A reassessment of industrial growth in interwar Turkey through first-generation sectoral estimates

Abstract This study presents the first sectorally disaggregated estimates of the industrial output growth for Turkey between World War I and II. These estimates indicate that at the aggregate level the existing official index overestimates the output growth. Secondly, the sectoral disaggregation shows that the industrial growth was balanced, as both textiles and food-processing branches, which comprised most of the value-added, grew significantly. Local industries expanded against the only modest gains in per capita consumption of manufactured goods and incomes. Output growth was positively correlated with higher initial import penetration and nominal protection rates, which implies that trade protectionism helped favorable relative prices induce domestic expansion. On the other hand, both import-competing and domestic-market-oriented sectors significantly expanded, which suggests that import repression and increasing domestic demand drove industrial growth.


Introduction
The conventional wisdom holds that the Ottoman economy that Turkey inherited after World War I underwent a significant de-industrialization during the second half of the nineteenth century, despite the establishment of a number of modern factories and the resistance of handicrafts in the final decades of the Ottoman Empire. 1 Yet, there were countries in the periphery at the time which witnessed the emergence of modern industry and even outperformed the industrialized countries as far as output growth was concerned (O'Rourke and Williamson 2017). In other words, the industrial catch-up of the peripheral economies was already underway before 1914. Yet, the Ottoman economy was not part of that process.
The subsequent performance of Turkey's economy in the interwar period was more than remarkable. 2 In the 1920s, it was a large agricultural economy, greatly reflecting the initial economic and demographic shocks of WWI and the following armed struggles. Close to the European markets, Turkey was also exposed to external shocks. The domestic market consisted of loosely connected local markets with relatively high commercial agriculture in the west and self-sufficient rural economies in the rest. The 1920s saw a rapid recovery in this environment due to population growth. The period was succeeded by the adverse shocks of the Great Depression on incomes and, finally, the recovery-growth cycle in the second half of the 1930s. By the end of the 1930s, per capita incomes exceeded the 1913 levels, which was a remarkable achievement considering the massive effects of the war and the unfavorable global economic environment after the Great Depression. Overall, the interwar period eventually saw massive railroad expansion, modern factory production funded mainly by the government, and the decreasing importance of foreign trade. 3 Even though most peripheral economies outperformed the industrial core (the US, the UK, Germany, France, and Japan) in the 1930s, Turkey stood out even within this group. As explained in detail below, the current estimates show that Turkey's industry was one of the best performers globally, with a compound growth rate of around 9.2 percent annually between 1923 and 1938. Figure 1 plots the GDP per capita growth and industrial output growth at constant prices between 1920 and 1938 for a handful of countries comparable to Turkey. It clearly shows that Turkey saw the highest per capita income growth, and its industrial output growth was one of the highest within the group.
Conventionally, Turkey's industrial take-off is dated back to the 1930s. The received wisdom is that Turkey achieved a big push in industrial development thanks to strong protectionism and state entrepreneurship (etatism) in the aftermath of the Great Depression (Boratav 1988). However, this view is only based on the aggregate industrial output estimates provided by Bulutay, Tezel, and Yıldırım (1974) and additional scattered evidence, mainly representing the government sector. Altogether, these estimates indicate a fast output growth, around 8-9 percent, one of the fastest in the world at the time. There are, though, two significant shortcomings of these estimates. First, they are biased by construction, as they primarily represent the modern factory output rather than the whole industry. Secondly, the lack of sectoral disaggregation prevents a more refined understanding of the drivers of industrial expansion in the interwar period.
To fill that empirical gap, this article presents the first sectorally-disaggregated estimates of industrial output and consumption of manufactured goods for 1925-1939. Referring to the relevant methodological and empirical challenges faced by the economic historians working on British industrialization during the nineteenth century and Italian industrialization after unification, I present a detailed discussion of the construction of the sectoral output measures and discuss their broader implications regarding the post-Great Depression economy of Turkey. In particular, I greatly benefited from the late Stefano Fenoaltea's methodological contributions to the statistical reconstruction of the output growth of post-unification Italian industry. 4 The present estimates are correct for the overestimated output growth in Bulutay, Tezel, and Yıldırım (1974). However, I show that the estimated average growth, around 6.8 percent annually, put Turkey among the best performers at the time, consistent with the cross-country evidence of the output growth at the beginning of industrialization in the periphery. I also argue that Turkey's performance was consistent with the massive change in relative prices in favor of industry globally and the resulting deterioration of the country's terms of trade. There was a clear positive correlation between initial import penetration and subsequent growth. Moreover, the aggressive import protection also helped the importcompeting sectors to grow, such as textiles. The estimated nominal protection rates indicate a positive relationship between tariffs and average growth.
In what follows, Section "Historical background and the existing aggregate output estimates" outlines the shortcomings of the existing estimates of output growth in interwar Turkey. Section "Methodology of the present estimates" discusses the methodology of the present estimates, while further details for sectoral series can be found in Data Appendix. Section "Interpretation of the new estimates" presents an interpretation of the present estimates regarding peripheral industrialization in the interwar period and protectionist policies.

Historical background and the existing aggregate output estimates
The late nineteenth century is traditionally portrayed as having witnessed the integration of the Ottoman economy globally via commercial agriculture in the west, helped by the foreign-funded railroad and port construction, and the gradual decline of some of the traditional handicrafts in the face of foreign competition. World War I and the subsequent war against the occupation devastated that predominantly agricultural economy until 1923, when the new state was founded, due to significant political, economic, and demographic shocks. Against this backdrop, economic nationalism and protectionist zeal were in currency in the 1920s due to the memories of the empire's disintegration. 5 For instance, the mood in favor of active industrial policy was visible in the policy framework announced in the Izmir Economic Congress in 1923, the first venue of the new elites' economic policy. But the policy options were limited until 1929 because of the Lausanne Treaty of 1924, which placed sanctions on import policies. Thus, the industrial policy remained restricted to some nationalizations, such as in sugar and tobacco processing and the introduction of credit programs in the 1920s.
The end of the treaty's sanctions and the Great Depression made 1929 a turning point in terms of economic policy in Turkey. The government quickly made a tariff reform increasing the effective and nominal rates on manufactured goods. In parallel, the economy faced severe terms of trade shock due to the decline in relative prices of agricultural goods, making up for most of the export revenues. The government, first, aimed to ease the balance of payment problems via currency controls and import repression. However, the initial ad hoc import repression policies became more systematic after 1932 with the preferential trade agreements. The imports of manufactured goods became increasingly difficult with the prohibitive tariff and quota policy. Another significant policy shift of the 1930s was the increasing state entrepreneurship. The first five-year plan, a list of investment projects rather than an elaborate planning document, was formulated with the help of Soviet experts and started to be implemented in 1934. From then on, the state rapidly emerged as an investor and entrepreneur in crucial sectors, such as textiles, leather, tobacco, sugar, and mining. During the 1930s, around 20 large state enterprises were established, and they began to assume an increasingly prominent role. The industrial activism of the government can also be seen in the data environment at the time. In line with the expanding official statistical capacity, the first industrial census was held in 1927; and the central statistical office published annual summary reports on the big manufacturing firms and output data on the government sector after 1932. Despite the scattered quantitative data on the industrial sector, there did not exist any systematic account of industrial activities for decades. Bulutay, Tezel, and Yıldırım (1974) provide the only available aggregate output estimates for the industry of interwar Turkey (1923)(1924)(1925)(1926)(1927)(1928)(1929)(1930)(1931)(1932)(1933)(1934)(1935)(1936)(1937)(1938)(1939). These estimates were produced as part of the extensive reconstruction of national income (agriculture, industry, and services). Although that was a private study, Turkey's statistical agency (State Institute of Statistics-SIS) adopted their series and integrated it with the official GDP estimates for the subsequent period. Their methodology of estimation follows two stages. First, for 1932-1939, they extrapolate backward the official output estimates of the 1940s, using the output of the enterprises that benefited from the Law for Promotion of Industry as a proxy. 6 The official summary statistics present the total production of the beneficiary industrial enterprises at current prices, so they use this index to extrapolate the official aggregate output figure. Those enterprises, which numbered around four-five thousand at the time, had to meet specific criteria, such as hiring at least ten employees to get subsidies. That practically means that they represented almost all mechanized factory production.
On the other hand, handicrafts, or small-scale manufacturing, did not benefit from the law. It is hard to measure the precise value-added share of handicrafts from the 1927 industrial census, though, by looking at the number of employees per enterprise in 1927, a reasonable guess would put it at more than half. Therefore, that creates a considerable selection bias in Bulutay's methodology. In the second estimation stage, the authors draw on transaction and income tax data as the proxy for 1923-1931. However, no sectoral breakdown of tax data is available. Therefore, it is far from obvious how well the tax data represents the industrial output. Thus, overall, the Bulutay series possibly overestimated the growth rates for 1932-39 due to possible higher growth rates of mechanized production, while it is not easy to gauge the direction of bias created by the tax series.
Subsequently, the only attempt to revise Bulutay estimates came from Zendisayek (1997), who suggested a correction by taking them as representing the factory output and combining it with a separate index for handicraft output. The handicraft series were obtained from Eldem (1947). Though it sounds like a reasonable improvement, the underlying handicraft index is no more than a simple guess, as Eldem neither explains his methodology nor specifies his sources. My extensive research of the primary sources implies that it is nearly impossible to measure the handicraft output separately because it was almost undocumented. Secondly, Zendisayek obtains 1938 weights from the official estimates in Die (1954). Her weighting procedure inherently overestimates growth rates since she applies these fixed weights of the factory and handicrafts output, produced for the year 1938, to the whole interwar period, even though factory output and its share of value-added probably grew faster than handicrafts. By reducing the share of factory output in favor of handicrafts for the 1920s, one plausibly comes up with growth rates of around 3-4 percent per annum, smaller than her estimate. In other words, due to 1938 weights taken as constant for the whole period, the factory output is over-represented in Zendisayek's common index.
Besides those problems, these two indexes do not provide any insight into the composition and structure of industrial growth due to the lack of sectoral disaggregation. At best, they are proxy estimates for the aggregate industrial value added. However, given a large amount of scattered quantitative evidence, it is plausible to conduct a detailed study and develop more accurate disaggregated output series.

Methodology of the present estimates
The census data are the primary data source on production, value-added, or employment, typically when the institutional capacity of data production is limited, as was the case in interwar Turkey. The first industrial census in Turkey was carried out in 1927, and the second in 1950. The coverage of the 1927 census is broad enough, as it covered both small and large enterprises. Only home production was excluded from the census. Population censuses were undertaken more often: 1927, 1935, and every five years after that. Among them, the 1935 census is particularly useful since it gives information about the employment composition within the industry, thus making it possible to compare the post-1929 situation with the previous one. Furthermore, direct output data are available for the sectors where the government exercised strict supervision or had a monopoly, such as tobacco, alcohol, salt, mining, and utilities. As for textiles and food-processing sectors, one finds some direct, yet fragmentary sectoral evidence, although its use requires care and caution.
Any attempt to estimate output growth based on such imperfect evidence must be imprecise and precarious. For several countries, like the UK and Italy, the first systematic quantitative estimates of industrial output, "first-generation" estimates in late Fenoaltea's terminology (Fenoaltea 2003), were later improved by successive better estimates. The level of imprecision in new estimates typically is reduced thanks to new evidence and better methods. The attempt to measure British industrial growth before the mid-nineteenth century is a good case. Commenting on the history of British industry, Harley (1982, 272) remarks "the data for this period are imperfect, and so any estimates of growth are controlled conjectures." 7 Hoffman's (1955) first-generation index provided the basis for understanding the extent of the British industrial revolution. The second-and third-generation indexes presented by Deane and Cole (1962), Crafts (1983), Harley (1982), and Crafts and Harley (1992) all used similar or reconstructed versions of the basic sectoral series that Hoffman and later Deane and Cole introduced. The sectoral series in those works are based either on the relevant input indices, which best reflect output growth in the corresponding sectors, or appropriate proxies. For instance, the cotton industry is represented by retained cotton imports, the wool industry by the sum of the estimated domestic clips and imports, clothing by a weighted average of textile output, and the whole food and drink industries by a transformation of population growth with an index of milling and baking. Crafts and Harley's previous papers improve the choice of weights rather than the basic series provided mostly by Deane and Cole. Another case in point is the Italian industries before World War I. 8 Given the data limitations, Gerschenkron's (1962) estimates of industrial growth before 1913 mostly used input consumption as proxies. For instance, the growth of the silk industry was measured by raw silk output, cotton by the net imports of cotton, and flour milling by wheat consumption. His series was later to be significantly improved. In his comprehensive revision of Italian industrial growth before 1914, Fenoaltea (1988Fenoaltea ( , 2000Fenoaltea ( , 2001Fenoaltea ( , 2002 masterfully uses a different technique. They reconstruct textile series by considering different stages of production, controlling for foreign trade and lags between output transformations in each stage of production. This procedure causes the estimated variation of final output across time to depend on the underlying input data and the changes in foreign trade of various intermediate goods. Thus, his method captures the effects of foreign exchange on output at different stages of production. For instance, the growth of cotton spinning, and weaving differs when weaving develops much faster than spinning due to cotton yarn imports. More generally, if one country specializes in a particular stage of a specific industry, then the raw material consumption fails to reflect the growth of the whole industry. In this sense, Fenoaltea's method successfully distinguish the intermediate production from that of final goods.

Individual sectoral series
This section outlines the construction of physical output proxies and measures for individual sectors, which I will combine into aggregate indexes in the subsequent stage. Overall, the physical output series estimation follows the procedure outlined in Figure 2. I first obtain the sectoral physical output estimates using the underlying data such as agricultural, industrial, and foreign trade statistics and censuses. At this stage, I have both the output series representing the final output for each sector and the elementary series that measure raw material and intermediary consumption. To combine the sectoral output series, I obtained the fixed value-added series from the 1927 industrial census in the second stage. Therefore, by combining the final series with the weights, I get the final aggregate output series, representing the movement of the total industrial value added.
Overall, the present reconstruction of industrial output growth aligns mainly with the first-generation output estimates in methodology and the nature of data sources. Referring to the "principles" of first-generation reconstructions of national accounts (Fenoaltea 2010; i) some of the present sectoral series are relevant proxies for input consumption; (ii) the known series represent the unknown series in most cases; and (iii) for aggregation, I use physical output indices with fixed value-added weights. However, to make further refinement just as described in Fenoaltea's textile series, I put considerable effort into improving the elementary series by measuring the traded intermediate goods, as in the case of textiles and leather processing. I measure yarn and cloth output separately, by keeping an eye on net imports as much as the trade data allows. Another point where the present estimates differ from the first-generation estimates is that I provide a more refined estimate of woolen goods by making a separate estimate of carpets and rugs instead of simply taking the available greasy wool output. The need arose because carpets and rugs absorbed a large part of raw wool consumption, and the processing was different for the coarse types needed for carpets and other goods. Also, carpet sales significantly declined in the interwar years because of the contraction in export demands. Therefore, to capture the effect of the massive collapse of the carpet market, I first derived a carpet demand index from the cotton cloth consumption and relative prices and was able to deduct the total greasy wool needed for carpets. 9 Moreover, as carefully as possible, I cross-checked the sectoral series with available, albeit scattered, qualitative and quantitative evidence for each sector. For instance, in the case of cotton, as the local data sources on raw cotton output conflict with the official aggregate figures, I reconstructed raw cotton output from scratch based on local data sources. Also, the known series here represent the unknown series in a specific way: the known food-processing series represents the unobserved food industries, and the observed textiles series the unobserved textiles, and then the estimated aggregate food-processing and textile series, together with others like utilities and mining, altogether represent the remaining unknown series like chemicals and metal making. Table 1 thus presents the sectoral measures, the details of which can be found in the Data Appendix. For instance, briefly sketching, I measure cotton textiles output with cotton yarn consumption. The cotton yarn consumption is, in turn, estimated based on the raw cotton output, net of foreign trade, and the net exports of the cotton yarn. I apply a similar procedure to other textiles. In other words, the output series in textiles and leather processing are reconstructed starting from raw material consumption. The net consumption of yarn and processed leather is used as the final output series. As for the olive oil, tobacco, alcohol, sugar processing, utilities, and mining sectors, I  Building iron and cement consumption use direct output estimates, either constructed from several sources or obtained from secondary sources. For instance, as explained in the Appendix, the direct data on the olive oil output is available in several sources. The production in some sectors is well-documented by official sources because of government monopolies or strict supervision. Tobacco, alcohol, and sugar processing are prime examples. Simple proxy measures must be preferred for food processing since it is hard to find systematic evidence about that sector. Although it constituted almost 40 percent of total industrial value-added in 1927, it was the most poorly documented branch. The small, traditional, and family-based enterprises involved in the sector diffused throughout the country, except the ones requiring large-scale capital investments like tobacco and alcohol processing. Therefore, in the present work, the wheat consumption represents the milling output and timber consumption the woodworking.
Finally, after estimating the output measures for individual sectors, I calculated the implied consumption levels from the output estimates to cross-check the validity of these output estimates by adding the net imports of final goods. I assume that consumption can be considered a function of incomes, relative prices, and their elasticities, thereby providing a ground to judge the validity of output estimates.
As usual with first-generation estimates, the present series contain a significant potential error. Besides the unavoidable crudeness of proxies, two additional problems stand out: First is the failure to consider technical change over time. This bias might be significant, particularly in the increasingly modernized sectors, such as cotton textiles, where technical change was supposedly more rapid. Secondly, I cannot sufficiently observe inventory changes due to the absence of data. In many cases, I take the three-year moving averages as a second-best solution, which leads to a downward or upward bias in year-to-year fluctuations, but not necessarily in trends.
To illustrate the estimated trends in output and consumption in a few sectors, Figures 3 and 4 show the cases of cotton cloth (the most widely used fiber in textiles) and food processing branches. The local cotton fabric output was stable at around 15 thousand tones until 1931 yet increased very fast until the war up to 45 thousand, all the while the fabric imports declined by about one half. The resulting per capita consumption had a U-shaped curve (ending at 3 kg per person in 1939). As a result, the per capita consumption barely exceeded the pre-Depression years by the end of the 1930s. However, the composition of consumption moves swiftly in favor of domestic output. Only high-quality fine cloths continued to be imported, while the domestic producers succeeded in providing most coarse fabrics before the war. This example well characterizes the nature of industrial growth in the 1930s: fast output growth was primarily due to the import replacement, and at that easy stage of import substitution, local production quickly replaced the imports of the cheap coarse product segment. Figure 4 presents the estimated variables in food processing. The case of refined sugar resembles the textiles series with the U-shaped per capita consumption and rapid increase in local output (from nil in 1925 to 60 thousand tons in the mid-1930s). Sugar took great attention from policymakers at the time, as it was one pillar of industrial policy. Tobacco and milling output increased to more modest levels, and the trend in olive oil is not discernible. Tobacco and vegetable oil production was already well established before the Depression, and the import volume did not change much. All these estimates give a consistent picture: The output growth varied from one sector to another, but there was substantial expansion in all branches. Moreover, the per capita consumption levels either stagnated or increased moderately.

Weighting and aggregate series
Once the individual sectoral output series are obtained, I combine them into an aggregate output index, which requires a sound weighting system. Given the lack of representative price data, the best alternative is to use a constant-weighted index via sectoral value-added shares. I derive these shares from value-added and employment data from the 1927 industrial census. The value-added composition of  Table 2 appears as one expects from traditional industry. Cereal milling accounts for almost one-fourth of the total value added, followed by tobacco, leather processing, and alcohol. These figures make sense since in the 1920s, these sectors had relatively established the tobacco and leather industries. Most modern capital-intensive sectors, such as iron, cement, glass, and chemicals, were mostly absent as they would be built after the war.

in
I adjust the weights of 1927 for the 1930s due to the significant variation between sectoral growth rates over time. In the absence of reliable data on employment at the present sectoral disaggregation, I use the movement of sectoral output and prices to track down the change in relative value-added in each sector. 10 Table 2 indicates a significant realignment between 1927 and 1935. Cereal milling remains the largest sector, yet its share declined from 25.5 to 18.9 percent, and it continues to be followed by tobacco processing, whose share decreased slightly. The relative decline of food processing took place in favor of textiles altogether. To compare the present series with Bulutay estimates, I derive the annual total value added at 1927 prices by extrapolating sectoral value added by quantity relatives, as suggested by Fenoaltea (2003). I estimate that the total value-added increased from about 200 million TL up to 500 million in 1939, both in 1927 prices ( Figure 5). My estimates diverge from Bulutay's after 1932, which has slower growth rates. On the other hand, Zendisayek's estimates seem much lower than both others. The main difference between Bulutay estimates and the present ones is that the former entirely depends on the official industrial statistics, which are biased toward factory production.    2 also shows the estimates of the time trend for each sector. It seems that all sectors except carpetmaking witnessed a significant trend growth, with sugar processing the highest. The latter, however, should be considered in connection with the local refined sugar production, which was extremely low at the beginning and remained strongly protected during the whole period. Secondly, the textile growth rates were higher than most other sectors.

Overall performance
How does the overall performance of Turkey's industry compare globally in interwar years? O'Rourke and Williamson (2017, 16) summarize the most recent findings on cross-country industrial performance since the late nineteenth century. Their figures forcefully show that some countries in the periphery outperformed the industrial leaders starting from the late nineteenth century. However, the Middle East, i.e., the Ottoman economy at the time, fell behind other developing regions before 1914. 11 Yet, in the interwar years, most countries in Asia and the European Periphery significantly outperformed the industrial core, while Latin America's performance was more moderate. Against this more extensive background, Turkey's interwar performance, with about 6.5 percent estimated annual growth, appears to be one of the highest at the time. However, since the initial output levels were relatively low, we can more suitably compare the high growth rates with the pre-1914 performance of some other developing economies. Turkey's interwar experience, therefore, is more consistent with the initial industrial take-off of, for instance, Austria, Hungary, Russia, Chile, Brazil, Argentina, and Mexico from 1870-1896, which all witnessed 5 percent or higher annual growth. (O'Rourke and Williamson 2017, 3) As for the Middle East, the estimate for Egypt, the only country for which systematic evidence exists, shows that it was among the best performers in the interwar period, albeit doing worse than Turkey. Yet, Egypt's industrial growth depended entirely on the textile sector, as the other sectors were mainly stagnant (Karakoç 2018).
Crucially, the acceleration of the industrial catchup of the peripheral economies in interwar years happened in the face of the global Depression and stagnant incomes. Although it sounds puzzling, the economic mechanisms behind it are not unknown (Feinstein, Temin, and Toniolo 2008). Two parallel forces were at work: First, the decline in agricultural prices was more pronounced than in industrial prices globally. Therefore, that stimulated resource allocation to shift away from agriculture and in favor of industry. Secondly, for the primary producers in the periphery, this relative increase in industrial prices also meant deterioration in terms of trade, as the exports consisted overwhelmingly of agricultural products, while imports were primarily manufactured goods. We know that the terms of trade booms had associated with higher gains from trade and de-industrialization in the periphery in the late nineteenth century (Williamson 2011). 12 Similarly, the interwar crisis significantly lowered the gains from trade while increasing the prospect of industrial development at home, consistent with the patterns in the previous era of the world economy.
On the other hand, the industrial performance in the periphery was not uniform globally. The immediate responses to the Great Depression consisted of a large spectrum of ad hoc policies, such as tariff and non-tariff protection and exchange rate controls, which initially aimed to reduce the current account balance and trade deficit (Lewis 1953). Yet, starting with the tariff reform of the UK in 1931, policy activism quickly spread to the periphery and constituted the policy paradigm during the decade. The now dominant view of the Great Depression (Temin 1989;Eichengreen 1996;Eichengreen and Sachs 1985;Eichengreen and Irwin 2010;Campa 1990) holds that countries resorted to the devaluation or monetary expansion after going off Gold in the early 1930s to stimulate the domestic demand. On the other hand, in the countries where such policies were not implemented (due to lack of policy capacity or by choice), trade protectionism helped to divert the demand toward local products. Therefore, Turkey's policy choice, heavily dependent upon import restrictions in the 1930s, was relatively well predicted by the persistent orthodox monetary and fiscal policies (balanced budget, tight money, and overvalued exchange rate). 13 Overall, protectionism meant a highly skewed price structure against the agricultural sector. Boratav (1988) also agrees that industrialization in the 1930s was financed by agriculture, where peasants had to endure low agricultural prices.

Initial import penetration and growth
Therefore, the change in relative sectoral prices, the terms of trade, and protectionism operated to the effect that domestic production replaced imported manufactured goods in Turkey. However, these two factors had varying effects within the industrial sector. The present sectoral disaggregation of industrial output helps to shed light on those varying effects. First, the scope of import substitution was determined by, sectorally, the initial level of import penetration. For instance, the share of locally produced fabrics in total consumption in the mid-1920s was as low as 41 percent for cotton and 24 percent for silk goods. However, import penetration was relatively low in most food processing branches. Sugar was the most extreme case, as almost all refined sugar was imported in the mid-twenties. Therefore, sugar and cotton received significant government attention, so much so that the first government factories, built after 1933 as part of the five-year plans, specialized in those sectors.
Moreover, some sectors like tobacco processing were already well-established, serving the domestic market without much foreign competition, and even had some export capacity. Figure 6 shows the negative relationship between the share of local output in total consumption and the trend growth rate from 1925-1939. Sugar processing seems to be the outlier with 33 percent per annum growth; however, one should remember that the local output was initially negligible.
The observed relationship between the initial import penetration and output growth is consistent with the view that one must distinguish between the import-competing and domestic-market-oriented sectors to understand the sectoral growth dynamics. While the favorable industrial prices and protectionism helped the import-competing sectors to expand output, the stagnant local incomes remained a significant constraint over the domestic-market-oriented branches, such as milling, vegetable oils, and leather processing.

Tariffs and growth
To what extent does the evidence on sectoral growth and nominal protection rates support the view that protectionism supported growth? The sectoral data made available with the present work also help to answer this question. To do so, I first calculate the equivalent of specific import taxes. The calculation of these rates needs some explanation because the import tax policy was complicated at the time.
Turkey's tariffs, inherited from the Ottoman wartime tariff scheme, had been mainly fiscal, and the government could not modify them due to the restrictions of the Lausanne Treaty, which ended in 1929. The new tariff scheme, introduced in 1929, increased specific rates to increase effective protection for intermediate and finished goods. 14 On the other hand, many consumer goods, particularly textiles, foodstuff, leather, and cement, came to be highly protected (Tezel 1982, 145-46). Eventually, the average aggregate equivalent tariff rate increased from 15 to 35 percent in a year and remained around 40-50 over the decade. However, no estimate of sectoral nominal protection rates exists.
I here approximate the nominal protection rates by the percentage of custom revenues to the import value for all goods. The trade publications report only the import volume (in kg) and import value in Turkish Figure 6. Growth rates and the initial import penetration. currency (TL), with further information on the origins of imports. Thus, the critical variable to be estimated for goods is customs revenues. 15 The primary empirical difficulty in deriving custom revenues is the existence of preferential tariff treatments. Bilateral trade agreements stipulated different tariff duties for various trade partners due to the special concessions. However, a closer look reveals that the preferential specific tariffs were more of a problem during the 1920s because some trade partners did not benefit from the most-favored-nation clause, which would have provided an equal tariff treatment. 16 I identify different specific rates for each year for the relevant statistical units and then multiply the imports from each country by the appropriate specific rate. Then, I add the revenues received from each country's imports and divide it by the total imports value to obtain the tariff rate. When more than one specific rate is present, all countries benefiting from the mostfavored-nation-clause enjoy the lowest rate, and the official not-discounted rate is applied to the others. This rule practically meant that the partners in the most-favored-nation sphere enjoyed the rates used to the parties of the Lausanne Treaty, for which tariff rates were frozen until 1929 (at 5 or 9 times the 1916 levels). For the countries outside the sphere, I apply the official rates. For the years after 1929, the lowest rate was equivalent to the official rate minus the discount determined by the trade agreements. If one trade agreement, say with Germany, specifies a specific discount on an item, this applies to Germany and all other countries benefiting from the most-favored-nation clause. If other trade agreements set more than one discount rate for the same good, the highest discount rate is considered due to the nature of the equal treatment. 17 Therefore, I obtain the nominal rates for the relevant statistical codes following this procedure. The second step is correctly aggregating the duties to get the aggregate rates, such as the nominal rate for cotton or woolen fabrics. As Federico and Tena (1998, 75-76) explained, the duties could be weighted with the composition of imports in a hypothetical free-trade situation. But obviously, such a free-trade environment never existed. Nonetheless, even if it was possible to find such "free trade" weights, one should consider that the import composition of the industry is inherently bound to change for different reasons, as we see in the rapid and early development of the textile sector in Turkey. We expect the production to begin with the lower quality products and develop toward the fine cloth types, meaning that a gradual change in the import composition occurs partly irrespective of the tariff policy. It is therefore controversial as to whether free trade weighting is ideal. Another alternative would be taking a simple average of duties across items, using the share of each item in domestic output, or assuming an ideal case of trade vectors, such as the composition of British exports or world trade. The first is the least perfect because it means no weighting, giving the most and least essential import items the same weights. Similarly, there is no reason that each country should have a similar import structure, meaning the latter method is also controversial. Therefore, I prefer to follow the more conventional approach by using the current values as weights to aggregate the rates of statistical codes. 18 It is true that this systematically biases the estimates downwards since the higher duties reduce the imports of the more protected goods. For instance, if a duty is prohibitive for a single-count cloth, and it is nil for the multiple-count, the resulting imports of the less-protected multiple-count will increase and substitute for the prohibited single-count fabric. That would yield a downward bias in the average rate. So, we should take the nominal rates, based on current values as weights, as lower bound estimates. Figure 7 shows that nominal tariffs positively correlated with output growth. The correlation was largest for cotton textiles, tobacco, woolen goods, and sugar processing, whereas it was lower for milling, alcohol, and vegetable oils. However, the tariffs-growth relationship is not necessarily linear because tariffs in interwar Turkey were complemented by quotas (quantitative import restrictions) between 1932-39 (Tezel 1982, 157), and increasing role of state factories in sectors such as sugar processing, textile and leather making. 19 However, how quota policy differed from tariffs regarding its protective impact remains unknown. In principle, regardless of the policy motivations and intentions, the effects of quotas had to operate by changing the association between local and import prices. If they were not redundant, that is, if quotas were not below the import levels that tariff levels implied (i.e., above the autarky level), they should have changed the elasticity of the local output to the import prices. 20 Likewise, the government-operated factories might have kept domestic prices artificially lower to increase domestic sales and output. The fast growth in sugar and cotton cloth supports this view. Yet, we do not observe a fast growth in alcohol and leather (other government priorities), indicating that a much more intricate public pricing policy might have been in place. These connections between quotas, government factories and output growth, though, need to be explored further detailed research.

Conclusion
Any attempt to make consistent quantitative estimates of past economic variables is bound to deal with the low quality and insufficient quantity of raw data sources. The potential conceptual mismatch between the original sources and the estimation objective is also critical. These problems make it necessary to make assumptions, cross-check alternative sources, and use novel statistical techniques. The case of interwar Turkey is not an exception. Even though the first applications of the national income accounting framework emerged in Turkey in the 1930s, we lack good data sources to estimate aggregate economic activities, such as the representative high-frequency census data or input-output tables. However, the abundance of quantitative sources, though scattered, on various aspects of the industry makes it possible to get relatively reliable estimates of output growth, which are comparable to the estimates available for other countries. Therefore, inspired by the scholars who worked on British and Italian industrialization in the nineteenth century, this study offers first sectorally disaggregated industrial output estimates, "first-generation estimates", for interwar Turkey.
On the one hand, these new estimates contribute to understanding peripheral industrialization. While the experience of the countries in the periphery of the interwar global economy was rather diverse and mixed, Turkey can be considered one of the large and highly agricultural countries which witnessed rapid industrialization. An accurate description of the early industrialization efforts of Turkey helps understand the economic development in the Balkans and Latin America at the time, where the highly prohibitive macro policies favored the local industrial sector in relatively large domestic markets. Maddison (1985) and Diaz-Alejandro (1984) usefully compare the peripheral economies regarding the experience of the Great Depression. Both point to the contrast between the small open economies in Asia, which adhered to orthodox policies and did not perform well, and the large economies in Latin America and the European periphery, which performed better while switching to policy activism. Turkey's experience conforms with this second group with the quick recovery of incomes accompanied by active industrial policy.
At the aggregate level, the present estimates indicate that the existing official industrial index has some upward bias. More importantly, it turns out that industrial development was balanced, as both textiles and food-processing branches, making up most of the value-added, grew significantly. Textiles outperformed food processing since the initial import penetration was more significant, while the latter was more dependent on domestic incomes and demand. Per capita incomes also increased in interwar Turkey, particularly in the second half of the 1930s; however, the income gains were much more limited relative to the size of industrial output growth. The per capita incomes just exceeded the pre-WWI levels by [1938][1939]. Therefore, while the growth was relatively balanced, the industry's diversity in growth rates was still considerable. For the sake of comparison, Egyptian industrial expansion in the 1930s was almost entirely due to the textiles' performance (Karakoç 2018). The overall picture reminds of Robert Tignor's (1984) description of the Egyptian economy in the 1930s as "development without growth," where industrialization happened within a stagnant economy. Turkey's experience is not as bleak as the statement implies since there was a steady increase in incomes in interwar Turkey. Yet, the contrast between modest income growth and fast industrial expansion is remarkable.
Finally, the present sectoral disaggregation clearly shows that the extent of output growth correlated with higher initial import penetration and higher nominal protection rates. The logic of import substitution dictates that the industrial output is a function of domestic demand and import replacement. Supporting this observation, I show that subsequent growth was higher in the sectors where the initial import penetration was more remarkable and in those where the nominal tariff rates were higher.

Acknowledgments
I am grateful to Şevket Pamuk (Bogazici University) and Peter Howlett (LSE, Economic History) for their constructive comments and continuous support during this study. I would also like to thank the participants of the Ph.D. Workshop at the Economic History Department at the LSE, where I presented earlier versions of this work. I also thank the anonymous referees for their comments and suggestions, which helped improve the presentation and discussion in this article.

Declaration of interest statement
There are no relevant financial or non-financial competing interests to report. Hayrettin, "Deri Sanayimiz." Torgut, Sevket, "Y€ unl€ u Mensucat Raporu." " _ Istanbul Dericileri Raporu" " _ Istanbul Pamuklu Mensucat Sanayi Raporu" "Izmir Dericileri Raporu" " _ Izmir Y€ un Mensucat Raporu" Yahya, H, "Zeytinya gı, Nebati Ya g ve Pirine Ya gı Sanayi." Notes 1. Issawi (1980Issawi ( , 1982 provide a conventional view of the Ottoman de-industrialization. Pamuk and Williamson (2011) assesses the magnitude of the decline of the local manufacturing. Karakoç, Pamuk, and Panza (2017) provide the most recent evaluation on the subject. However, the conventional view is not without criticisms. Quataert (2002) emphasizes that it is misleading to understate the dynamism of rural manufacturing in the nineteenth century. By a similar account, Pamuk (1986) points out that the resistance of handicrafts to the rising foreign competition was due to the low wages and abundant labour, along the lines of the labour-intensive industrialization models. 2. For an overview of macroeconomic developments in interwar Turkey, see Tezel (1982). For a detailed discussion of the policy shifts during the Great Depression and in the 1930s, see Tekeli and _ Ilkin (1977) and Tekeli (2009). 3. The railroad tracks in use increased from 3756 to 6948 km between 19236948 km between and 19396948 km between . 4. Fenoaltea (2011 is the culmination of author's decades-long work on the history of Italian industry after the unification up to the World War I. 5. Boratav (1988, 13) reminds that the economic idea called "national economy" that had been formed before World War I was partly implemented during the war and received official acceptance after 1923. 6. The annual detailed summary statistics were published in several volumes with the title Sanayi _ Istatistikleri by SIS between 1932 and 1941. 7. For a full list of the data sources, see Crafts (1985, 180). 8. An extensive historiographical survey can be seen in Fenoaltea (2003). 9. Please see the Appendix for details of the estimation of the demand for carpets.

Sources
10. To do so, I calculate the value added in each sector in 1927 using total value added in 1927 and the weights calculated so far. Then, these value-added figures are multiplied by the rate of change of output value (output times the average representative output prices) between 1927 and 1935, which is chosen as the reference point for the 1930s. As for the price data, I identify representative final goods for all sectors and obtain their prices in 1927 and 1935 from various sources. For the wholesale prices of sugar, olive oil, sheepskins, carpets, cotton yarn and cotton fabrics in Istanbul, see Sicilli Ticaret Gazetesi ve Piyasa B€ ulteni.
The prices of raw silk and hemp seeds measure the price change for silk spinning/weaving and hemp processing (). The average mining, timber, construction, electricity, and wool prices are obtained from Bulutay estimates. Alcohol prices are assumed to follow the aggregate price index. Note that Harley (1982) offers a similar weighting adjustment method to measure the British industrial output during the industrial revolution. 11. The figures in the book representing the Middle Eastern growth before 1914 only reflect the estimates of the authors of the relevant chapter in the book. 12. Williamson's neo-Ricardian model predicts the deindustrialization when the terms of trade move in favour of exports. The model, as elaborated in the book's chapters on each part of global periphery, predicts well the long cycles of industrialization and deindustrialization before 1914. 13. Tezel (1982). Pamuk (2000) finds the adherence to the orthodox monetary and fiscal policy stance rather curious. Eichengreen (1996, 23), however, points out that the economies which had experience with inflation in the aftermath of the World War I were more likely to avoid expansionary policies in the 1930s. France was the prime example. 14. For a discussion of the different proposed tariff schemes and the views of merchants, experts, and policy makers on tariff reform, see Kurmuş (1978). For the distinction between nominal and effective protection rates, see Kitson, Solomou, and Weale (1991), and Federico and Tena (1998). 15. The only existing estimate of nominal tariffs for this period are provided for 1916 and 1929 by Kurmuş (1978). Note that transaction and consumption taxes in imported goods were also of significant size. However, they were applied to both local output and imports, so they did not particularly favour the local output (Tezel 1982, 147). The Law 2458 and its subsequent revisions clearly show that the consumption tax was not discriminatory. Laws 2546, 2731 and 3101 can be found on the digital archives of https://www.resmigazete.gov.tr/ 16. I limit the analysis only to the major trade partners, to the countries whose exports to Turkey added up to around 90 per cent of the total imports of each good. This is just a practical definition to reduce the computational costs. Defining this way, the major partners appear as the European countries, Japan, Syria, Egypt, and the US. Within this group, the number of countries that were outside the sphere of most favoured nation clauses decreased every year so that by 1929 only Syria and Egypt remained excluded. 17. All the necessary information is combined in a number of volumes published by various contemporary experts. These volumes outline how the import taxes were calculated, the changes of specific rates over time and the discounts rates (Nuri 1929(Nuri , 1931(Nuri , 1934Y€ ucelir 1935;Emilsili 1938). 18. See examples in Eichengreen and Irwin (2010) and Irwin and Temin (2001). 19. During 1932-39, the quota system evolved from being product-specific to the bilateral logic. In 1937, government abolished them for the countries with which Turkey had more than a 20 percent trade surplus, effectively for all major trade partners. 20. Towle (1940, 614) identifies several reasons why quota policy was preferred to tariff by interwar governments. Crucially, they allowed strict and predictable control over imports, unlike tariffs which permitted demand and supply factors to work.