp1976; Rieber Reference Rieber1991; Raskov and Kufenko Reference Raskov and Kufenko2017). The geographically limited nature of such ties is quantitatively corroborated by Hillmann and Aven (Reference Hillmann and Aven2011), who show how business opportunities depended crucially on local network strength in the provinces. Secondly, the exhaustive study on the State Bank by Bugrov (Reference Bugrov2012, pp. 180–93) narrates how, after the start of Witte’s policy, business owners had an incentive to cultivate access to bankers and officials, often managing to obtain a seat on the board of their local State Bank branch. Political capture of local branches by business interests was facilitated by the low rate of turnover of branch managers (due to the difficulties of attracting qualified banking staff) (Bugrov Reference Bugrov2012). Finally, in a painstaking quantitative study of archival records on the linkages between bankers and industrialists, Lychakov (Reference Lychakov2018, p. 3) details the “personal connections, or more formally interlocks, between members at banking boards, government officials, and company board members.” As a result, those factory owners who managed to become close to bankers had an advantage when it came to obtaining coveted funds.Footnote 12
Abrogation of Witte’s Policy
Witte faced a high degree of resistance against his policies, both from the rural poor who paid taxes and from the landed nobility who feared industrialization might undermine their rural power base. However, the eventual end of the experiment was not related to the policy itself. The cause for Witte’s dismissal in 1903 was disagreements over Russian imperial expansion into the Korean peninsula, which he opposed. Witte lost the argument, and Russian expansion went ahead. The resulting war with Japan (1904–05) wreaked disaster on the Russian economy (Wcislo Reference Wcislo2011). Facing mounting fiscal pressures, Witte’s successors rapidly dismantled the State Bank’s expensive credit drive. Industrial loans were once again curtailed (Figure 2c). Instead, the Bank increasingly acted as an orthodox central bank attempting to defend the stability of the ruble by raising interest rates (Ananich Reference Ananich and Lieven2006).
Industrial growth returned after 1907, with larger participation from the now rapidly expanding commercial banking sector. Joint-stock commercial banks had developed sluggishly until the late 1890s (Figure 2b). The slow growth of private banking had partly been due to government restrictions and partly due to the fact that many banking houses were linked to established enterprises, therefore having little incentive to invest in new ventures. Nonetheless, even after private banks had become important actors in industrial finance after Witte’s exit, the State Bank did not entirely abandon its commercial interventions. The Bank continued to act as a lender of last resort to commercial banks and large strategic enterprises in times of crisis (Crisp Reference Crisp1976; Boiko Reference Boiko2011; Salomatina Reference Salomatina2014b). However, the purpose was now to effect emergency bail-outs, rather than foster industrialization.
DATA
Our data set consists of three key components. The Russian manufacturing census of 1908, conducted four years after Witte left office and his credit policies were terminated, provides the endline data. Secondly, we use a similar census from 1890, two years before Witte’s tenure, as a baseline. Thirdly, we employ data on the branching and financing activities of the State Bank from its foundation in 1860.
Geocoded 1908 Factory Data
The “List of Plants and Factories in the Russian Empire” (Varzar Reference Varzar1912) provides the universe of factories in the Russian Empire in 1908.Footnote 13 The unit of observation is the physical plant or factory, rather than the legal entity owning it.Footnote 14 For each factory, the census provides three types of variables.
Firstly, there is data on revenue (in rubles), workforce size (headcount), and installed machinery (by propulsion type and horsepower). These factory-level outcomes serve as dependent variables in later analysis. We calculate labor productivity as revenue per person, including this as an outcome variable due to Witte’s focus on improving industrial productivity.
Secondly, the census details several control variables for analysis, including the owner’s name and social status (noble, merchant, townsmen), type of establishment (workshop, factory), corporate form (none, publicly owned, partnership, share-issuing corporation, co-operative), and industry classification.Footnote 15 We also approximate the owner’s ethnicity from their name (Russian or non-Russian). For many establishments, there is additional data on proximity to railway, riverine, postal, and telegraph stations. We code these factory-level controls as indicator variables.
Thirdly, each factory has a precise address, including its first-level administrative subdivision (province or governorate) and second-level subdivision (district or uezd). The third level, comprising the municipal area (volost in rural areas or gorod in urban regions), is where geocoding occurs.Footnote 16 To geocode, we consult a broad array of sources on local Russian history to match historic town and village names to modern ones, enabling geocoding of factory locations at the municipal level. We geocode 88 percent of factories to this level. Maps B.1 and B.2 in the Online Appendix plot these factories.
1890 Baseline Data and Matching
The 1890 manufacturing census (Orlov and Budagov Reference Orlov and Budagov1894; Orlov Reference Orlov1895) provides a baseline for our analysis. Similar in structure to its 1908 successor, it includes many of the same variables.Footnote 17
We match the factories of the 1908 census to those in the 1890 census in three steps. First, we use a learning algorithm that matches factories by province and the owner’s last name, adjusting for different spellings in each iteration. Second, we manually verify the algorithm’s matches, utilizing the factory’s economic sector and size. Third, we exclude multiple matches (cases where one 1890 factory matches several 1908 factories or vice versa), which can occur due to factory splits or mergers during this period.Footnote 18 This leaves us with 2,677 conservative matches.Footnote 19 The high rate of factory establishment and dissolution implied by the matching quotient is unsurprising in a dynamic economy. The period 1890–1908 also includes several economic downturns, leading to many industrial bankruptcies (Gregg and Nafziger Reference Gregg and Nafziger2024). Importantly, our matched factories are drawn from most districts and nearly all provinces of the Russian Empire (Online Appendix Table A.1), suggesting broad geographical representation.
We compare the characteristics of the 1908 census factories matched to the 1890 census with those that remain unmatched (Table 1). One concern is that our hand-matching might favor larger establishments, as they may be better documented. Yet, there is no statistically significant difference between the means of the matched and unmatched groups for revenues, workforce, urbanization, and distance to the provincial capital. However, unmatched 1908 enterprises use more machinery. This does not necessarily suggest an oversampling of unmechanized factories in the matched group, given that the unmatched group contains many new factories established after 1890. In an era of rapid technological progress, these newer establishments may be more capital-intensive—a phenomenon noted for late Imperial Russia as early as Gerschenkron (Reference Gerschenkron1962).
Table 1 Matching 1908 factories to the 1890 census: Balance
Notes: Non-geocoded factories excluded from both groups. Factories with multiple matches are excluded from both groups. * p<0.10, ** p<0.05, *** p<0.01.
Sources: Authors, see Online Appendices Tables A4 and A5 for detail.
Financial Data
We take the founding year and location of each State Bank branch from Bugrov (Reference Bugrov2012). Most provinces received only one branch, located in the capital. By 1908, 11 provinces had not received a branch of the State Bank in the capital.Footnote 20 In order to control for the presence of private commercial banks, we employ data from Salomatina (Reference Salomatina2014a). She provides data on the location, foundation date, and capitalization of private commercial banks and their branches across the Empire. For the analysis, we simply employ a dummy variable that takes the value of 1 if a municipal area (volost) is home to a private bank.
As a supplementary measure of our banking treatment, we have also collected financial data for all 120 branches of the State Bank in the Russian State Historical Archive in St. Petersburg for the period 1881–1913 (RGIA 2018). This information consists of the branches’ balance sheets at the end of each financial year, as well as their annual turnover of deposits and advances. Annual turnover is further subdivided by the type of financial instrument: promissory notes (vekselia) and industrial loans. Industrial credit includes only longer-term loans. Promissory notes, a shorter-term instrument, are also widely considered to have been employed in industrial credit as well.Footnote 21 Unfortunately, data on individual financial instruments are not always available for all branches, especially before 1900 (see Online Appendix D). While the financial data provide an overview of general trends in credit provision, we therefore do not rely on it for our main results, for which we measure access to finance provided by geography, rather than finance itself. Similarly, wherever we do control for the availability of finance at the branch level in our empirical analysis, we prefer to use a “coarse” binary indicator variable, splitting the sample at the median into branches with high or low credit provision. This avoids a fine-grained continuous measure that might, to a greater degree, be contaminated by measurement error.
In addition, we include demographic and socio-economic variables at the district and province levels. Summary statistics, definitions, and sources for all variables are provided in Online Appendix Tables A.3–A.7, while full data is available at Suesse and Grigoriadis (2025).
EMPIRICAL STRATEGY
Identifying Assumptions
We are interested in measuring the effect of State Bank credit on Russian industry. As the historical overview has shown, the timing of the State Bank’s branching was endogenous to local conditions. This rules out a staggered treatment analysis of branching on provincial outcomes. Moreover, the credit volume each branch extended might have been determined in response to local industrial characteristics. This rules out a simple comparison of credit volumes between provinces (as was done in Figure 1b). Our core strategy is to treat both the branch network and factory location as given in 1890, before the start of Witte’s credit policy, and examine subsequent changes in outcomes at the factory level.
Firstly, we need a measure of the exposure of each factory to the bank that is not driven by factory characteristics. For example, even if individual-level loan data were available, the extension of loans is likely to have been driven by factory revenues or expectations about revenues. We circumvent this problem by using the geographical distance from each factory to its local bank branch as a measure of exposure to finance. Larger distances will be reflected in higher transaction costs for factory owners in applying for loans and in higher monitoring costs on the part of the bank. This is especially salient in Imperial Russia, where distances were large and transport links were still developing (Kahan Reference Kahan1989).Footnote 22 Moreover, the historical overview demonstrated that cultivating personal connections to state bankers was an important factor in receiving loans and that these networks were highly localized (Hillmann and Aven Reference Hillmann and Aven2011). As localized business network strength generally decreases with geographic distance (Chaney Reference Chaney2014), so will the probability of receiving credit.
Secondly, we need to fix the location of the bank branch. Clearly, policymakers could have placed branches closer to important clusters of factories to minimize the costs of accessing finance. In this case, distances would not be exogenous to firm characteristics. We therefore exploit the administrative rule for the location of bank branches: “Every Town a Bank.” In practice, this meant that if a province received a bank branch, it was placed in the town serving as the administrative capital of the province.Footnote 23 We can therefore use the distance from each factory to its provincial capital, rather than distance to a bank branch, as a measure of its access to finance. As the designation of towns as provincial capitals had been historically determined before the creation of the State Bank, this measure is not influenced by bankers’ assessments of the economic potential of a region.
Thirdly, we have to confront the likelihood that distance to the provincial capital coincides with access to administrative services or markets for inputs and outputs, all of which could spur the growth of factories. In this case, we would be picking up a general “capital” effect, rather than the specific “bank” effect. To circumvent this, we interact distance to the provincial capital with the presence of a State Bank branch in the capital, thus using factories located in provinces without a bank in their capital to identify the “pure” effect of being located close to a capital.
Fourthly, we insert fixed effects at the province—and therefore the branch—level. These serve a dual purpose. For one, these control for the possibility that the unobservable characteristics of a region (such as economic potential) could influence the decision on whether to invest in the provincial capital with a bank branch. Furthermore, these fixed effects control for the specific geography of a province. For example, branches in outlying rural provinces may be systematically worse in allocating loans (a valid concern given the difficulties of finding qualified staff in remote locations). In this case, we would still pick up a “real” effect of finance on growth, but the interpretation would be different. Once we include province dummies, however, we capture the variation in access to finance by individual factories, rather than the supply of credit in an entire province.
Finally, having found an exogenous location for the bank branch, we need to fix the location of factories in space. If factories were free to relocate (or could be newly founded), owners could choose to locate close to the bank to minimize transaction costs. This might be a problem if the propensity to do so correlates with factory outcomes (i.e., more successful factories find it easier to relocate). We avoid this threat to our identification strategy by matching factories in 1908 to those already existing in their present location in 1890, before the start of Witte’s policies. As noted, before the State Bank started to dispense credit liberally under Witte, there was little reason to locate close to the State Bank for access to credit. Moreover, to preclude factory owners from locating close to the bank in anticipation of this policy change, we limit the sample to those plants already established in their location before their local branch was founded, as a robustness check.Footnote 24
We can now state our identifying assumption. We estimate the causal effect of banking on factory-level outcomes if the factories that will experience a stronger growth of revenue, machine use, or productivity in the 1890s do not systematically sort closer to the provincial capital in those provinces that eventually receive a branch of the State Bank. Note that causality does not require factories in those two groups of provinces to be identical—it merely requires them to be identical in the dimensions of geographical sorting that are correlated with future growth. We now present evidence that this claim is plausible.
Table 2 shows that there exists no statistically significant relationship between factory-level outcomes in 1890 and the interaction of distance and bank placement. Factories located closer to a capital city with a bank did not exhibit higher levels of revenue, machinery use, or productivity before the start of Witte’s policy. This suggests that there was no geographical sorting by high-performing plants and that the development-distance link did not differ systematically between banked and unbanked provinces. We then run this regression on the same set of plants in 1908, after Witte’s policy of cheap credit. Now we do observe a relationship between factory outcomes and distance in banked towns. Given that distance between factory and bank is fixed by construction, this implies that the bank’s lending activity has changed the importance of distance. Being far away from a bank now carries a penalty that it did not carry before. In Columns (3), (6), and (9) of the same table, we show that this insight is not due to sample selection stemming from our procedure of matching factories across census years. Running the same regression on the full set of geocoded factories in the Russian Empire, we find a similar result (with statistically similar coefficients): differential access to banking is associated with differential outcomes in 1908.
Table 2 No evidence of selection into treatment: Falsification test
Notes: Sample: governorates of the Russian Empire, excluding Central Asia and Finland, 1890–1908. Full sample is not restricted to factories matched across 1890 and 1908 census years. Dep. Var. (1)–(3): natural logarithm of factory-level revenue (in rubles); (4)–(6): horsepower of installed machinery; (7)–(9): natural logarithm of labor productivity (revenue per worker). All regressions are Ordinary Least Squares, with fixed effects at the level of industry and governorate. Factory-level controls include dummies for ownership categories (noble, merchant, townsman, citizen), owner ethnicity (Russian or otherwise), corporation type (public, cooperative, shareholding, partnership), factory type (workshop, retail establishment, factory, craft shop), and infrastructure availability (rail, river, post office, telegraph). All revenue regressions control for incidental revenues. 1908 regressions control for (district level) revenue, machinery and productivity in 1890. Distances measured in kilometers. Bank in capital refers to the presence of a branch of the State Bank in a governorate’s capital prior to 1908. Factory location defined as the factory’s municipality (volost). Standard errors clustered at district (uezd) level (418–651 clusters). Standard errors in parentheses: * p<0.10, ** p<0.05, *** p<0.01.
Sources: Authors, see Online Appendices Tables A4 and A5 for detail.
Specification in Differences, 1890–1908
The preceding analysis of 1908 data offers preliminary evidence of the importance of the State Bank. For our main analysis, we express our dependent variables in differences. This has three advantages. First of all, by examining changes in revenues, workers, or machinery, we can control for the starting level of these variables in 1890. Secondly, this focuses the analysis on factories whose location is fixed in 1890, before Witte’s policy of industrial credit. Thirdly, by utilizing the 1890 census, we gain access to data on the founding year of each factory, which we use for robustness checks. Based on the requirements of our identification strategy, our benchmark specification is:
where t 0 and t 1 are 1890 and 1908, respectively, and i refers to the individual factory, k to the industry, and s(j) to the municipality s that is a part of the province j. Fixed effects (µ j, µ k) are therefore at the province and industry levels. Distance between the factory and the provincial capital d ij is interacted with the presence of a bank b j in the provincial capital [0,1].Footnote 25 The parameter β 2 is our coefficient of interest. It will identify the causal effect of the State Bank subject to the assumptions discussed previously. Dependent variables Y include either growth in revenue, machinery use, or labor productivity, all measured at the factory level.Footnote 26 Factory baseline outcomes Y i,k,s(j)t0—revenue, machine use, or productivity in 1890—are inserted as controls. Establishment-level covariates in X i, such as ownership type, are time invariant.
We cluster standard errors at the level of the district (uezd), resulting in a maximum of 480 clusters. Alternatives to this clustering are explored in Online Appendix C.6.
RESULTS
Principal Results: Access to Banking Improves Factory Outcomes
The empirical results suggest that the State Bank supported the growth of Russian industry. As Columns (2), (5), and (8) of Table 3 demonstrate, factories located farther away from a branch of the State Bank in the provincial capital display a lower pace of growth in revenue, invest less in additional machinery, and experience slower growth in labor productivity. The coefficients on the interaction of interest are statistically significant at conventional levels, and do not change noticeably upon the inclusion of a rich battery of factory-level controls in Columns (3), (6), and (9). Throughout, we control for baseline levels of revenue, machine use, and productivity. The sign on the coefficients of these baseline variables suggests convergence between factories for revenue and productivity, with initially larger and more productive plants growing less rapidly on average. This suggests that the process of industrial growth was not confined to plants that had a head start.
Table 3 Explaining change in enterprise-level outcomes 1890–1908: Benchmark
Notes: Sample: governorates of the Russian Empire, excluding Central Asia and Finland, 1890–1908. Dep. Var. (1)–(3): change in natural logarithm of factory-level revenue (in rubles); (4)–(6): change in horsepower of installed machinery; (7)–(9): change natural logarithm of labor productivity (revenue per worker). See Table 2 for more details on estimation and variable definitions. Standard errors clustered at district (uezd) level (418–480 clusters). Standard errors in parentheses: * p<0.10, ** p<0.05, *** p<0.01.
Sources: Authors, see Online Appendices Tables A4 and A5 for detail.
The coefficient on distance supports the role of the State Bank. In Columns (1), (4), and (7), before inserting the interaction of interest, the coefficient on distance is negative and statistically significant. After interacting with the presence of a bank in the capital, the main effect of distance switches sign and turns positive. This implies that, after accounting for the effect of the bank branch, there is no longer a penalty associated with being far away from the provincial capital. This is quite in line with the discussion on identifying assumptions—industrialists had little to gain from proximity to government services (which were rudimentary, and sometimes predatory) in the absence of Witte’s credit policies. In fact, once we have accounted for the presence of a bank, the results suggest that factories located farther from provincial capitals posted higher growth in revenues and productivity. Apart from a possible escape from government predation, the historical literature suggests two reasons for the benefits of geographical dispersion. Firstly, many industries were dependent on a steady supply of raw materials, the transport of which was costly given Russia’s unreliable infrastructure. In industries where freight costs for inputs outweighed those for the finished product, locating close to raw material sources was beneficial (Spechler Reference Spechler1980; Gregory Reference Gregory2014).
The most important reason for choosing a rural location, however, was access to workers. In Online Appendix Table C.1, we use growth in a factory’s workforce as the dependent variable, and find that plants farther from provincial centers experienced larger increases in employment once we account for the presence of a bank. This apparent paradox is well-established in Imperial Russian history. Much to the chagrin of Lenin and his comrades-in-arms, Russia’s urban proletariat was small, and much industrial labor was provided by peasants. This work was often of a seasonal nature. Before its reform in 1906, the rural commune system also placed restrictions on the distance that peasants could travel to work in manufacturing. Moreover, few were willing to migrate permanently to cities where food and housing costs were exceedingly high. Factory owners, therefore, had an incentive to choose rural sites close to their workforce (Spechler Reference Spechler1980; Chernina, Dower, and Markevich Reference Chernina, Dower and Markevich2014; Gregg and Matiashvili Reference Gregg and Matiashvili2022). Although these location decisions reflected sound economic fundamentals at the time they were made, they proved costly once provincial centers became a lucrative source of finance in the 1890s. Factories close to capitals with a bank are able to expand their workforce at a faster rate, as shown in Online Appendix Table C.1.
The standardized β-coefficients on the interaction of distance and bank presence in Table 4 suggest that the effects we find are economically large. A one standard deviation increase in distance to the bank decreases revenue and productivity growth by more than 0.2 standard deviations, which is a sizable effect. However, to properly evaluate the marginal effect of distance from the bank, we have to take into account two countervailing forces. First, there is the benefit of proximity to bank credit, apparent in the interaction term. Working against this is the (numerically smaller) benefit of being close to rural inputs of raw materials and labor discussed earlier, expressed in the main effect of distance. Table 4 calculates the net marginal effect. Counterfactually moving a factory one standard deviation away from the bank (87km) decreases the growth rate of revenue by 0.13 percentage points (from a mean growth rate of 0.97). The effect on machinery is similar in size (a decrease of 8.9 horsepower from a mean growth of 58hp), while the effect on productivity is comparatively larger. Overall, this exercise suggests that the State Bank had a noticeable effect on factory outcomes.
Table 4 Marginal effects and economic significance
Notes: Coefficients and marginal effects of distance from benchmark regressions (3), (6), and (9) from Table 3. Robust standard errors on marginal effects in parentheses: * p<0.10, ** p<0.05, *** p<0.01.
Sources: Authors, see Online Appendices Tables A4 and A5 for detail.
Potential Threats to Causal Identification
We identify the causal effect of distance to a bank branch conditional on the assumption that the extent of geographic sorting by future factory outcomes does not differ between banked and unbanked locations. We now investigate potential confounds to this assumption. In Table 2, we have already presented evidence that high-performing factories had not sorted closer to provincial capitals with a bank in 1890, before Witte’s policy. However, if factory owners with better growth potential anticipated the policy, they could nonetheless have chosen to locate closer to a bank branch. This would have required a large degree of foresight on the part of entrepreneurs, which seems unrealistic given that Witte’s policy broke with established monetary orthodoxy in Russia (Crisp Reference Crisp1976). Nonetheless, if such sorting occurred, it would not be picked up by examining outcomes in 1890. In Online Appendix Table C.2, we therefore examine the growth of factories that were already established in their location before their provincial capital received a State Bank branch. The coefficient on revenue decreases slightly, the impact on machinery increases, and the effect on productivity is unchanged. In all cases, the coefficient on the interaction remains statistically significant. We go one step further by restricting the sample to factories established before 1860, when the idea of a State Bank was conceived. We again find similar results, albeit with a decreased degree of precision due to the diminishing number of observations.
In the same Online Appendix Table C.2, we also control for proximity to any other State Bank branch (including those not located in a provincial capital). However, we find our coefficients of interest to be largely unaffected when inserting a variable that measures the sum of inverse distances to all other State Bank branches.Footnote 27
Finally, our causal interpretation could be threatened if proximity to banked capitals became more important during the 1890s for reasons unrelated to banking. For example, banked capitals might coincidentally receive better access to railways or experience higher population growth, and these benefits might spill over to closer enterprises. This would not be picked up in our benchmark specification. In Table 5, we investigate these potential confounders. We interact distance to the province capital with an indicator denoting (1) whether a capital was connected to a railway network in 1890; (2) whether a capital received a rail connection after 1890; or (3) whether the capital lay at a junction in the rail network. We find that being close to a rail-connected provincial capital is indeed beneficial for factory growth. The coefficient on our interaction of interest, if anything, increases in size once we control for rail access. In Columns (4) and (5) of the same table, we interact distance with two measures of population growth in the capital.Footnote 28 The results show that proximity to faster-growing capitals is not correlated with higher factory revenue growth. Importantly, the effect of the State Bank does not change upon the inclusion of this control variable. In conclusion, neither railway access nor population growth seems to confound our estimates.
Table 5 Railroads and population growth in province capital 1890–1908: Confounders
Notes: Sample: governorates of the Russian Empire, excluding Central Asia and Finland, 1890–1908. Dep. Var.: change in natural logarithm of factory-level revenue (in rubles). Rail stations and rail junctions (intersections of rail lines) are dummy variables taking the value of 1 if these facilities exist in the provincial capital. Population growth is the compound average growth rate over the period. See Table 2 for more details on estimation and variable definitions. Standard errors clustered at district (uezd) level (424–499 clusters). Standard errors in parentheses: * p<0.10, ** p<0.05, *** p<0.01.
Sources: Authors, see Online Appendices Tables A4 and A5 for detail.
Direct Evidence of Credit Allocation
To what extent do the State Bank’s branch accounts support our conclusions? In Table 6, we introduce a binary indicator capturing the extent to which promissory notes or industrial credit emission at a branch was below or above the median.Footnote 29 In Columns (1), (4), and (7), the indicator for credit based on promissory notes replaces our banking indicator and is interacted with our distance variable. The data suggest that credit at the branch level indeed improved growth in revenue and productivity (but not machinery).
Table 6 Access to State Bank branch and branch credit emission: Promissory notes and industrial credit
Notes: Sample: governorates of the Russian Empire, excluding Central Asia and Finland, 1890–1908. Dep. Var. (1)–(3): change in natural logarithm of factory-level revenue (in rubles); (4)–(6): change in horsepower of installed machinery; (7)–(9): change in natural logarithm of labor productivity (revenue per worker). See Table 2 for more details on estimation and variable definitions. Extent of promissory notes (1896–1900) and industrial credit (1893–1908) emission measured at branch level and converted into indicator variables at the median. Base category for all interactions: provinces without State Bank branches. Standard errors clustered at district (uezd) level (352–476 clusters). Standard errors in parentheses: * p<0.10, ** p<0.05, **** p<0.01.
Sources: Authors, see Online Appendices Tables A4 and A5 for detail.
In the remaining columns, we implement a triple interaction between banking presence, distance, and one of our two financial variables. Comparing the coefficients on this interaction within each column, we find larger coefficients for promissory notes on revenue and productivity, while industrial credit mattered more for machinery investment.Footnote 30 The differential effect between financial instruments ties in with their term structure, as industrial credit was often extended for longer durations and was therefore more suitable for capital investment (Online Appendices G and H). While distance provides a useful proxy of bank access, it is reassuring that the State Bank’s credit data point in a similar direction.
Effects of Factory and Region Characteristics
We now show that the effect of the State Bank is particularly large for factories lacking alternative sources of financing. This is especially true for growth in machinery, indicating that better access to finance mattered especially for long-term investment.
We first investigate whether the effect of the State Bank differs by factory age and size in Table 7. On the one hand, empirical research shows that smaller and newer firms benefit more from improved access to external finance, as they are unable to refinance themselves with retained earnings. On the other hand, the importance of business networks for obtaining credit might have given larger and more established firms an insider advantage (Beck et al. Reference Beck, Demirguc-Kunt, Laeven and Levine2008). For ease of interpretation, we divide factories into two groups and interact them with the treatment.Footnote 31 The results in the upper and middle panels suggest that the benefits of incumbency do not outweigh the benefits that external finance accords to smaller and newer plants: there is little difference in the coefficients according to either size or age. Similarly, we find little evidence that access to banking has differential effects according to the social status or ethnicity of the factory owner (Online Appendix Table C.3). Factories owned by nobles benefit as much from the State Bank’s presence as others. Similarly, factories owned by entrepreneurs with a Russian name (rather than a name associated with ethnic minorities) do not benefit disproportionately from proximity to the bank.Footnote 32 Although the social status of factory owners does not seem to matter, ownership structure does. We show this in the lower panel of Table 7. Factories that have been incorporated as separate legal entities do not suffer from a penalty if they are located away from the bank. This is what we would expect: the purpose of incorporation was to raise capital, either by bringing in partners or by publicly issuing shares. These enterprises were, therefore, not dependent upon State Bank loans. For the majority of establishments, however, incorporation was too costly given the extraordinarily high administrative hurdles attached to the process—which often required the corporate charter to be signed by the Tsar himself (Gregg and Nafziger Reference Gregg and Nafziger2019). Lacking alternative access to capital, these non-incorporated plants benefited from being close to a branch of the State Bank.
Table 7 Access to State Bank branch and factory characteristics: Age, size, and corporation status
Notes: Sample: governorates of the Russian Empire, excluding Central Asia and Finland, 1890–1908. Dep. Var. (1)–(3): change in natural logarithm of factory-level revenue (in rubles); (4)–(6): change in horsepower of installed machinery; (7)–(9): change in natural logarithm of labor productivity (revenue per worker). See Table 2 for more details on estimation and variable definitions. Regressions control for revenue, machinery, and productivity in 1890. Factory characteristics are converted into indicator variables at the median. Base category for interactions: 0-value of respective indicator variable in provinces without a State Bank branch. 1-value of indicator in provinces without a branch omitted from table for brevity. Standard errors clustered at district (uezd) level (403–480 clusters). Standard errors in parentheses: * p<0.10, ** p<0.05, *** p<0.01.
Sources: Authors, see Online Appendices Tables A4 and A5 for detail.
Gerschenkron’s original conjecture and the literature on the Big Push emphasize that the role of state aid in industrialization should decrease with prior levels of development and market size. If markets were large and well-developed, there would be no role for intervention (Murphy, Shleifer, and Vishny Reference Murphy, Shleifer and Vishny1989). The historical literature on Russia agrees that “the main aim [of the Bank] was to increase credit to all branches of the economy where private credit was deficient” (Crisp Reference Crisp1976, pp. 134, 155). What does the empirical evidence say? Table 8 provides a first glance by splitting the sample according to provinces below and above the median value of aggregate industrial output.Footnote 33 We find that being located far from a branch of the State Bank decreases growth in revenue, machinery use, and productivity only in provinces with weakly developed industries. In those provinces already containing substantial industries, there is no penalty for lacking access to the State Bank. We cannot pin down the exact mechanism in this table—industrialized provinces may offer better infrastructure, more developed credit or input markets, or a larger customer base. Yet, these results are in line with the theory on late industrialization, which sees existing market size and government intervention as substitutes (Murphy, Shleifer, and Vishny Reference Murphy, Shleifer and Vishny1989).
Table 8 Access to State Bank branch and 1897 regional development: Industrial output
Notes: Sample: governorates of the Russian Empire, excluding Central Asia and Finland, 1890–1908. Dep. Var. (1)–(2): change in natural logarithm of factory-level revenue (in rubles); (3)–(4): change in horse power of installed machinery; (5)–(6): change in natural logarithm of labor productivity (revenue per worker). See Table 2 for more details on estimation and variable definitions. Regressions split sample at the regional median. Regional revenue is industrial output (in rubles) at the governorate level according to the 1897 census. Standard errors clustered at district (uezd) level (132–332 clusters). Standard errors in parentheses: * p<0.10, ** p<0.05, *** p<0.01.
Sources: Authors, see Online Appendices Tables A4 and A5 for detail.
We improve our insight into the relationship between access to state finance and prior development in Table 9, where we split market development into various components. We focus on machinery use as our dependent variable, that is, on investments requiring a longer time horizon. Columns (1) and (2) show the effect of the State Bank on factories without and with a commercial bank in their municipality. For plants with a nearby commercial bank, there is no statistically significant penalty associated with remoteness from a branch of the public banking system. For plants not close to private banking establishments, there is a very large penalty for also being far removed from public banking: the coefficient on the interaction of interest increases by a factor of 3.3. This suggests that public capital did indeed substitute for private capital in Russia’s industrialization. We provide further evidence of this phenomenon in Online Appendix Table C.4, where we interact the presence of a bank with two proxies for informal inter-enterprise credit (geographic and ethnic proximity to other factory owners). We find suggestive evidence that while access to inter-enterprise credit tends to benefit factory-level outcomes, this advantage disappears in provinces with a public bank.
Table 9 Access to State Bank branch and investment in machinery: Prior market development
Notes: Sample: governorates of the Russian Empire, excluding Central Asia and Finland, 1890–1908. Dep. Var.: change in workforce at factory-level revenue (headcount). See Table 2 for more details on estimation and variable definitions. Presence of commercial banks refers to a non-state bank in the municipality. Other regressions split sample at the regional median. Noble landholdings refer to the share of land held by nobles in a district in 1877. Annual fairs refer to the volume of goods sold during historical private annual fairs in a governorate in 1830. Market potential refers to a district’s population-weighted distance to internal Russian markets. Standard errors clustered at district (uezd) level (160–449 clusters). Standard errors in parentheses: * p<0.10, ** p<0.05, *** p<0.01.
Sources: Authors, see Online Appendices Tables A4 and A5 for detail.
Returning to Table 9, Columns (3) and (4) proxy the strength of landed elites based on land ownership data. We concentrate on the share of land in a district owned by the nobility—the segment of the Russian elite that often opposed industrialization.Footnote 34 Districts with extensive noble landownership had lower shares of merchants or townspeople owning land. In a society where land possession was still a mark of status and economic influence, these were therefore districts where mercantile elements held less power (Von Laue 1968). The results indicate that proximity to the State Bank was important in regions where nobles were economically powerful and mercantile interests were weakly developed. Once again, the size of the coefficient of interest indicates that proximity to the State Bank significantly eased machinery investment in areas where the landed nobility held sway.Footnote 35
We now use a more direct measure of prior market development in Columns (5) and (6), where we examine differential effects according to sales at market fairs in a province. Periodic fairs were the traditional means in Imperial Russia to market local produce, and fairs saw a large offering of agricultural products, raw materials, and basic tools. Places at which such fairs were held would therefore have had an ample supply of inputs for industry. Many fair locations later grew into prosperous industrial towns (Fitzpatrick Reference Fitzpatrick1990). We use detailed data on the volume of sales (in rubles) at these fairs in 1834 as a measure of historic development of markets for industrial inputs. The results suggest that there was little effect of public funds in areas where turnover at these traditional markets was already voluminous, again suggesting that public credit was more important when private markets were less developed.Footnote 36
Finally, we explore heterogeneity in the development of markets for industrial output. Factories located close to customers will find it easier to market products and, therefore, generate sales without government assistance (Donaldson and Hornbeck Reference Donaldson and Hornbeck2016). We, therefore, compute a measure of market access for each district in the Russian Empire by calculating its distance to other districts, weighted by the size of the population in these districts. The results in Columns (7) and (8) demonstrate that successful factories in districts with lower market potential were more dependent on funding from the State Bank. For those factories located in districts close to the major population centers of the Empire, however, access to the State Bank does not seem to have spurred growth.
In summary, these results suggest that the State Bank was particularly important in funding long-term industrial investment in areas where private financial and input markets were weakly developed, where revenues were harder to market, and where mercantile interests were politically weak. In other words, public capital substituted for private capital.
Further Robustness
Some final checks on our main results may be in order. Levels of urbanization at the sub-provincial level could have affected factory outcomes and, therefore, the precision of the estimates. Online Appendix Table C.7 controls for population in the factory’s municipality, as well as dropping the Empire’s dominant provinces of Moscow and St. Petersburg. The results are qualitatively unaffected, despite large variations in sample composition in the latter case.
A more pressing concern may be that our method of matching factories across the 1890 and 1908 census years induced sample selection bias. We have already presented two pieces of evidence that allay this concern. Firstly, 1908 factory characteristics are broadly balanced between the matched and unmatched groups (Table 1). Secondly, the results for predicting factory outcomes in 1908 are very similar when using either the matched sample or the full sample (Table 2). We now experiment with alternative matching procedures. In Online Appendix Table C.8, we replicate our results while omitting the manual check of the validity of matched pairs and unmatched factories. Fully automated matching limits researchers’ discretion, but invariably induces a larger measurement error. We find smaller effects for machinery investment, but broadly similar results for revenue and productivity growth. As a further step, we allow for entrepreneurial dynasties by including factories that may have been split or merged between the census years in Online Appendix Table C.9. We again find slightly smaller coefficients across specifications, but no reason to revise the substance of our conclusions.
We also investigate whether distance might have non-linear effects. In Online Appendix Table C.10, we implement binned interactions, where our bank indicator is multiplied by indicators for middle and upper tertile distances (with the lower tertile being the omitted category). These suggest that the distance effect is particularly noticeable in the upper tertile. Reassuringly, though, dropping very large outlier distances (95th-percentile) does not affect our results qualitatively; neither does dropping provinces with large internal distances (Online Appendix Table C.11). In a similar vein, results are unaffected by dropping randomly selected groups of provinces or industries (Online Appendix Tables C.12–C.15). Some industry-specific effects are apparent, however: We find that the impact of the bank is somewhat more pronounced in industries with greater capital requirements. For example, the effects of the bank are least pronounced in labor-intensive industries such as alcohol and leather, but more pronounced in capital-intensive industries such as mechanical wood processing.
CONCLUSION
Our results are in line with a cautious Gerschenkronian view of the state in fostering industrialization. We find evidence that the presence of State Bank branches did raise the growth rate of revenue and labor productivity in nearby factories (although Witte’s policy seems to have affected a broader range of factories than the large heavy industrial conglomerates Gerschenkron had in mind). The results on the mechanization of production also suggest that public credit did not merely boost sales, but was also utilized for long-term investment. The latter was especially important in regions where local markets were smaller and private capital scarcer.
Our conclusions rely on the identification assumption that the extent of geographic sorting by future factory outcomes did not differ between banked and unbanked locations. By studying the dates of factory foundations and the potentially confounding effects of railways and population growth, we have offered evidence that this assumption is plausible.
In a narrow sense, Witte’s policy was successful. His stated aim was to spur Russia’s industrial growth, and loans from the State Bank did aid this goal. Whether industrialization was the “correct” objective is, of course, a different question. Future research should evaluate the distributional consequences of state-led industrial policy more directly. The social cleavages of Imperial Russia and the violent upheavals they engendered in 1917 suggest this is a pertinent question.
Open access
