The second half of the nineteenth century was a turbulent period for international trade. While trade flows were generally liberalized following the Cobden-Chevalier treaty and its successors (Bairoch Reference Bairoch1972; Jacks Reference Jacks2006; Lampe Reference Lampe2009; Tena-Junguito, Lampe, and Fernandes Reference Tena-Junguito, Lampe and Fernandes2012), cheap (American) grain imports prompted many (European) countries to reintroduce protectionist tariffs after the 1880s (Bairoch Reference Bairoch1972; O’Rourke Reference O’Rourke1997, Reference O’Rourke2000).Footnote 1 As these policies coincided with rapid industrialization, it raises the question of whether tariff increases were driving economic growth. Indeed, some observers link the strong growth of, for example, the United States and Germany during this period to their adopted protectionism (O’Rourke Reference O’Rourke2000). While protectionism is generally seen as hurting the economy (e.g., Edwards Reference Edwards1998; Frankel and Romer Reference Frankel and Romer1999), higher tariffs presumably helped industrialization in the late nineteenth century (e.g., Bairoch Reference Bairoch1972; Lehmann and O’Rourke Reference Lehmann and O’Rourke2011; O’Rourke Reference O’Rourke2000). This empirical relationship is known as the tariff-growth paradox but is heavily contested in the literature (e.g., Irwin, Reference Irwin, Dormois and Lains2006; Schularick and Solomou Reference Schularick and Solomou2011).Footnote 2 Lampe and Sharp (Reference Lampe and Sharp2013) synthesize the contradictory findings. While the previous literature relies on cross-country regressions, they analyze the effect of tariffs in individual countries, documenting a large heterogeneity in the relationship between tariffs and economic growth. What explains such heterogeneity?
Disregarded in the historic tariff-growth literature is a focus on firms.Footnote 3 This is unfortunate as cross-country regressions implicitly assume that import tariffs have a uniform impact across firms. However, given large differences in the historical performance of firms (Online Appendix Figure A.1), this is a questionable assumption. Contemporary trade models also highlight a heterogeneous impact of tariffs across firms. Conditional on their initial performance, tariffs differentially shape the incentives to innovate and efficiently organize production. As higher import tariffs lower domestic competition, initially low-productivity firms might be encouraged to innovate more, thus increasing their productivity, while the opposite holds for initially high-productivity firms (e.g., Aghion, Antonin, and Bunel Reference Aghion, Antonin and Bunel2021; Iacovone Reference Iacovone2012; Shu and Steinwender Reference Shu, Steinwender, Lerner and Stern2019). Thus, firm-level heterogeneity within and across countries might be one explanation for the varying findings in the historic tariff-growth literature.
I contribute to the historic tariff-growth literature by estimating the causal impact of a substantial increase in import tariffs on establishment-level performance. To that end, I leverage the Historical Manufacturing Census of Sweden, a novel and highly detailed panel data set covering Swedish manufacturing establishments in the late nineteenth century. I combine these data with industry-level import tariff rates collected by Persarvet (Reference Persarvet2019). I mainly rely on two-way fixed effects regressions, controlling for establishment and year characteristics, to estimate the establishment-level impact of a sharp rise in tariffs between 1891 and 1900. Sweden is an ideal choice for studying the existence of a potential tariff-growth paradox. While the country was among the most rapidly growing economies in the late nineteenth century (Schön Reference Schön2012, pp. 82, 136–37), it also sharply increased its (industrial) import tariffs in the 1890s (Figure 1) to promote domestic manufacturing. As such, the economy became effectively protected (Persarvet Reference Persarvet2019, pp. 83, 160–71).
Figure 1 AVERAGE SWEDISH IMPORT TARIFF RATE
Notes: Highlighted are the years 1887 and 1891, after which the main tariff increases occurred.
Source: Persarvet (Reference Persarvet2019).
To preview my findings, I first document that—consistent with previous research—tariff increases had no overall impact on establishment-level labor productivity, which is calculated as the natural logarithm of establishment-level sales divided by the number of workers. However, I subsequently show that this result is driven by heterogeneous establishment-level responses to tariff changes. My main finding is that a 1 percentage point increase in tariffs led to a productivity increase of about 3 percent for the initially least productive establishments. In contrast, initially high-productivity establishments saw a relative productivity decline of about 3 percent. Establishments with initial productivity levels between the extremes noticed smaller changes. Considering that the average Swedish tariff rate rose by about 3 percentage points between 1891 and 1900 (Figure 1), these are also economically significant findings.Footnote 4 I substantiate this finding in several robustness checks, using different measures of establishment-level productivity and tariff rates, other regression specifications, two-step system GMM regressions, and an Oster bound analysis (Oster Reference Oster2019).
To explain the results, I draw on insights from contemporary trade theory (e.g., Aghion, Antonin, and Bunel Reference Aghion, Antonin and Bunel2021; Chen and Steinwender Reference Chen and Steinwender2021; Iacovone Reference Iacovone2012) to provide suggestive evidence that the tariff increases differentially shaped the incentives of managers in initially low-and high-productivity establishments to innovate and (re)organize production. While higher tariffs increased the incentives of initially less-productive establishments to innovate and efficiently (re)organize production, initially relatively more-productive establishments saw a decrease in their innovative efforts. Ultimately, firms exhibited heterogeneous productivity dynamics. I also present evidence against alternative explanations, such as increased input prices, tariff retaliation, regional shocks, and changes in the establishment-level capital stock. Overall, my results show that heterogeneous establishment-level responses contribute to a potential tariff-growth paradox.
The history of Luth & Roséns Elektriska Aktiebolag nicely illustrates how tariffs plausibly shaped establishment-level performance, innovation, and learning. Luth & Roséns was based in Stockholm and active in the machinery and instrument industry, which was highly innovative according to patent data (Andersson, Berger, and Prawitz Reference Andersson, Berger and Prawitz2023). While the firm was initially one of the least productive establishments among its industry-level competitors, it saw an impressive rise in productivity between 1891 and 1900 as sales per worker increased from about 400 to 4.300 kronor. At the same time, it developed from merely installing lighting to producing advanced technological goods such as dynamos. As its industry-level tariff rate increased by about 40 percent in 1892, it is plausible that higher tariffs were conducive to learning processes within the establishment. Since the owners of Luth & Roséns used the previous decades to travel abroad and acquire technical knowledge (Hildebrand Reference Hildebrand1905; Wannberg Reference Wannberg1984), higher tariffs potentially gave the establishment room to utilize this knowledge to experiment and eventually succeed in producing advanced technological goods. Indeed, my establishment-level data support the presence of learning processes within Luth & Roséns. For example, in 1891, the firm was owned by an engineer, who was likely in a strong position to improve and optimize production. When tariffs were increased in 1892, the firm started using electrical power. As only about 40 firms used electricity in that year, Luth & Roséns increasingly became a forerunner of technological change and innovation.
Foremost, my paper relates to the literature on the tariff-growth paradox by highlighting the importance of considering establishment-level heterogeneity when estimating the historical impact of tariffs. Overall, the results are consistent with dynamics postulated in the modern trade literature and suggest that tariffs did not have a unique impact in the late nineteenth century. The paper thereby highlights that establishment-level heterogeneity can be one contributing factor to the largely inconclusive results of previous studies relying on cross-country regressions to estimate the historical impact of tariffs. Ultimately, the sign and magnitude of the tariff-growth relationship depend on the characteristics of firms across countries. As late-industrializing countries have a relatively larger need for state intervention to shape the industrialization process (Gerschenkron Reference Gerschenkron1962), this paper highlights the ambivalent role tariffs can play. While some firms profited from tariffs, thereby supporting arguments for (temporary) protection, trade policy also created losers. At the aggregate level, this had (at least in the short run) no large positive or negative consequences. Given that my data end in 1900, analyzing the long-run consequences is left for future research. The paper also relates to modern studies on the effectiveness of tariffs on firm-level development (e.g., Chen and Steinwender Reference Chen and Steinwender2021; Iacovone Reference Iacovone2012) by extending this literature back into the nineteenth century. Finally, the paper adds to previous studies on the effectiveness of Swedish tariffs at the aggregate level (e.g., Bohlin Reference Bohlin2005; Persarvet Reference Persarvet2019).
HISTORICAL BACKGROUND AND THEORETICAL FRAMEWORK
Swedish Tariffs and Trade Patterns
SWEDISH TRADE POLICY
Traditionally, Sweden was a relatively poor agrarian country on the European periphery. Its industrialization after 1850 was ultimately linked to growing exports of staple goods, while production for the domestic market also expanded. While growth slowed in the 1880s, Sweden was among the fastest-growing countries worldwide after 1890. Economic growth also became more broad-based, centering around emerging industries such as engineering, machinery, pulp, paper, and chemicals, while the older industries experienced slower growth (Schön Reference Schön2012, pp. 73–125). Overall, after 1890, Swedish output was increasingly consumed on the domestic market rather than exported (Persarvet Reference Persarvet2019, p. 73).
In conjunction with international developments, the Swedish economy and trade policy were essentially liberalized after the 1860s (Schön Reference Schön2012, pp. 92–93). While imports were historically restricted through tariffs and other prohibitions, these restrictions were now lifted. In 1857, Sweden removed tariffs on raw materials, inputs, tools, machinery, steam engines, agriculture, and foodstuffs, and the joining of the Cobden-Chevalier network in 1865 further increased trade liberalization (Persarvet Reference Persarvet2019, pp. 80–97).Footnote 5 Swedish trade openness increased in the following decades, with imports exceeding exports (Persarvet Reference Persarvet2019; Schön and Krantz Reference Schön and Krantz2012; Online Appendix Figure A.2). This coincided with a reorientation of Swedish trade toward Germany at the cost of Great Britain (Online Appendix Table A.1). Also, the types of traded goods changed (Online Appendix Tables A.2 and A.3). Around 1860, Sweden mainly imported agricultural and food products, but over time the share of inputs (e.g., mining, oil, coal, metal, and chemicals) and final goods such as textiles or machines increased. After the 1880s, imports mainly consisted of inputs rather than final goods. While Swedish exports were mainly centered around wood, iron, and oats, new goods such as pulp, machinery, butter, and matches were increasingly exported toward the end of the century, reflecting the economy’s industrial upgrading (Persarvet Reference Persarvet2019, pp. 69–73).
As the influx of cheap American grain in the 1880s threatened domestic rents, Sweden, alongside many other countries, returned to protectionism (O’Rourke Reference O’Rourke1997). Because grain imports mainly hurt farmers, the debate over whether tariffs should be raised first concerned agriculture (Persarvet Reference Persarvet2019, pp. 80–84). The intensity of the debate increased over time and became one of the most important political conflicts in modern Swedish history (Lehmann and Volckart Reference Lehmann and Volckart2011). Interest groups were formed to influence public opinion and, consequently, the government. The groups published widely, for example, in newspapers, lobbying for or against higher tariffs.Footnote 6 The debate was generally motivated by the economic interests of individuals and groups (Montgomery Reference Montgomery1921, pp. 137–82). Generally, capital owners and workers supported free trade, whereas farmers favored protectionism (Lehmann and Volckart Reference Lehmann and Volckart2011). Proponents of agricultural tariffs maintained that tariffs would increase agricultural incomes, which would enable farmers and agricultural workers to increase their consumption of industrial goods (Montgomery Reference Montgomery1921, pp. 137–82). Additionally, as jobs in agriculture would be more secure, industrial workers would have less competition in the labor market (Esaiasson Reference Esaiasson1990, pp. 83–94). In contrast, free traders feared that higher food prices would hurt the overall strength and productivity of (industrial) workers (Esaiasson Reference Esaiasson1990, pp. 83–94) and increase industrial wage demands (Gårestad Reference Gårestad1987, pp. 111–12). Additionally, the Swedish industry had successfully grown without protectionism previously (Esaiasson Reference Esaiasson1990, pp. 83–94). Over time, representatives from industries with high import competition began supporting agricultural tariffs, hoping to receive tariff protection as well.Footnote 7 Analyzing letters sent by firms to the Swedish tariff commission, Gårestad (Reference Gårestad1987, pp. 111–24) finds that industrialists supported infant industry protection to catch up with more advanced production methods and technologies from abroad.
In the end, the Swedish shift back into protectionism occurred somewhat accidentally and followed two elections in 1887, which almost resembled a referendum on the tariff question (Lehmann and O’Rourke Reference Lehmann and O’Rourke2011).Footnote 8 The liberal party, favoring free trade, won both elections.
But after the second election, it was discovered in unrelated events that one of their members did not fully pay his taxes. According to the election law, he and several other members were expelled from government, giving their opponents a majority, who then raised tariffs in 1888.Footnote 9 Tariffs were first increased in agriculture as industrial tariffs were still bound by previous international agreements. However, a long period of tariff increases followed, and industrial tariffs were mostly raised in 1892 when the previous agreements ended. Tariffs on agricultural and industrial goods were again raised in 1895 (Bohlin Reference Bohlin2005; Lehmann and Volckart Reference Lehmann and Volckart2011; Persarvet Reference Persarvet2019). By this time, the motivation for tariffs had shifted, however, and tariff legislation was sought to promote the production of finished goods (Lindberg Reference Lindberg1983, p. 73). Sweden now followed a solidaristic tariff system, which granted industries relatively higher tariffs if they relied on inputs subjected to tariffs. Thus, inputs were least protected or even duty-free, capital goods faced higher tariffs, and final products were most protected (Bohlin Reference Bohlin2005).Footnote 10 Over time, the debate between free traders and protectionists shifted and centered more around the height of tariffs than whether they should be imposed at all. The preference of industrial workers for industry-level tariffs prevented substantial decreases in later years (Carlsson Reference Carlsson1961, p. 571).
Figure 2 illustrates the development of the Swedish tariff structure for the sample of industries analyzed in this paper.Footnote 11 Focusing on nominal tariff rates, the figure highlights that tariff protection was substantial across industries and expanded mostly after 1891. It also reflects that the Swedish tariff structure favored the production of finished goods, as the highest tariffs were granted to the food and beverages, textile, and paper industries.
Figure 2 SWEDISH INDUSTRY-LEVEL NOMINAL TARIFF RATES AND ERP
Notes: The figure shows the Swedish nominal tariff rates and effective rates of protection across industries. Highlighted are the years 1887 and 1891, after which the main tariff increases occurred. Note the different scales across industries.
Source: Persarvet (Reference Persarvet2019).
Nominal tariff rates are potentially an unsatisfactory measure, however, since they do not account for tariffs on inputs. Effective rates of protection (ERP) are a popular way to address this issue as they measure the effect the tariff structure has on value added.Footnote 12 However, since ERPs are based on strong assumptions (Corden Reference Corden1966; Laird Reference Laird, Joseph and Kenneth1997; Persarvet Reference Persarvet2019), this paper mainly uses them as robustness checks. Figure 2 shows that ERPs were generally higher than nominal tariff rates, which is indicative of tariff escalation. ERPs are usually evaluated relative to their average value across all sectors (Corden Reference Corden1966; Laird Reference Laird, Joseph and Kenneth1997). Persarvet (Reference Persarvet2019, pp. 160–71) provides such comparisons, suggesting that the late-nineteenth-century Swedish tariff structure favored industry, especially the food, textile, wood, paper, and metal industries. In contrast, agriculture and input products were disadvantaged. This further strengthens the possibility of a positive relationship between tariffs and industrial production.
THE EFFECTS OF INCREASING SWEDISH TARIFFS
Several studies analyze the effects of the Swedish shift back into protectionism, which, similar to the general tariff-growth literature, remain at the aggregate level and are somewhat inconclusive. A public investigation, including Eli Heckscher and Arthur Montgomery, concluded that the tariff increases hurt Swedish economic development as they supported unprofitable industries producing for the home market at the cost of more profitable export industries (Tull- och traktatkommittén 1924). Both authors later diverged from their assessment, arguing that the tariff increases in 1888 and 1892 did not play a large role in Swedish economic development (Heckscher Reference Heckscher1941; Montgomery Reference Montgomery1947). Jörberg (Reference Jörberg1966, pp. 30–31) maintains a similar view, while Hammarström (Reference Hammarström1970, pp. 48–52) argues that the tariffs created import substitution in the consumer goods industries by decreasing the imports of final products. This conclusion is echoed by Schön (Reference Schön and Dahmén1989, p. 242), who argues that tariffs contributed to the emergence of a mass market for domestically produced consumer goods. Using aggregate data, studies generally find no clear effect of tariffs on growth in Sweden (Lampe and Sharp Reference Lampe and Sharp2013; Potrafke, Ruthardt, and Wüthrich Reference Potrafke, Ruthardt and Wüthrich2021).
Two studies analyze the effects of the Swedish tariff increases in more detail. Bohlin (Reference Bohlin2005) provides the first quantitative estimates of Swedish tariffs for a sample of commodities, showing that the tariffs successfully reduced import penetration across many industries. As the new tariffs mainly protected the consumer goodsFootnote 13 and food industries,Footnote 14 imports in these industries declined sharply. As raw materials and other inputs were less protected or even duty-free, the imports of such products increased. Thereby, the tariff structure supported import substitution in the Swedish domestic consumer goods industries. Persarvet (Reference Persarvet2019) provides the most comprehensive analysis of the late-nineteenth-century Swedish tariff policy, also suggesting that the tariffs decreased import penetration, especially in the consumer, capital goods, and food industries. The import of duty-free inputs, raw materials, and fuels increased. Again, the simultaneous expansion in the domestic production of especially consumer products suggests a process of import substitution. However, tariffs had a rather negligible impact on labor productivity within industries or on moving labor from agriculture to industry.
When collapsed to the industry level, the establishment-level data used in this paper generally support the view that tariff increases did not have an impact on, for example, sales or productivity (Online Appendix Table A.4).Footnote 15 However, such findings hide a potentially heterogeneous impact of tariffs.
Theoretical Framework
A traditional argument for increasing import tariffs is infant industry protection. Given learning effects or economies of scale, tariffs can help protected industries move down their average cost curve until they are internationally competitive (Persson and Sharp Reference Persson and Sharp2015, pp. 178–79). Indeed, there is evidence that this channel explains the positive relationship between tariffs and industrial growth in the late nineteenth century (Lehmann and O’Rourke Reference Lehmann and O’Rourke2011).Footnote 16 Recent advances in international trade theory suggest that tariffs have differential effects depending on the initial characteristics of firms, which could help explain the heterogeneity uncovered in the literature on the tariff-growth paradox. Drawing on, for example, Schumpeter (Reference Schumpeter1942), Arrow (Reference Arrow and Nelson1962), and Aghion et al. (Reference Aghion, Bloom, Blundell, Griffith and Howitt2005), Shu and Steinwender (Reference Shu, Steinwender, Lerner and Stern2019) build a framework of two channels through which import tariffs change the learning and innovation practices of establishments, thereby ultimately affecting establishment-level outcomes.Footnote 17
According to the Schumpeterian effect, lowering tariffs increases domestic competition, thereby decreasing the market share of domestic establishments. This reduces the post-innovation rents of establishments and consequently their innovative efforts and productivity.Footnote 18 This effect is more pronounced for establishments with initially low productivity since they have less chance of catching up with leading firms. As I analyze an import tariff increase and not a decrease, the opposite should hold. A tariff increase reduces domestic competition, which should increase the post-innovation rents, especially for initially low-productivity firms. Thus, initially low-productivity establishments are expected to exercise more innovative efforts and learning, so their productivity likely increases.Footnote 19
The escape-competition effect is especially relevant for initially high-productivity firms (Shu and Steinwender Reference Shu, Steinwender, Lerner and Stern2019). Accordingly, a decrease in tariffs lowers the market share of domestic establishments, leading to a decline in their pre-innovation rents. Firms respond by increasing their innovative activities, leading to an increase in productivity. The effect is likely more pronounced for initially high-productivity establishments, which are typically closer to the technological frontier and more capable of innovating. Conversely, given an increase in tariffs, especially initially high-productivity firms should see a decline in their productivity as they now have a lower incentive to keep innovating.Footnote 20 Using Mexican establishment-level data and NAFTA as a case, Iacovone (Reference Iacovone2012) finds support for these channels. Especially frontier firms in innovative industries (defined as industries with high R&D intensity where process and more demanding product innovations are conducted) expanded their innovative efforts and productivity after liberalization.Footnote 21
DATA AND EMPIRICAL STRATEGY
Data
This paper leverages newly digitized data from the Historical Manufacturing Census of Sweden (Fabriksberättelserna), which provides detailed information on the individual performance of late-nineteenth-century manufacturing establishments. The data were collected through yearly questionnaires, which the state sent to the establishments.Footnote 22 Each establishment was mandated to return the completed questionnaire to local authorities. The data include a variety of variables, giving rich insights into the performance of establishments. Specifically, the data state the location, name, and industryFootnote 23 of individual establishments as well as the name, address, and title of their owner(s). The data also state the number of sales,Footnote 24 profits,Footnote 25 workers,Footnote 26 and whether the establishment used any power sources, such as steam engines.Footnote 27
As the Historical Manufacturing Census of Sweden was separately collected each year, the data do not contain an identifier tracing individual establishments over time. I use automatic record linkage techniques to link establishments into a yearly panel.Footnote 28 This procedure takes three steps and is further outlined in Reference Almås, Berger, Boppart, Burchardi, Ejermo, Eriksson, Larsson, Malmberg, Maukner, Olsson and OstermeyerAlmås et al. (forthcoming). First, I calculate Jaro-Winkler and Levenshtein edit distances for the names of establishments and their owners in each location-industry cell across all possible pairs of years. These edit distances measure the similarity of establishment and owner names within similar locations and industries over time. Since they are bound between zero and one (with the latter indicating that two names are identical), I can link two given observations across years if they provide a unique match. This is the case if the edit distances for either the name of the owner or establishment are larger than the specified threshold of 0.9, both establishments operate in a similar location and industry, and there is no other potential match according to these criteria. Finally, the resulting panel is manually checked and corrected if necessary.
I calculate my main outcome variable of establishment-level labor productivity as ln(Sales/Workers). As further discussed in the next section, my main sample includes all establishments active in 1891 and in at least one later year. By construction, I thus disregard establishments that entered after 1891.Footnote 29 I also drop observations where establishment-level productivity is not available.Footnote 30 As establishments can be active in multiple industries, I use the modal industry over their lifetime for classification. For a minority of establishments, this is not possible, and they are dropped.Footnote 31 Overall, my approach to constructing the main sample is similar to previous work by, for example, Atack, Bateman, and Margo (Reference Atack, Bateman and Margo2008).
I combine my establishment-level data with yearly industry-level Swedish tariff data by Persarvet (Reference Persarvet2019) that cover the same period. Using trade and price data for individual goods, Persarvet (Reference Persarvet2019) calculates unweighted nominal ad valorem tariffs, which I merge with my establishment-level data on an industry-year level basis.Footnote 32 Persarvet (Reference Persarvet2019) also calculates ERP for a broader industrial classification than the one used in this paper. As such, ERP cannot be matched to all industries contained in my establishment-level data and are mainly used in robustness checks.Footnote 33 To study the channels driving my results, I use two further data sources to develop two proxies for the relative innovativeness of establishments. First, I use the number of patents granted by the Swedish Patent and Registration Office (PRV) to identify relatively more innovative industries. The data are taken from Andersson, Berger, and Prawitz (Reference Andersson, Berger and Prawitz2023) and measure the industry-level number of granted patents between 1891 and 1900. Second, I rely on information from the Swedish population census to identify establishments that were owned by highly skilled individuals. These highly skilled individuals are defined as individuals with an HISCO code of either 1 or 2 and can be seen as relatively more professional managers.
Empirical Strategy
To study the impact of tariffs across establishments, I estimate regressions following Equation (1), where the outcome variable is establishment-level labor productivity calculated as ln(Sales/Workers) for establishment i in industry j and year t, and Tariff Rate jt measures the industry-level tariff. To examine whether tariffs impacted establishments heterogeneously, I interact the tariff rate with a categorical variable Initial Productivity ijt0, denoting the quartile of the industry-level productivity distribution into which a given establishment initially fell in 1891, before tariffs were increased in 1892.Footnote 34 This categorical variable ranges from the reference category one—denoting the initially least productive establishments—to four. Depending on the specification, I include a vector of control variables C ijt consisting of the industry-level growth rate as a proxy for the general industrial development, the industry-level Herfindahl–Hirschman index as a measure of domestic competition, and a dummy for whether an establishment exits in the next period.Footnote 35 I also include establishment-level and region-by-year fixed effects, accounting for time-invariant establishment-specific factors as well as temporal shocks common to all establishments in each region. I follow Chen and Steinwender (Reference Chen and Steinwender2021) by using two-way cluster-robust standard errors at the establishment and industry-year levels, accounting for correlation within establishments over time and across establishments in the same industry, as well as possible heteroscedasticity.
The coefficient β 1 in Equation (1) states the effect of a 1 percentage point tariff increase for the initially least productive establishments, while the coefficient β 2 shows how this effect changes with increasing initial productivity. If initially less-productive establishments benefited relatively more from tariff protection, as hypothesized, β 1 should be positive, but β 2 should be increasingly negative as initial establishment-level productivity increases.
One concern is that successful establishment-level lobbying for tariff protection could lead to reverse causality and bias the estimated coefficients. A priori, the direction of this bias is unclear (Amiti and Konings Reference Amiti and Konings2007). However, this is unlikely to be a large issue. First, it is questionable whether individual establishments have enough power to influence policymakers (Trefler Reference Trefler2004), and the results hold when excluding the initially most and least productive establishments, which arguably had the largest incentive to lobby.Footnote 36 To further rule out that tariffs followed previous productivity developments, I follow Topalova and Khandelwal (Reference Topalova and Khandelwal2011) by testing whether lagged productivity at the industry level predicts current tariff rates. Online Appendix Table A.5 shows no statistically significant relationship for different periods, which speaks against a potential endogeneity of tariffs.
Table 1 shows the summary statistics of the main sample used in the regressions. Establishments with an initially low productivity are smaller, as measured by the number of workers, sales, or profits compared to initially high-productivity establishments. Online Appendix Figure A.4 shows the number of active establishments included in the analysis. While establishment-level exit reduces the sample size over time, no large jumps are observed. About 4 percent of the observations denote establishment-level exits, which aligns with exit rates observed in Imperial Russia or in modern economies (Gregg and Nafziger Reference Gregg and Nafziger2024). Overall, the panel is well-balanced.
Table 1 SUMMARY STATISTICS
Notes: Summary statistics of the main sample used in the analysis. It includes establishments active in 1891 that survived for at least one year. Establishment exit (0/1) is a dummy taking the value one if an establishment is not observed in the next year. Note that this variable is missing for 1900, as this is the last year contained in the data set. See the section describing the data for more details.
Sources: Fabriksberättelserna.
RESULTS AND DISCUSSION
Main Results
Table 2 states the main results of this paper. I begin in Column (1) by estimating the effect of tariffs on establishment-level productivity based on Equation (1), but without the interaction term. Similar to the results at the industry level (Online Appendix Table A.4), the coefficient for tariffs has no large and statistically significant effect on establishment-level productivity. However, as discussed, this might be driven by establishment-level heterogeneity. Indeed, as shown next, tariffs had large, economically and statistically significant, but heterogeneous effects across Swedish establishments.
Table 2 THE EFFECT OF TARIFF CHANGES ON ESTABLISHMENT-LEVEL PRODUCTIVITY
Notes: The table shows regression results for variants of Equation (1). The outcome variable is establishment-level productivity. Productivity is calculated as ln(Sales/Workers). The dummies for initial productivity denote the quartile of the industry-level productivity distribution into which a given establishment fell in 1891. Two-way cluster-robust standard errors at the establishment and industry-year levels are stated in parentheses. *** p < 0.01, ** p < 0.05, and * p < 0.1.
Sources: Fabriksberättelserna and Persarvet (Reference Persarvet2019).
To motivate that tariffs impacted Swedish establishment-level development heterogeneously, I begin by estimating Equation (1) separately for each industry. As each establishment within a given industry faced the same tariff rate, this effectively amounts to estimating the effect of tariffs on establishments with initially higher levels of productivity relative to the initially least productive establishments. Figure 3 shows the results for each industry using ERP as a proxy for the industry-specific tariff rate. Following a tariff increase, productivity declined as hypothesized among initially more productive establishments relative to the initially least productive establishments.Footnote 37
Figure 3 THE EFFECT OF ERP ON ESTABLISHMENT-LEVEL PRODUCTIVITY BY INDUSTRY
Notes: I separately estimate Equation (1) for each industry using ERP. The figure shows point estimates and 95 percent confidence intervals for each quartile of the initial establishment-level productivity distribution relative to the initially least productive establishments. Establishment-level productivity is calculated as ln(Sales/Workers). I include establishment and region-by-year fixed effects but no additional control variables and cluster the standard errors at the establishment level.
Sources: Fabriksberättelserna and Persarvet (Reference Persarvet2019).
The main results of this paper are reported in Columns (2) and (3) of Table 2, where I estimate Equation (1) using data across all industries. Figure 4 plots the corresponding average marginal effects. Irrespective of the included control variables, the results show that tariff changes had economically and statistically significant, but heterogeneous, effects across Swedish establishments. While an increase in tariffs is associated with a rise in productivity among initially low-productivity establishments, the initially high-productivity establishments saw a relative decline, and the remaining establishments noticed changes between these extremes. Specifically, a 1 percentage point tariff increase is associated with an increase in productivity of about 3 percent for the initially least productive establishments, whereas the initially most productive establishments saw a relative decrease in their productivity of 3 percent. As discussed in the introduction, these are statistically and economically significant effects.
Figure 4 THE EFFECT OF TARIFF CHANGES ON ESTABLISHMENT-LEVEL PRODUCTIVITY
Notes: The figure visualizes the average marginal effects of increasing the tariff rate by 1 percentage point using the specification presented in Column (3) of Table 2. The figure shows point estimates of the main and interaction effects and 95 percent confidence intervals.
Sources: Fabriksberättelserna and Persarvet (Reference Persarvet2019).
Robustness Checks
This section documents the robustness of the main result, whereas the next discusses potential channels. The main robustness checks are presented in Table 3, where I begin by using different proxies for establishment-level productivity. Since ln(Sales/Workers) could be an imprecise measure of establishment-level productivity, I use in Columns (1) and (2) TFP Footnote 38 and ln(Profits/Workers) Footnote 39 as outcomes. Notably, the results stay robust and, if anything, increase. This also holds when using ERP instead of nominal tariffs in Column (3)Footnote 40 and when altering the econometric specifications in the remaining columns. To show that attrition is not driving the results, I use in Column (4) a balanced sample that only includes establishments observed in each year between 1891 and 1900.Footnote 41 As the historical background outlines, Sweden increased its tariffs in two steps. Since manufacturing tariffs were mainly increased in 1892 (Bohlin Reference Bohlin2005), I base my main sample on all establishments active in 1891 and afterward. Column (5) shows that this choice does not drive the results, as they are similar when using the first period of tariff increases. Here, establishments active in 1887 and afterward are identified, and 1887 is used to determine their initial productivity. Since the main approach assumes that tariffs impact establishment-level outcomes in the same year as they are changed, I allow for a dynamic impact by lagging tariffs by one year in Column (6). Again, the results remain similar.
TABLE 3 ROBUSTNESS CHECKS OF THE EFFECT OF TARIFF CHANGES ON ESTABLISHMENT-LEVEL PRODUCTIVITY
Notes: The table shows regression results for variants of Equation (1). The outcome variable is establishment-level productivity unless otherwise specified. Productivity is calculated as ln(Sales/Workers) unless otherwise specified. The dummies for initial productivity denote the quartile of the industry-level productivity distribution into which a given establishment fell in 1891. These dummies are calculated based on profits instead of sales in Column (2). In Column (9), I use the deflated value of sales to determine these dummies. Two-way cluster-robust standard errors at the establishment and industry-year levels are stated in parentheses. *** p < 0.01, ** p < 0.05, and * p < 0.1.
Sources: Fabriksberättelserna and Persarvet (Reference Persarvet2019).
TABLE 4 MECHANISMS: THE EFFECT OF TARIFF CHANGES ON INNOVATION AND PRODUCTIVITY
Notes: The outcome variable is establishment-level productivity calculated as ln(Sales/Workers). The dummies for initial productivity denote whether an establishment’s labor productivity was above the median of its industry in 1891. The titles in each column of the table state the variable included in the triple interaction. In Column (2), the innovative sector consists of those industries that have relatively more patents per establishment than the median industry (industries 4, 6, 9, 10, 11, and 12), and in Column (3), the industries in the fourth percentile (industries 6, 10, and 11) between 1891 and 1900. Figure 2 states the name of each industry. Engineer = 1 indicates establishments where at least one owner was an engineer at any point between 1891 and 1900. Skilled Owner = 1 denotes establishments where, for at least one observation between 1891 and 1900, an owner had a HISCLASS code of 1 or 2. Note that this regression only includes establishments that were successfully matched to the census and reported an HISCLASS value. Two-way cluster-robust standard errors at the establishment and industry-year levels are stated in parentheses. *** p < 0.01, ** p < 0.05, and * p < 0.1.
Sources: Fabriksberättelserna and Persarvet (Reference Persarvet2019).
Until now, I have determined the initial productivity level of establishments using a dummy. While this eases the interpretation, it is questionable whether establishments close to the borders of the respective quartiles are very different. Similarly, grouping establishments from different industries into one dummy might lead to a mix of heterogeneous establishments; although this is likely a minor issue, as discussed previously. Similar to Chen and Steinwender (Reference Chen and Steinwender2021), I interact the initial productivity level of establishments with the tariff rates and run the regression in the first difference in Column (7). Again, this transformation does not affect the result. Two further potential threats to identification are regression to the mean effects and outliers. However, the results remain similar when I winsorize the sample in Column (8) by excluding the initially most and least productive 5 percent of establishments and recalculating the dummies denoting initial productivity.Footnote 42 Another issue is that price changes over time could distort the value of sales and, thus, my measure of productivity. This is unlikely to be the case, however, as the analyzed period is not very long, and the results remain similar when deflating the value of sales with an index taken from Ljungberg (Reference Ljungberg1990) in Column (9). Furthermore, initial establishment-level size rather than productivity could drive the findings. Online Appendix Table A.6 rules this alternative explanation out, however.
While the findings are robust, further unobservable time-varying variables can still bias the estimates. Oster (Reference Oster2019) develops a method to assess how large the selection on unobservables would need to be to drive the observed effects to zero.Footnote 43 Intuitively, the method assesses how stable the coefficients and R-squared values are when moving from a model that includes control variables to one that excludes them.Footnote 44 This is done by calculating a value δ i for each coefficient β i. A value of δ i > 1 suggests that to invalidate the findings, selection on unobservables must be stronger compared to selection on observables. Online Appendix Table A.7 shows that this holds in all cases, so that unobservables likely do not bias the estimated coefficients.
Online Appendix Table A.8 shows that the results also remain robust when using two different dynamic models. The first is based on fixed effects estimates that include the lagged dependent variable, whereas the other uses a two-step system GMM estimator.Footnote 45 I specify the system GMM estimator by including two lags of the dependent variable to ensure that there is no second-order autocorrelation. Therefore, I include (if available) observations from the years 1889 and 1890. Year fixed effects are treated as strictly exogenous regressors and are included in the level-equation of the estimator. All other variables are included in the first-difference regression.Footnote 46 Online Appendix Table A.8 suggests that the model using the two-step system GMM estimator is well-specified, as I find no signs of second-order autocorrelation, and the insignificant Hansen test in Column (4) suggests that the overidentifying assumptions are valid. Overall, the estimated coefficients are larger than those reported in Table 2 and are statistically significant. If anything, this suggests that the main estimates are a lower bound.
Finally, I use similar regressions as before to study how tariffs affected other establishment-level characteristics (Online Appendix Table A.9). The results are similar when using sales and profits as outcome variables. Again, this suggests that tariffs had a differential impact on the growth and development of Swedish manufacturing establishments, which is not apparent at the aggregate level. For workers, the effect is much smaller and opposite to the previously documented results, suggesting that tariffs affected productivity by influencing the output rather than the hiring practices of establishments. This increases the plausibility that changes in the innovative efforts of establishments underlie the findings, as discussed in the next section.
Mechanisms
TARIFFS SHAPED ESTABLISHMENT-LEVEL INNOVATION STRATEGIES
The previous section shows that tariffs impacted establishments heterogeneously. Following an increase in tariffs, initially low-productivity establishments saw a relative increase in their productivity, while initially high-productivity establishments saw a relative decline. This section argues that the observed dynamics are consistent with the Schumpeterian and escape-competition effects discussed earlier. Because I do not directly observe the innovation strategies pursued by establishments, I use an indirect strategy following Iacovone (Reference Iacovone2012) to show that my findings are consistent with the two channels. The intuition behind this strategy is that if the Schumpeterian and escape-competition effects are relevant explanations for my findings, the results should be stronger among relatively more innovative establishments. Using different proxies for the relative innovativeness of establishments, Table 4 shows this is indeed the case.
I begin in Column (1) of Table 4 by running a similar regression as before, except that I collapse, for ease of interpretation, the four levels of initial productivity into two. Next, similar to Iacovone (Reference Iacovone2012), I use triple interactions to show that the results are stronger for relatively more innovative establishments, which is taken as indirect evidence that tariffs affected establishments by changing their incentives to innovate.Footnote 47 The first proxy for the relative innovativeness of establishments comes from industry-level patent data. In Column (2), I identify innovative industries as those industries with more patents per establishment than the median value across all industries between 1891 and 1900. In Column (3), I increase this threshold to the 75th percentile. Incidentally, these relatively more innovative industries correspond to the emerging industries of the 1890s, as discussed in the historical background. As hypothesized, the results are stronger in relatively more innovative industries, particularly when using the stricter second definition. While initially low-productivity establishments in innovative industries saw an additional increase in productivity following a tariff increase, there was an additional decrease among the initially high-productivity establishments.
I substantiate these results by directly identifying relatively more innovative establishments from my data. In Columns (4) and (5), I specifically focus on establishments that were owned by engineers or highly skilled individuals. The intuition behind this strategy is that engineers and skilled owners are presumably in a relatively better position to successfully lead innovation strategies at the establishment level. Therefore, the results should be stronger for these subsets of establishments. Again, this is the case and is consistent with the Schumpeterian and escape-competition effects.Footnote 48
ALTERNATIVE MECHANISMS
The other channels discussed—input tariffs and tariff retaliation—likely do not explain my findings as the results hold when using ERP as previously stated, and there is no evidence of extensive tariff retaliation during my period of analysis (Persarvet Reference Persarvet2019, pp. 80–84, 101–2, 181–92). Moreover, several other explanations for the results can be ruled out. First, shocks to individual regions likely do not drive the findings, since I include region-by-year fixed effects. Second, instead of establishment-level learning, changes in the establishment-level capital stock can also increase productivity (Iacovone Reference Iacovone2012). As a crude proxy for the effects on establishment-level capital, I analyze whether tariffs differentially shaped the probability of establishments to adopt steam power, which was the key technology of this time (Crafts Reference Crafts2004). The results in Online Appendix Table A.9, Column (4), speak against the presence of this channel.
CONCLUSION
Studies on the historical impact of import tariffs on economic growth provide inconclusive results. I argue that this is at least partly driven by the use of aggregate data. In contrast, firm-level data can help us better understand the channels through which tariffs affect economic development. Leveraging detailed establishment-level data and using Sweden’s shift back into protectionism during the late nineteenth century as an example, I show that initially low-productivity establishments profited from tariff protection, whereas initially high-productivity establishments experienced a relative decline. Consequently, these opposing outcomes help explain why tariffs had no impact at the aggregate level. In addition, I provide suggestive evidence that the results are driven by tariffs differentially shaping the incentives of establishments and managers to innovate and learn to organize production efficiently. Higher import tariffs reduced the incentives to innovate and learn for initially more productive establishments, but increased them for initially low-productivity firms. More broadly, the findings suggest that heterogeneous establishment-level responses to tariff changes contribute to the contradictory findings in the historical tariff-growth literature. Studying the long-run effects of these tariff changes provides fruitful avenues for future research.
Open access
