Policy Research Working Paper 11118 Chinese Imports and Industrialization in Africa Evidence from Ethiopia Marina Mavungu Ngoma Development Economics Development Research Group May 2025 Policy Research Working Paper 11118 Abstract The rise of China in the global economy has been linked with leads to a 15.2 percent increase in industry employment. negative impacts on employment across many high- and The inputs effect is disentangled from the other two effects middle-income countries. However, evidence for African by decomposing total Chinese imports by their end-use countries is limited. This paper investigates the causal category using input-output tables. The evidence shows relationship between Chinese imports and manufacturing that imported intermediate inputs are driving the employ- employment in Ethiopia. Imports may harm domestic firms ment gains. The findings are consistent with the idea that through a revenue effect (lower market shares) or benefit employment gains are a result of productivity gains and them, indirectly if competition spurs innovation or directly increases in capacity utilization. These employment gains through access to better quality or cheaper inputs. The anal- appear to benefit large firms and labor-intensive industries ysis shows that a one unit increase in import penetration disproportionately. This paper is a product of the Development Research Group, Development Economics. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The author may be contacted at mmavungu@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Chinese Imports and Industrialization in Africa: Evidence from Ethiopia Marina Mavungu Ngoma∗ JEL Classification: O14, F14, F16, L60, C13, C23, C26, D57 Keywords : Imports, Inputs, Manufacturing, Ethiopia, China, Employment. ∗ The World Bank, Development Research Group. Email: mmavungu@worldbank.org. I am sincerely grateful to my Tufts University PhD advisors Margaret McMillan, Jenny Aker, Federico Esposito and Cyn- thia Kinnan, for invaluable guidance and support throughout this project. I thank Mats Ahrenshop, Paulo Bastos, Paul Brenton, Alison Campion, Kyle Emerick, Mesay Gebresilasse, Susan Godlonton, Abel Ku- tangila, Brian McCaig, Eoin McGuirk, David McKenzie, Nathan Nunn, Adam Storeygard, Shinsuke Tanaka, and John Ulimwengu for insightful comments and suggestions. I am grateful for the constructive observa- tions and inputs from participants at Tufts University’s Economics and Public Policy Workshop, the 2023 STEG Annual Conference in London, the HEC Montr´ eal (University of Montreal) Seminar, the Toronto Metropolitan University Economics Seminar, the Conference on Structural Change in Industrial Policy in ´ Africa, and the Cercle des Economistes Congolais. I thank Kwok Wai So, Jared Fang and Mia Ellis for excellent research assistance; as well as Trinity Manning and Vasudha Ramakrisha for their work at the earlier stages of this project. I thank Seneshaw Tamru, Tewodros Gebrewolde and the IGC-Ethiopia team for helpful support in accessing and discussing Ethiopian datasets. I am grateful to Gaaitzen de Vries and Emmanuel Mensah for sharing valuable information on imports classifications. I am also grateful for the generous grant from the International Growth Center. 1 Introduction After joining the World Trade Organization in 2001, China quickly grew to dominate inter- national markets, becoming the largest exporter in the world by 2009. The rapid ascension of Chinese exports on the global stage introduced new competition for manufacturing in- dustries, and the opportunity for high- and middle- income countries to offshore production to China. The increased competition with China has led to a reduction in manufacturing employment in high- and middle-income countries (Autor et al., 2013; Acemoglu et al., 2016; Iacovone et al., 2013; Blyde & Fentanes, 2019). However in low-income countries, like many across Africa, there is little causal evidence on how the boom in Chinese imports over the past two decades has affected domestic manufacturing employment. Despite the well-established negative effect observed in advanced economies, the theo- retical relationship between Chinese imports and manufacturing employment in any given country is ambiguous. The effect of Chinese imports can operate along two main channels: the competition channel, in which domestic producers compete against Chinese producers of similar goods; and the inputs channel, in which domestic producers upgrade their production process by leveraging the Chinese-imported goods as inputs. The competition channel can either lead to a decline in manufacturing employment due to the revenue effect (Leamer et al., 1995; Autor et al., 2013; Pierce & Schott, 2016; Acemoglu et al., 2016), meaning domes- tic firms lose out to the Chinese imports, or an increase in manufacturing employment due to the innovation effect, if competition pushes domestic firms to improve their production process (Bloom et al., 2016; Aghion et al., 2005; Raith, 2003; Schmidt, 1997). The inputs channel can lead to increases in manufacturing employment by improving domestic firms’ access to better quality and more affordable intermediate inputs that can lower production costs while also enhancing productivity (Topalova, 2007; Goldberg et al., 2010; Redding et al., 2006; Amiti & Konings, 2007). The net impact of these countervailing effects may be quite different in African countries as compared to more advanced economies, a question which is understudied in the literature. This paper investigates the impact of increased Chinese imports on manufacturing em- ployment in Africa. Most African economies are still in the early stages of structural transfor- mation with under-developed and growing manufacturing sectors (Diao et al., 2021; Rodrik, 2018; Erten et al., 2019). As such, we may expect African economies to be less prone to the competition channel than more advanced economies that have more developed manufac- turing sectors. African economies also have greater potential for boosting industrialization through better access to high technology inputs they otherwise lack, meaning the inputs 1 channel may dominate. The distinctive nature of African economies may lead to a different net effect from that observed in high and middle-income economies trading with China. Studying the case of Ethiopia, this paper provides new evidence that, in contrast with the experience in wealthier countries, exposure to Chinese imports led to employment gains in the Ethiopian manufacturing sector. Like in other countries across the world, trade between Ethiopia and China has grown significantly since the early 2000s. Chinese exports to Ethiopia increased from 254 million USD in 2001 to 4.07 billion USD in 2016 (BACI database, CEPII). Moreover, manufacturing imports from China exceed total foreign direct investment in Ethiopia by a factor of 30 on average over the same period. At the same time, the Ethiopian manufacturing sector remains small. The average share of manufacturing employment was 7.87% over the period from 2001 to 2016 (World-Bank, 2021). To measure industry employment and Chinese import exposure, two main datasets are combined. First, the Ethiopian Large- and Medium-scale Manufacturing Establishment Census is used to compute manufacturing employment and other industry outcomes. Manu- facturing employment is measured by aggregating employment year by year across firms that operate in the same industry. Second, Chinese imports to Ethiopia are extracted from the BACI dataset. Imports are aggregated to the 2-digit industry level. The sample of analysis is an annual panel covering 22 manufacturing industries from 2002 to 2017 1 To causally identify the impact of exposure to Chinese imports on Ethiopian manu- facturing employment, this paper relates the variation in outcomes across manufacturing industries to the variation in industry exposure to Chinese import competition. Exposure to Chinese import competition is measured by the import penetration ratio, calculated at the industry level as contemporaneous Chinese imports relative to the initial size of an industry in the domestic market2 . The baseline regression specification includes year and industry fixed effects to control for shocks common to all industries and time-invariant industry-specific unobserved factors affecting employment. One potential concern with this estimation strategy is that increasing exposure to Chi- nese imports could be driven by domestic demand shocks which could simultaneously be correlated with industry outcomes. To address this concern, this paper instruments Chi- nese import penetration in Ethiopia with Chinese import penetration in other Sub-Saharan African economies3 . The instrument isolates the component of the variation in Chinese 1 Examples of industries include the textiles, electrical machinery, wearing apparel, or foods and beverages. 2 Referred to as Chinese import penetration for the remainder of the paper 3 Chinese import penetration in other Sub-Saharan African economies is computed as the average across other countries’ industry-level ratio of Chinese imports over Ethiopia’s initial size of the industry 2 import penetration that is influenced solely by productivity shocks in China. My empir- ical strategy follows the same logic as a vast body of literature, such as Acemoglu et al. (2016); Autor et al. (2013); Bloom et al. (2016); Pierce & Schott (2016), analyzing the ef- fects of Chinese imports on the domestic manufacturing sector in high and middle-income economies4 . To support the identifying assumptions, evidence is provided of a strong first stage, demonstrating that Chinese imports to other Sub-Saharan African economies strongly predict Chinese imports to Ethiopia. To further support the exclusion restriction assumption, an over-identification test is conducted, treating each country included in the instrumental variable as a separate instrument. The paper’s main findings suggest that looking across industries and years, increased exposure to Chinese imports increases manufacturing employment. On average, a one unit increase in industry import penetration leads to a 15.2 percent increase in industry em- ployment. This impact is economically and statistically significant, although base levels of manufacturing employment are relatively small with an industry-level average employment of 7,783 workers. The estimated employment gains associated with a one standard deviation increase in Chinese import penetration is 1,790 workers. This result is robust to several checks. First, the analysis test whether the results are driven by a specific industry. For example, industries like food and beverage are top importers while they also employ the highest share of manufacturing workers. In contrast, the wood industry is a bottom importer and exporter. The findings demonstrate that no single industry is driving the results. Second, the estimation is conducted using different subsamples of the data, varying the firms and the years included in the analysis sample. The coefficients are robust to almost all alternative samples. Third, robustness is tested using alternative import exposure measures. A different import penetration ratio is computed, where initial absorption is replaced with initial output or initial employment. Additionally, contemporaneous imports are used as the explanatory variable. The coefficients remain consistently positive and are larger in magnitude. The effect of Chinese imports on manufacturing employment is the result of two different channels at play, namely the competition channel (revenue effect or innovation effect) and the inputs channel (inputs effect). To disentangle the employment effects arising from the competition channel from those arising from the inputs channel, , the analysis proceeds in three steps. First, Chinese imports are categorized as either intermediate goods or final goods, using the Broad Economic Categories (BEC) concordance table. By construction, 4 This literature instruments Chinese imports to the country under analysis by Chinese imports to other countries 3 the input usage channel is present only for intermediate goods, while final goods will isolate the competition channel. Over my period of analysis, 78.5% of Chinese imports to Ethiopia consisted of intermediate goods. Second, the direct impact of intermediate goods is estimated separately from the direct impact of final goods. The results indicate that imports of intermediate inputs have a positive impact on employment, whereas no detectable effect is observed from the import of final goods. Specifically, one unit increase in exposure to Chinese-imported intermediate goods leads to a 19.7% increase in industry employment. This effect could be driven by either the competition (innovation effect) or the inputs channel. Auxiliary evidence suggests that the positive impact of intermediate goods is likely due to the inputs channel. Although the data do not allow explicit separation of domestic producers by intermediate and final goods, suggestive evidence on the category of products indicates that the majority of goods produced by manufacturing industries in Ethiopia are final goods. This could suggest that there is limited competition between intermediate goods producers and importers of the same products. Additionally, the share of own industry input usage is relatively high within the manufacturing sector. For example, the textile industry sources 57% of its manufacturing inputs from the textile industry. Overall, 55% of its total inputs are supplied by the agricultural and mining sectors, 15% from the services sector, and about 30% from the manufacturing sector (including more than half from the textile industry). Because domestically produced intermediate inputs are more likely to be locally produced agricultural products or other non-technologically advanced goods, it is likely that the imported intermediates are not in direct competition with domestically produced intermediate inputs. To clearly identify the inputs effect and calculate the industry imports usage, a new measure of imports exposure is constructed using the input-output table. Specifically, im- ports of intermediate inputs are proportionally allocate to industries based on the industry’s input usage share in total intermediate usage. Industry imports usage is the weighted aver- age of Chinese intermediate imports from each upstream industry supplying inputs to the exposed industry. Findings indicate that employment results are driven by the inputs chan- nel, consistent with the composition of Chinese imports dominated by intermediate goods. In particular, industries exposed to Chinese imports through downstream linkages, namely their input suppliers, record employment effects of 14.1%, consistent with the inputs usage channel. IPositive effects are observed only among firms using imported inputs while there is no effect on firms that use domestic inputs. This result suggests that the technologies in- corporated by the use of imported inputs are efficiency enhancing (Amiti & Konings, 2007). 4 Moreover, this is consistent with the fact that in 2001, shortage of raw materials and other issues related to difficulty accessing inputs were listed among the top reasons that prevented Ethiopian manufacturing firms from operating at full capacity. Accordingly, Chinese import penetration is found to reduce the proportion of firms facing a shortage of raw materials by 13.8%. Next, the paper analyzes the mechanisms linking Chinese imports and Ethiopian man- ufacturing employment, with supporting evidence provided for the inputs channel operating through increased productivity and capacity utilization. Industry production functions are estimated using the Levinsohn & Petrin (2003) methodology to correct for simultaneity in the choice of inputs. The findings reveal that firms utilizing intermediate imports from China through upstream linkages exhibit higher productivity. Moreover, a positive impact on skills upgrading is reported for the industries affected through the downstream shock. This suggests that, with the assumption of complementarity between labor and inputs, the use of higher quality inputs from China might require firms to employ more skilled workers. Although no evidence is found of changes in firm entries and exits in response to Chinese imports, the results vary across industry characteristics. In particular, the positive employ- ment effects are driven by large and labor intensive firms, whereas ownership type has no impact on the effect. Finally, this paper provides evidence that firms are substituting away from traditional trade partners and towards cheaper Chinese imports. This paper makes three contributions. First, it contributes to the literature on the impact of Chinese import competition on domestic manufacturing employment by studying the case of a relatively low-income African country. Previous literature on this topic has largely focused on high-income economies, examining rising wage inequality and manufac- turing job losses driven by import competition from low-wage countries in general (Revenga, 1992; Bernard et al., 2006; Ebenstein et al., 2014), and from China in particular (Autor et al., 2013; Acemoglu et al., 2016; Iacovone et al., 2013; Mion & Zhu, 2013; Bloom et al., 2016). In contrast, this paper analyzes the impact of Chinese import competition in Sub- Saharan Africa, a region where input constraints and industrial underdevelopment shape a different set of dynamics. In Ethiopia, the manufacturing sector is small and growing, and China is a relatively more technologically advanced economy. Studying this context provides insight into how low-income countries are affected by increased trade with more industrially advanced partners and highlights how trade can serve as a channel for industrial upgrading rather than deindustrialization. One related study in the African context is Edwards & Jenkins (2015) who analyze South 5 Africa—a more developed economy relative to most of Sub-Saharan Africa. Their analysis centers on the competition channel, whereas this paper finds the inputs channel to be more salient in less advanced settings. In contrast with most findings in the literature, results from this study suggest that in low-income countries facing input constraints, increased imports of intermediate goods can raise domestic manufacturing employment by enhancing production capacity and firm-level efficiency. Second, this paper contributes to the literature on the labor market impacts of the “China Shock” by distinguishing between imports of final goods and those of intermediate inputs. While much of the existing work estimates the aggregate competition effects of total imports, this paper demonstrates that such aggregation obscures important heterogeneity: whereas imports of final goods may exert neutral or adverse employment effects, imports of intermediate goods can generate substantial employment gains. In doing so, this study builds on prior research, including Mion & Zhu (2013) and Biscourp & Kramarz (2007), as well as more recent contributions such as Taniguchi (2019), who separately examines final and intermediate imports to explain employment heterogeneity in Japan, and Aghion et al. (2024), who decompose the China Shock into input supply and output competition effects at the firm level. While Mion & Zhu (2013) emphasize offshoring and Biscourp & Kramarz (2007) focus on the competition channel via final goods, this paper extends the analysis to intermediate inputs and their supply-side effects. This paper further distinguishes itself by applying a comparable decomposition in a low-income country context—an area largely underexplored in the existing literature. In addition, it leverages input-output linkages to isolate the productivity effects of intermediate input use from those attributable to product market competition in a structurally different environment. It also goes beyond prior work by examining the specific channels through which improved access to imported inputs enhances manufacturing employment, specifically through firm-level productivity and capacity utilization. These mechanisms are especially salient in contexts marked by persistent input shortages and underutilized production capac- ity. The findings offer robust evidence that, even in low-income settings, expanding access to affordable, high-quality intermediate goods can yield meaningful improvements in both firm performance and employment outcomes, complementing existing evidence from advanced economies. Third, this paper advances the literature on importing and productivity, specifically the studies that provide empirical evidence that imports of intermediates or the decline in input tariffs are associated with productivity gains (Goldberg et al., 2010; Abreha, 2019; Goldberg et al., 2010; Redding et al., 2006; Nocke & Yeaple, 2006; Topalova & Khandelwal, 2011; 6 Topalova, 2007; Amiti & Konings, 2007; Kasahara & Rodrigue, 2008). While the existing literature primarily utilizes input or output tariffs as the treatment variable, this paper complements and extends these studies by employing the value of imports as the treatment variable. It also demonstrates that skills upgrading results not only from the competitive effects of trade, but also from access to higher-quality inputs. The remainder of the paper is organized as follows. Section 2 presents the background and conceptual framework. Section 3 describes the empirical strategy and presents the data. Section 4 presents the baseline results and section 5 presents the results on the two channels of Chinese imports. Section 6 discusses the mechanisms through which imports impacted employment. Section 7 concludes. 2 Background and Conceptual Framework This section outlines the nature and evolution of Chinese imports in Ethiopia. It highlights that the increase in imports from China was driven by reforms in China rather than specific trade policies in Ethiopia. Additionally, it explores the variation in exposure to Chinese imports across industries, which forms the basis for the empirical strategy. Finally, back- ground information on the Ethiopian manufacturing sector and its industrialization efforts is provided. 2.1 Chinese imports in Ethiopia Over the last two decades, Ethiopia has experienced a sharp increase in economic relations with China, especially through trade. This fact is common to most countries around the globe. Figure 1 plots the value of Chinese manufacturing imports5 and FDI inflows to Ethiopia from 1996 to 2017. Since 2001, Chinese exports to Ethiopia have increased at an annualized rate of 27%, from 59 million USD in 1996 to 4.07 billion USD in 2016. Even as Ethiopia increasingly welcomed Chinese Foreign Direct Investment, total FDI to Ethiopia is still dwarfed by the value of Chinese imports. Chinese FDI approximated 327 million USD per year over from 1996-2017, about 30 times less than the value of Chinese imports. The Chinese imports trend remained positive throughout the 2002-2017 time period, although a slowdown occurred a few years following the 2009 financial crisis and a major drop in 2016. Imports have generally been declining in Ethiopia since 2016 due to a severe shortage 5 In this paper, I focus on manufacturing imports. Of the Chinese imports in Ethiopia, 99.7 % are manufactured goods. The other 0.3% are goods classified as agriculture, electricity, mining and services. 7 of foreign exchange. The shortage was caused by a drought and a weak global environment, causing a rapid decline in foreign reserve buffer accompanied by general economic growth slowdown (IMF, 2016). Chinese imports reflect that macroeconomic trend.6 The rise of Chinese imports to Ethiopia coincides with China’s accession to the World Trade Organization in 2001. As depicted in Figure 1, the rate of increasing Chinese imports picks up in 2002. The timing of this increase suggests that it is unlikely to be the result of a demand shock in Ethiopia. Rather the increase was likely due to the reforms China undertook to make its economy more market-based and competitive (Ianchovichina & Martin, 2001). My empirical approach accounts for this feature while also focusing on the post-2002 period. Ethiopia imports a mix of manufactured goods from China. Imports range from fertiliz- ers and other types of chemical products, to plastics, rubber articles, clothing, and mechan- ical appliances. Electrical and non-electrical machinery, communication equipment, wearing apparel and textile goods are among the largest share of imported goods. For example, in 2016, about 36% of total Chinese imports consisted of electrical machinery and equipment, televisions, video projectors, and mechanical appliances (Figure A.1). Not all the goods imported to Ethiopia were also produced domestically. For example, some types of power engines, marine engines, milling machines and other textile industry goods, machinery and medical equipment have no domestic production. The majority of goods not produced in Ethiopia are in the textile, machinery, and medical equipment industries. This suggests that not all imported goods directly compete with domestically produced goods. There is variation in the degree of Chinese import exposure across industries and through time in Ethiopia. Table 1 reports four main characteristics across industries.7 The first column captures industry size as measured by the industry share of total average, annual manufacturing employment in the country. The second column presents relative exposure to Chinese imports as measured by the industry’s average annual imports share of total manufacturing imports. The last two columns report the average annualized growth of employment and imports between 2002 and 2017. As shown in the table, there is great variation in the share of imported goods, as well as in the annual growth of imports across industries. For example, the machinery industry originally has a large share of total imports, but grows at the same pace as the average industry between 2002 and 2017. In contrast, the 6 The major drops from 2015 to 2016 are concentrated in the tobacco industry (74%) and the transport equipment industry (69%). While between 2015 and 2016 many industries imports kept growing, almost all experienced drops from 2016 to 2017 when the major drops were observed in the metals industry (70%) and the transport equipment industry (60%). 7 Industries use the the 2-digit ISIC codes. Further details on the industry coding are presented in section 3 8 transport equipment industry represents a relatively small share of total imports in 2002 but grows much faster than the average industry. My empirical analysis exploits this variation across industries to measure the effect of exposure to Chinese imports. The overall correlation between manufacturing imports and employment across indus- tries is positive, but not systematic. For most industries around the 50th percentile of both employment and imports, the table suggests a positive correlation between employment and imports. For example, the textile and non-metallic mineral industries employ an important share of manufacturing workers (37%, 19% and 13% of the total manufacturing workers re- spectively) and are also exposed to Chinese imports. The industries that experienced large growth of Chinese imports also grew in terms of employment. For example, the vehicles in- dustry experienced a 62% growth in imports and a subsequent 22% growth in employment. Meanwhile the fabricated metals industry grew less in imports (38%) and subsequently less in employment (16%). Nonetheless, some of the biggest manufacturing industries - as measured by employ- ment share - display a relatively small exposure to imports. This is the case for the food and beverages industry, representing 27% of total manufacturing employment while their share of imports is less than 1%. In parallel, some of the top importing industries, such as the communications equipment and machinery industries, have a small share of domestic employment. In this paper, I employ a rigorous empirical approach to shed light on the direction and magnitude of the relationship between industry exposure to Chinese imports and industry employment. 2.2 Ethiopia’s industrialization As discussed in the previous subsection, the Ethiopian manufacturing sector remains small, although the overall trend of the manufacturing employment is increasing. Importantly, most of the studies analyzing the impact of China are realized for high-income countries where manufacturing employment is declining. In contrast, in Ethiopia, efforts are made by the government to increase the contribution of the manufacturing sector to GDP and employment growth. The Ethiopian industrialization efforts have focused on infrastructure development rather than targeting trade protection. While the policy instruments that prevailed since the impe- rial period focused on high tariff and import substitution, post-2002 industrial policy favors direct support for select export sectors and provision of economic incentives and credit scheme (Gebreeyesus, 2013). The major trade policy change in Ethiopia was implemented between 9 1993 and 1998.8 During this time period, the average import tariff (both output and input) declined from 41.6% to 19.5%. Before 1993, the maximum tariff recorded was 230%. After 1998, tariffs kept declining but at a much lower, practically negligible, pace as shown in Figure A.5 (Bigsten et al., 2016). Following the reforms and trade liberalization efforts of the 1990s, three industries main- tained relatively high tariff rates: the wearing apparel, footwear, and tobacco industries (Ianchovichina & Martin, 2001). This policy is consistent with the Ethiopian government’s focus on promoting some of the most labor intensive industries within the country, in light of its ”National Import Substitution Strategy” (Ministry of Industry, 2013, 2023). These tariffs are not necessarily predictive of these sectors’ shares of total Chinese imports. The wearing apparel industry has a relatively large import share, while the tobacco industry imports less. Notably, these tariffs were not specific to trade with China. Figure A.5 shows an almost flat trend in tariffs between 2002 and 2017, while Chinese imports were growing steadily during this period. My empirical analysis accounts for these potentially confounding factors as well as the relative importance of each industry in the industrial policy. Together, these features reinforce the assumption that the rise in Chinese imports in Ethiopia is likely driven by reforms undertaken in China, rather than specific policies in Ethiopia. They also show why it is more appropriate to capture the Chinese trade shock through the directly observed industry imports rather than through tariffs. As a country, China is currently Ethiopia’s top export and import partner: Chinese im- ports represent 25% of total imports. The share of Chinese imports has grown substantially relative to other trader partners. Prior to the China shock, Ethiopia was primarily import- ing from Europe, but the relative share of imports from China grew substantially such that in 2016, the total value of imported goods from China was higher than those from Europe (Figure 2). Specifically, the Chinese share of total imports grew from 6.48 % to 27.96 % between 1998 and 2016. The reallocation away from Western economies could reflect the fact that China has become more competitive in producing the goods that Ethiopia needs to import, hence supplying similarly sophisticated goods but at a lower price. 2.3 Conceptual framework Conceptually, the total impact of imports on domestic manufacturing employment is an em- pirical question. The direction and magnitude of impact depend on the relative importance 8 See Gebreeyesus (2013), Fenta (2014) and Bigsten et al. (2016) for extensive details about the economic reforms undertaken by the Ethiopian government to move towards a private and marketed economy. 10 of two main channels, namely the competition channel and the inputs channel. Suppose a firm’s residual demand, representing their market share, is given by: 1 ¯) Q=S − b(P − P N where S represents the total industry output (suppose constant), N the number of firms ¯ the price set by Chinese firms, and within their industry, P the price set by domestic firms, P b a coefficient capturing product differentiation (greater substitution, greater competition). In equilibrium, if domestic and Chinese firms set an equal price, then the market will be equally shared across firms in the industry. Formally, if P = P ¯ , then Q = S/N . Competition arises from domestic industries producing the same goods as those being imported from China, leading Ethiopian producers to compete against Chinese producers within the same industry. This channel may have two opposite effects on manufacturing employment. Competition may reduce employment through revenue effects driven by lower sales, potentially in conjunction with increased firm exit. This revenue effect is consistent with the Heckscher-Ohlin factor-proportions theory of comparative advantage (Leamer et al., 1995; Feenstra, 2003). According to these models, Chinese competition reduces the relative prices of goods in competing Ethiopian sectors. In the framework presented above, Chinese comparative advantage will allow Chinese firms to produce at lower price than domestic firms, or P ¯ < P . As such, high-cost Ethiopian producers see their sales fall - Q < S/N - and are eventually forced to exit, ultimately leading to lower employment in those sectors. As in Autor et al. (2013); Pierce & Schott (2016) and Acemoglu et al. (2016), increased competition imposed by imports from China led to declines in manufacturing employment. However, competition may also increase employment due to an innovation effect whereby domestic producers upgrade their technology to compete with Chinese producers. This is particularly true if innovation is not labor-saving. The innovation effect is supported by Schumpeterian models which predict that firms willing to escape competition will innovate Aghion et al. (2005). Moreover, competition can increase firms’ incentives to expand their market share (Raith, 2003) or minimize agency costs (Schmidt, 1997), which induces inno- vation. Such moves by domestic firms will lead to the lowering of their prices. Illustratively, innovation will allow domestic competing firms to reduce the price gap with Chinese firms, or at most, produce at lower prices. In the latter scenario, P < P¯ such that Q > S/N . In Europe, Bloom et al. (2016) found that innovation - as measured by patenting - rose within firms that were more exposed to increases in Chinese imports. In Peru, Medina (2022) pro- vides evidence for quality upgrading in response to competition from China and documents 11 increases in annual sales and employment. In Canada, Yang et al. (2021) in response to the rise in Chinese imports, firms that prioritize product innovation as compared to process innovation have higher profits if they survive, with little impact on their likelihood of exiting. According to the inputs channel, imported intermediate goods may increase employment through productivity gains (Topalova, 2007) arising from access to better quality (Amiti & Konings, 2007), cheaper (Grossman & Rossi-Hansberg, 2008), or a greater variety (Goldberg et al., 2010) of inputs. Such intermediate goods produced and exported from China can be considered to be higher quality, to have lower prices, or to be completely non-existent in Ethiopia, and therefore may increase productivity in Ethiopian industries using Chinese inputs. Quality aside, if Chinese imports have lower prices, access to these inputs can improve firms’ efficiency by lowering their production costs (Halpern et al., 2015). Higher productivity may allow firms to expand their production and, consequently, employment. An increase in productivity can be expressed as an increase in product differentiation (a reduction in b) or a reduction in domestic firms’ prices P, such that the firm’s residual demand will increase. Furthermore, if inputs are complementary to labor (Atalay, 2017), then greater access to inputs may lead firms to demand more labor conditioned on existing installed capacity. Additionally, increased access to inputs may induce an expansion of firms’ capacity utiliza- tion. The inputs channel is potentially more salient in low-income countries with poor access to good quality inputs. One way to disentangle the competition and inputs effects is by separating the effects of imported goods by their end use category. Imports of final goods from China will impact competition in domestic markets. As discussed above, this competition can have two opposite effects: a reduction in employment through lower sales (revenue effects), or an increase in employment through technology upgrading (innovation effect). On the other hand, imports of intermediate goods can both spark competition and induce an input effect. As a result, we can observe three possible effects on domestic producers: a revenue, innovation, and inputs effect. This paper proceeds in two steps. The first part estimates the overall impact of total Chinese imports. The second part delves into the two channels through which these imports affect Ethiopia. 12 3 Empirical Strategy and Data 3.1 Industry exposure to Chinese imports To measure industry exposure to Chinese imports, this paper follows Acemoglu et al. (2016) and Bernard et al. (2006) and defines the import penetration ratio as the increase in Chinese imports by industry relative to the initial domestic industry demand. It is a measure of the degree to which domestic demand is satisfied by Chinese imports. Specifically, the industry import penetration is given by: China ImportsChina it IP Pit = (1) Outputi,1998 + Importsi,1998 − Exportsi,1998 China where IP Pit denotes the import penetration of Chinese imports in industry i and year t. ImportsChina it is the total imports from China in industry i and year t. Outputi,1998 , Importsi,1998 and Exportsi,1998 , are the production, total imports and exports for industry i in 1998, respectively. Together, the denominator represents the industry-specific initial domestic absorption. The year 1998 is chosen as the base year because it predates the surge in Chinese imports in 2002 and provides a sufficient number of observations across industries. 9 Figure 3 shows that, consistent with the trend in Chinese imports, import penetration increased significantly during the time period of analysis. In 2002, less than 10% of the initial domestic demand was met by Chinese imports. In contrast, in 2016, Chinese imports represented 135% of initial domestic absorption. This trend is observed across all industries taken individually (Figure A.2). 3.2 Main estimating equation My empirical strategy leverages the variation in industry exposure to Chinese import com- petition to explain the variation in a range of outcomes across manufacturing industries. The baseline specification is as follows: China Yit = β1 IP Pit + Xit β2 + θt + θi + it (2) 9 As a robustness test, two alternative base years are considered: 2001 and the average domestic absorption from 1998 to 2001. 13 In this context, Yit is the outcome for industry i and year t. The main outcome of interest is the logarithm of industry employment. IP Pit is Chinese import penetration in Ethiopia for industry i in year t. To address concerns of omitted variables bias, the regression equation includes θi and θt as the industry and year fixed effects, respectively. Industry fixed effects control for industry-specific unobserved factors affecting employment, such as the availabil- ity of raw hides in the leather processing sector which is variable due to animal disease. Year fixed effects control for nationwide shocks common to all industries, such as the global financial crisis of 2008. The regression equation includes a vector Xit of industry-level time- varying controls. The baseline specification controls for yearly imports from Africa, America, Europe, and Asia, excluding China. These control for confounding factors that may be cor- related with Chinese imports to Ethiopia while also driving industry aggregate employment. Figure 2 also shows that, while China has become Ethiopia’s top source of imports, Ethiopia still trades with other regions in the world. it is the error term, clustered at the industry level as the import shocks may be correlated within industries. The coefficient of interest is β1 , which provides an estimate for the percentage change in industry employment associated with the industry’s increase in Chinese import penetration. 3.3 Instrumental variable As specified in equation 2, the estimates of β1 could be biased if rising Chinese import competition is endogenous to domestic factors in Ethiopia that also correlate with indus- try outcomes. For example, the Ethiopian manufacturing industries can experience shocks leading to increased demand for Chinese goods while also boosting domestic demand, and subsequently domestic employment. To address this concern, the analysis utilizes an instru- mental variable approach. The goal of the instrument is to capture the component of the rise in Chinese exports to Ethiopia that is unrelated to domestic factors. As discussed in Section 2, the rise in exports from China is more likely driven by reforms undertaken by the Chinese government to transition to a market based economy, which resulted in its accession to the WTO in 2001. If Chinese exports to Ethiopia grow over time, exports to China’s other trading partners should similarly grow. To address potential endogeneity, Chinese import penetration in Ethiopia is instru- mented using Chinese import penetration in other Sub-Saharan African economies. The analysis includes a set of 12 Sub-Saharan African countries that also import from China and whose gross domestic product per capita is not greater or smaller than twice that of Ethiopia. Countries experiencing major conflicts are excluded from the sample10 during the 10 According to the World Bank’s list of ”High institutional and social fragility” countries (2021) 14 period of analysis.11 The assumption is that this instrument will isolate the component of the variation in exposure to Chinese imports that is influenced by productivity shocks in China. Under the identifying assumptions discussed below, this instrument will yield causal estimates of the Chinese exports shock on manufacturing employment. The instrument is given by: China O ImportsChina it O IP Pit = (3) Outputi,1998 + Importsi,1998 − Exportsi,1998 China O where IP Pit denotes Chinese Import Penetration of industry i in year t in the set of China O other Sub-Saharan African economies. Likewise, Importsit is total imports from China in industry i and year t. Outputi,1998 , Importsi,1998 and Exportsi,1998 , are production, total imports and exports for industry i in 1998 in Ethiopia, respectively. 3.4 Identifying assumptions In order for the aforementioned instrument to provide causal estimates of the impact of increased exposure to Chinese imports, it should satisfy two conditions. First is the relevance condition, which is satisfied if Chinese import penetration in other African economies causes variation in Chinese import penetration in Ethiopia. I test for weak instruments by assessing the joint significance of the instrument’s coefficients via the Kleibergen-Paap F-statistic. A first-stage F-statistic smaller than 104.7 indicates the presence of a weak instrument (Lee et al., 2022). My instrument satisfies this condition, as shown in Figure 4 displaying the first-stage of the instrument. The second main assumption is the exclusion restriction. The exclusion restriction im- posed by the instrument is that rising Chinese import penetration in other African economies affects the Ethiopian manufacturing industry only through the increase in Chinese imports in Ethiopia. In other words, import demand shocks need not to be correlated across countries. Arguably, rising imports from China is associated with supply-driven shocks such as the pol- icy efforts to lower barriers to trade, among others. In fact, Chinese imports in Ethiopia did not start rising until 2002, following China’s accession to the WTO. Moreover, being a small economy, Ethiopia is unlikely to have caused the rise in Chinese exports to other African countries. Although this assumption is plausible, steps are taken to ensure the consistency of estimates and the validity of the instrument. The analysis begins by testing the endogeneity of Chinese import growth. Next, an overidentification test treats each country separately as an instrument. Finally, robustness checks use an alternative instrument. 11 The countries included are: Chad, Eritrea, Guinea, Guinea-Bissau, Liberia, Madagascar, Malawi, Rwanda, Sierra Leone, Togo, Uganda, and Tanzania. 15 3.5 Data To examine how exposure to Chinese imports affected the Ethiopian manufacturing sec- tor, this paper uses data from two main sources: the manufacturing sector data from the Ethiopian Large and Medium Scale Manufacturing establishment census, and the trade data from the BACI Bilateral trade flows database. Trade data Data on trade flows is compiled from the BACI bilateral trade dataset of the CEPII. 12 BACI is an international trade database providing yearly data on bilateral trade flows at the product level. There are advantages to using the BACI data over the United Nations COMTRADE data, the most commonly used in trade related studies. BACI is a harmonized version of the UN-COMTRADE data. It uses statistical approaches to reconcile the discrepancies in the reporting of trade flows between importers and exporters, and verifies the reliability of the reported flows. This harmonization yields a higher level of reliability and greater coverage of products (more than 5,000) and countries (more than 200) compared to other similar datasets (Gaulier & Zignago, 2010). For every importer-exporter-year combination, the dataset contains information on the trade quantity and trade value of each traded product for the period from 1994 to 2018. Trade flows are adjusted to 2017 US dollars. Products are classified using the Harmonized commodity description and coding System (HS) at six-digits. The data is available in all HS revisions. This study uses the 1996 revision nomenclature (HS96). The revisions of the HS nomenclature are only available for the corresponding years. Since the analysis relies on trade data from earlier years, the earlier HS revisions are more appropriate. Manufacturing data Data on the Ethiopian manufacturing sector comes from the Large and Medium Scale Manufacturing establishment census (LMSM) collected by the Central Statistical Agency (CSA) from 1996 to 2017. The data are an unbalanced panel of all registered firms engaged in the mechanical, physical, or chemical transformation of materials, substances, or components into new products. The survey is limited to establishments that employ at least ten workers (permanent and seasonal)13 and use power-driven machinery. The dataset provides information, among other things, about each firm’s capital, number of workers by occupation, sales, inputs, industry, etc. (output, capital, labor, raw material, energy inputs, and other industrial costs are all in the dataset). Both public and private 12 The dataset was downloaded from the website http://www.cepii.fr/anglaisgraph/bdd/baci.htm. 13 Firms continue to be surveyed even if their number of employees falls temporarily below 10. Firms are removed from the survey in subsequent rounds if they continue to employ fewer than 10 workers and are added back to the survey if they return to employing 10 workers. 16 firms are included. Sample construction The dataset is constructed by aggregating both the imports data and the manufacturing firm data to the 2-digit level of the International Standard Indus- trial Classification (ISIC, Revision 3.1). Although the recent literature has emphasized the benefits of firm-level analysis to control for various factors, this analysis is conducted at the industry level for three reasons. First, the research question is targeted at understanding the impact of exposure to Chinese imports on industry manufacturing employment. Second, my treatment varies at the industry level. Third, aggregating the manufacturing data at the industry level addresses some concerns inherent to firm level. In particular, the manufactur- ing data exhibits high turnover. The average annual entry rate is estimated to be 32% with an average exit rate of 35%. Moreover, roughly 16% of the manufacturing firms appear only once in the dataset over the 16-year window. Firms’ employment numbers range from 1 to more than 3,000. This pattern could be driven by firm ID entry errors.14 Aggregating the firm data at the industry level accounts for these entries and exits. To aggregate the imports data to 2-digit ISIC codes, the WITS’ crosswalk from the 6-digit HS96 product codes to the 4-digit ISIC codes is used.15 The crosswalk file includes 5,113 products, of which 4,236 imported products are successfully mapped to 34 industries (including 22 manufacturing industries). These represent manufactured goods imported by Ethiopia from China between 1996 and 2017. To construct the LMSM data at the firm level, this paper follows the cleaning steps described in Abebe et al. (2022). This yields a sample of 9,220 firms spanning 20 industries throughout the period from 2002 to 2017, with an average of 1,821 firms per year, indicating high firm turnover as mentioned above. As many as half of the firms are surveyed 5 times across the entire time period, a fact also established by other authors using the Ethiopian manufacturing dataset (Diao et al., 2021; Abreha, 2019; Gebreeyesus, 2013; Abebe et al., 2022). Imports and manufacturing data are mapped by the 2-digit industry and year. To estimate the direct impact of exposure to Chinese imports, the sample of analysis is restricted to manufacturing industries.16 As a result, the analysis excludes all non-manufacturing industries from the imports data, since they do not have any corresponding industry in the 14 In other contexts, firm mergers could also explain such a pattern. However, this is unlikely the case in Ethiopia 15 The World Integrated Trade Solution Concordance table can be downloaded from the website :https://wits.worldbank.org/product concordance.html 16 In Section 5, non-manufacturing imports are also mapped to account for indirect effects through the industries’ input-output linkages. 17 manufacturing census data. Ultimately the sample contains 352 observations (22 industries from 2002 to 2017). Excluding the observations where there was zero employment due to nonexistent domestic production, 17 my final estimation sample contains 278 observations. Data is winsorized at 1% overall. Summary statistics Summary statistics of the primary analysis sample are presented in Table 2. Panel A reports summary statistics on trade variables. Chinese imports represent 25.6% of total imports in Ethiopia. On average, each industry imported goods values at 86 million real 2017 USD over the time period of analysis. This corresponds to an average import penetration measure of 0.925, suggesting that for the average industry, the initial domestic demand was met almost fully by imports from China. Panel B reports summary statistics on the manufacturing sector. The sector remains small overall. Mean employment across industries is 7,783 with a high standard deviation of 9,518. The average number of firms across industries is 94; however, similar to employment, firm counts vary dramatically across industries. About 40% of firms are considered “large” as they employ more than 20 workers. The average industry is more capital intensive than skills intensive. The ratio of capital (including buildings) expenditures to labor costs is 5.6 on average, whereas the ratio of labor costs to capital expenditures, measuring labor intensity, is 0.5. For the average industry, the ratio of skilled workers is 0.31. The ratio of permanent workers to total workers (both permanent and seasonal) is 0.94. The average entry and exit rates within industries are 15% and 19%. 4 Estimating the Impact of Total Chinese Imports Ex- posure This section argues that exposure to Chinese imports led to an increase in manufacturing employment. The analysis estimates a manufacturing employment gain of 0.23 standard de- viation attributable to a 1 standard deviation increase in import penetration as compared to the absence of such a shock. This result is robust across alternative instruments, alternative measures of import exposure, and alternative samples. A discussion is presented on how this result diverges from the most commonly reported results in the literature on the impact of 17 The transport equipment and the computing machinery industries do not have domestic production in Ethiopia. In addition, the following industries did not start counting firms until in the later years: electrical machinery industry, the petroleum and fuel, communication equipment, medical equipment. In this sense, my analysis is relevant for the industries with domestic production. 18 Chinese import competition on manufacturing employment. 4.1 Baseline results Results on the average impact of Chinese import penetration on manufacturing employment in Ethiopia are reported in Table 3. All regressions in this table follow the main estimating equation 2. They control for year fixed effects, industry fixed effects, as well as industry controls. Standard errors in parentheses are clustered at the 2-digit industry level. The OLS estimates of equation 2, reported in column (1) of Table 3, indicate that, on average, a one unit increase in Chinese import penetration leads to a 13% increase in manu- facturing employment. This estimate of the correlation between Chinese import penetration and industry employment is statistically and economically significant. The analysis next focuses on the results from the instrumental variable approach. Col- umn (2) of Table 3 presents the results of the first stage regression. In this regression, the independent variable is Chinese import penetration in Ethiopia, and the explanatory variable is Chinese import penetration in other African economies as measured in the same way as equation 3, but replacing industry imports from China by Ethiopia with industry imports from China by other African countries. The positive and significant coefficient on the first-stage estimate demonstrates that the instrument has strong predictive power on Chinese import penetration in Ethiopia. The Kleibergen-Paap F-statistic for the excluded instrument is 150.85, greater than 104.7, indicating that my estimates are unlikely biased by weak instruments. The first-stage coefficient suggests that a unit increase in Chinese import penetration in other African economies increases Chinese import penetration in Ethiopia by 0.302 unit. The two-stage least squares estimates shown in Column (3) suggest that exposure to Chinese imports led to an increase in manufacturing employment. The Ethiopian import exposure is instrumented with Chinese imports from other African economies. The results suggest that a one unit increase in Chinese import penetration leads to a 15.2% increase in manufacturing employment. When multiplied by the mean of log(employment), this effect represents a 0.23 standard deviation increase in employment in response to a 1 standard deviation increase in import penetration. The estimated employment gain associated with a one standard deviation increase in Chinese imports penetration is 1,790. The impact is therefore statistically and economically significant. Lastly, column (4) presents the reduced form results, where the outcome variable is manufacturing employment and the explanatory variable is the instrument in equation 3. The reported coefficient indicates that Chinese im- 19 ports lead to an increase in manufacturing employment by 4.6%. This effect is economically significant and statistically significant at the 1 percent level. Table 3 indicates that the IV estimate in Column (3) is bigger than the OLS estimate in Column (1). This is consistent with a downward bias on the OLS estimates. One might expect downward bias in OLS estimates if the industries that face more import competition from China have other characteristics causing them to also have lower employment. Alter- natively, measurement errors in the Chinese imports variable may cause attenuation bias, in which case OLS estimates will be attenuated towards zero. Finally, IV estimates provide a Local Average Treatment Effect (LATE). In particular, the instrumental variable is iden- tified off of compliers, i.e. industries experiencing Chinese import competition in Ethiopia while also experiencing Chinese import competition in these same industries in other African countries. Together, these results suggest that on average, Chinese import competition caused the most exposed industries to hire more labor. Although the 2SLS impact is large in magnitude and economically significant, its impact is not very big. As presented in Table 2, the average industry employment is 7,783. Therefore, evaluated at the sample mean, one unit increase in Chinese import penetration is predicted to increase manufacturing employment by 1,183. 4.2 Robustness checks The main results presented above are robust to many checks. I run three groups of robust- ness tests on the main results. The first group addresses potential remaining endogeneity concerns from my empirical approach. The second provides evidence that my results are robust to employing alternative measures of the trade shock. The third group provides ro- bustness checks from running the analysis on different data samples. Every specification instruments the Chinese import penetration shock with a similarly constructed measure for African economies. Potential endogeneity of Chinese imports. One concern about the exclusion restric- tion assumption of the instrument is that the increase in Chinese exports to Ethiopia may be correlated with the characteristics of the Ethiopia manufacturing industries. In this case, the estimates of Chinese import penetration would be biased. As discussed in Section 2, the rise in Chinese imports in Ethiopia has more was mainly driven by reforms undertaken in China rather than domestic factors in Ethiopia. To support this hypothesis, a test for the correlation between initial employment and future growth in imports is provided. Results from Column (1) of Table B.3 suggest that initial industry employment does not predict 20 future industry growth in Chinese imports. In column (2), using initial employment growth defined as employment growth from 1996 to 2002, the results reveal a statistically significant but very small negative coefficient on future imports. Overall, this supports the idea that Chinese imports to Ethiopia are less likely driven by initial industry characteristics. Nonetheless, given that the small coefficient on the correla- tion between initial employment growth and future imports growth is statistically significant, I present results controlling for initial industry employment * trend; as well as initial industry employment in the robustness checks in Column(4) of Table B.3. The results remain robust after including the interaction term to control for potential endogeneity in the Chinese im- ports expansion. Over-identification test of instruments. The validity of my instrument is not testable. However, the test of overidentifying restrictions can be performed in the presence of an overidentified model (Bowsher, 2002). To implement this test, each country included in the instrumental variable is treated as a separate instrument. A J-test is then performed, where Chinese import penetration in Ethiopia is instrumented with Chinese import penetration in each of the 12 Sub-Saharan African countries. I test the null hypothesis that, for each instrument, the remaining instruments are exogenous. The results suggest that, for all countries, except Eritrea,18 the exclusion restriction hypothesis cannot be rejected. Table B.4 indicates that the results are robust to excluding Eritrea from the set of countries used as instruments. In addition, excluding Eritrea, the overidentification test statistics are greater than the critical value at 10% significant level, as indicated by the p-values. Alternative instrument. The tables in the Appendix subsection A.2 show results in- strumenting Chinese import penetration in Ethiopia with Chinese imports penetration in high-income countries, following Autor et al. (2013).19 This test addresses the concern that, given their regional proximity, Ethiopia and the other African economies can have an eco- nomic relationship that could affect Chinese exports to Ethiopia. In this case, one can argue that the rise in Chinese imports in Ethiopia could be correlated with Ethiopia’s domestic demand. For example, the African Continental Free Trade Area - although only recently operational - could influence Chinese imports in many African countries. When using these alternative instruments, the results remain consistent overall. How- ever, the instrument on total imports in Table B.5 fails the weak instrument test. To address 18 This could be explained by the geographic proximity of Eritrea to Ethiopia, unlike other African countries that are rather spread out across the continent and not neighboring Ethiopia. 19 Australia, Denmark, Finland, Germany, Japan, New Zealand, Spain and Switzerland. 21 this, a subset of high-income economies is used, where the corresponding instrument suc- cessfully predicts the Chinese exports to Ethiopia. 20 Using this subset, the total imports estimates yield similar results, although the reduced form estimate lacks precision. Excluding one industry at a time. Table C.11 reports the results of the test for whether specific industries drive the findings. As discussed in sections 2 and 3, there exists impor- tant variations in industry exposure to Chinese imports as well as the dynamics of industry manufacturing employment in Ethiopia. In particular, despite the overall positive correlation between employment and Chinese imports (see Figure 3), the high fluctuations in permanent employment observed across industries, especially in the textile, machinery, wearing apparel, and basic metal industries could obscure heterogeneity across industries. This test is imple- mented by running the 2SLS specification on different subsamples, excluding one industry each time. Focusing on the top and bottom importers as well as employers, columns (1) to (8) exclude the Textiles; Leather & footwear; Machinery; Chemicals, Coke, petroleum and nuclear equipment; Computing machinery; Wood; Food & beverages respectively. The 2SLS results are robust across all these specifications. More sample adjustments. Table C.12 presents the results accounting for more sample adjustments, given the limitations in the datasets in use. The 2SLS specification is performed on different subsamples. Column(1) presents the baseline results where the analysis includes years 2002 to 2017, hence excluding observations before 2002. Column(2) includes all years (1996-2017). Column(3) excludes the following industries due to lack or negligible imports and/or employment data: Tobacco products, Wood, Petroleum and nuclear fuel, Computing machinery, Communication equipment, and Medical equipment. Column(4) excludes firms where employment varies inconsistently (by more than 5 times the average over time). Col- umn(5) excludes firms that only appear once in the data. Compared to the main analysis sample, all samples provide similar results in sign, magnitude and statistical significance, except for the sample including all years available in the data. Although the coefficient remains positive, its magnitude and precision decrease. This is consistent with the fact that imports from China were almost nonexistent prior to 2002. Alternative measures of import penetration. Table C.13 captures the Chinese import competition shock through alternative measures. Instead of normalizing industry import shares with initial industry absorption, as in equation 3.1, I normalize the measure using initial industry employment as well as initial industry output. In addition, I use a more 20 This includes New Zealand, Spain and Iceland 22 direct measure of imports, the total industry imports from China, without any adjustment. All results are similar to the main results in sign and statistical significance. The magnitudes of the estimated coefficients are very similar when using total imports, whereas they differ slightly when using imports adjusted by initial employment or initial output. However, this is not inconsistent with the main results given different measures used to adjust imports. Lagged imports shock. Table C.14 presents the results of including the lagged values of the baseline import penetration as well as the import penetration measures described in the previous paragraph in the estimating equation. Lagged values of the trade shock account for two main facts. First, they account for any delay in transmission of the shock. Second, they also account for endogeneity. Results are overall similar, they are of the same sign and even larger in magnitude. The pattern of a robust and significantly positive coefficient of total Chinese imports on manufacturing employment in Ethiopia is in contrast with the experience of the high and middle income countries examined in the literature, including in North America (Autor et al., 2013; Acemoglu et al., 2016; Bernard et al., 2006), Europe (Bloom et al., 2016; Mion & Zhu, 2013), and Latin America Blyde & Fentanes (2019); Iacovone et al. (2013). In those countries, existing literature finds that Chinese imports have had disruptive impacts on domestic manufacturing employment. The next next section investigates a key feature of the Ethiopian manufacturing sector that may explain this divergence - the composition of its imported goods. 5 Separating the Competition Channel from the Input channel In light of my conceptual framework, import exposure can affect employment through two main channels: the competition channel and the inputs channel. One way to disentangle the two channels is to break down total imports into final and intermediate goods. Because imported final goods will compete with final goods produced by domestic manufacturers in Ethiopia, the impact of final good imports provides one estimate of the competition channel. Final goods are not used as inputs to production, so they will not affect the inputs channel. The effect of intermediate goods can reflect both the competition and inputs channels as these imports can both serve as inputs to domestic firms’ production process as well as compete with inputs produced and sold in Ethiopia. This is particularly true if the domestic market has important supply chain linkages. The competition channel is not expected to 23 dominate the effect of intermediate goods imports due to Ethiopia’s high reliance on such imports and the relatively low use of domestic intermediate goods. This section argues that the positive total impact of increased Chinese import exposure is dominated by the inputs channel. The analysis reveals that 78.5% of total Chinese imports in Ethiopia are made of intermediate goods. On average, a one unit increase in exposure to Chinese intermediate imports leads to a 9% increase in industry employment. This effect is statistically and economically significant. On the other hand, exposure to final goods imports has no detectable impact on manufacturing employment. 5.1 Decomposing total imports by end use The end use category of imported goods is determined by mapping the imports dataset at the 6-digit product level to the Broad Economic Categories (BEC) classification. This classification yields two categories of goods: final goods and intermediate goods. The initial BEC end use categories include capital goods as well. This analysis groups the intermediate goods together with the capital goods because both types of goods enter in the production process. The proportion of imported capital goods is relatively small (13%).21 Only 0.8% of all imports are unclassified. The decomposition reveals that on average, 78.5% of total Chinese imports to Ethiopia from 2002 to 2017 were intermediate goods. Figure 5 displays the evolution of the share of each category of imports over the total value of Chinese imports. The proportion of inter- mediate imports by Ethiopia increased over time, reaching a peak of 89.1% in 2008. When compared to other countries, this composition of imports is relatively high in intermediate goods. For example, over the same time period, the average share of intermediate imports over total Chinese imports in the United States was 51.3% (see Figure A.4). When compared to other import origins, the composition of Chinese imports is proportionate to European imports (Figure A.3). Ethiopia imports a higher share of final goods from the rest of Asia.22 This is consistent with the fact that Ethiopia has been substituting its imports away from Europe (Figure 2), probably due to lower prices. The large majority of goods produced in Ethiopia are final goods. Although the LMSM data does not allow me to neatly distinguish between the Ethiopian producers of intermediate goods and the producers of final goods, an inspection of Table B.1 suggests that the large 21 A few example of goods classified as capital goods: machines (which constitutes the majority of capital goods), electrical transformers, radiators, turbines, engines. 22 Major countries are India, the United Arab Emirates and Viet Nam 24 majority of imported intermediates are not produced in Ethiopia. For this reason, the competition channel is expected to be less prevalent in Ethiopia when it comes to imported intermediate goods. 5.2 Impact of final and intermediate goods imports After decomposing Chinese imports into intermediate and final goods, their impacts on man- ufacturing employment are jointly estimated. A new measure of Chinese import penetration is constructed by first aggregating the imports of final or intermediate goods by industry and year. These import types are then adjusted by the initial absorption, following the approach outlined in equation 3.1. For example, consider the textile industry. Goods such as mattresses, blankets, towels, rugs, tents, bedspread, and other similar goods imported for retail sale will be classified as textile final goods. Conveyor belts, staple fibers, woven fabrics, and other similar unfinished fabrics not destined for retail sale but used in the preparation, spinning, or weaving of textiles will be classified as textile intermediate goods. Table 4 presents the breakdown of intermediate and final goods import shares across industries. As shown in the table, for the majority of industries, the largest share of imports are intermediate goods. The exception is for the wearing apparel, tobacco, as well as the leather and footwear industries where the share of intermediate imports was zero or close to zero. To evaluate the separate direct impact of final good and intermediate goods imports, I estimate the following regression: Yit = β1 IP P IntermediateChina it China + β2 IP P F inalit + Xit β3 + θt + θi + eit (4) China where IP P F inalit is the Chinese import penetration in Ethiopia for final goods exclu- sively, for industry i and year t. It is computed by aggregating all products classified as final goods within the 2-digit industry. Similarly, IP P IntermediateChina it is the Chinese import penetration in Ethiopia for intermediate goods exclusively, for industry i and year t. Using the BACI dataset, which provides product-level information across all available coun- tries, separate instruments are constructed for final good imports and intermediate goods imports. Specifically, Chinese final good imports to Ethiopia are instrumented with Chinese final good imports to other African economies. Similarly, Chinese intermediate goods im- ports to Ethiopia are instrumented with Chinese intermediate goods imports to other African economies. All other terms are the same as in equation 2. The coefficients of interest are β1 and β2 , which provide estimates for the percentage change in industry employment as- 25 sociated with an industry-level increase in Chinese import penetration for intermediate and final goods respectively. Overall, findings suggest that the positive results are driven by Chinese intermediate goods imports. Table 5 presents the estimates of β1 and β2 from equation 4, where Chinese intermediate goods imports to Ethiopia are instrumented by the Chinese intermediate goods imports to other African economies. Similarly, Chinese final good imports to Ethiopia are instrumented by the Chinese final good imports to other African economies. column (1) presents the OLS results where the effect of the trade shock is positive and statistically significant for the Chinese intermediate imports and positive and statistically insignificant for the Chinese final imports. column (2) shows a strong first stage on Chinese intermediate imports, with the F-statistic of 109.8 allowing the rejection a weak instrument. Likewise, column (3) shows a strong first stage on Chinese final imports, with the F-statistic of 212.85 allowing the rejection a weak instrument. column (4) presents the 2SLS coefficient, which is positive and statistically significant for the Chinese intermediate imports and negative and statistically insignificant for the Chinese final imports. The findings suggest that a unit increase in Chinese intermediate good import penetration leads to 19.7% increase in industry employment. The standardized effect associated to this impact is 0.20. The standardized effect associated with the impact of final goods is -0.12. This suggests that the competition channel is not likely driving the aggregate positive impact on employment. In sum, these results suggest that it is unlikely that the competition channel is driving the positive effect on employment demonstrated in Section 4. Because the estimated impact of intermediate goods is so much larger and more signifi- cant than that of final goods, the overall positive impact of Chinese imports on employment is likely driven by the import of intermediate goods.The intermediate goods estimates could still contain both the inputs and the competition effects. The estimates could be driven by input effects if each industry is importing a large share of intermediate inputs also classified in that industry. This would reflect competition effects if domestic industries manufacture and sell intermediate goods to other domestic industries or to export markets. There are a few ways for separating out these two effects. First, it is important to know whether domestic industries produce intermediate manu- factured goods that they sell in the domestic market or not. If they do, then competition may also exist over intermediate goods. One way to check this is by examining the products sold in Ethiopia to separate out the producers of intermediate inputs and the producers of final goods. Unfortunately, it is not possible to determine the end use category of domestically manufactured goods from the Ethiopian manufacturing census data. The manufacturing cen- 26 sus does not consistently provide information on the products sold by firms. In addition, the data is not detailed enough to determine whether the goods are intermediate or final goods. Nonetheless, the large majority of intermediates are not produced in Ethiopia. Therefore, these estimates are unlikely to provide competition effects on imported intermediate goods. We can also find suggestive evidence by analyzing the input-output table. The input- output table provides the distribution of aggregate demand between intermediate demand from firms and final demand from households. The 2005 input-output table indicates that only 24.3% of aggregate demand is served in the intermediate market (including the non- manufacturing sectors). The remaining 75.8% is allocated to household consumption. Within the manufacturing sector, the share of own-industry input usage is relatively high (see Figure 7). For example, as shown in Figure 6, about 20% of the textile industry inputs are supplied by the textile industry. Note that this percentage is higher (about 57%) when restricted to the manufacturing sector. Because the manufacturing sector in Ethiopia is underdeveloped, the majority of manufacturing inputs are domestically sourced from the agriculture and services sectors. Overall, 55% of its total inputs are supplied by the agriculture and mining sector, 15% from the service sector, and only 30% from the manufacturing sector (including more than half from the textile industry). This breakdown implies that, relative to the competition channel, the inputs channel is likely dominant. Lastly, the competition and inputs channels are distinguished by examining the indirect effect through the supply chain to compute an estimate of input usage. Instead of assigning intermediate imports from each industry only to the corresponding domestic industry, I adjust the exposure measure by accounting for imports from all other industries entering the production process of the domestic industry. For example, since the textile industry sources inputs from the metals and the rubber industries among others, my textile import exposure measure will account for imports from metals and rubbers according to their usage shares in the textile industry. This method is preferred for identifying the inputs channel as it most comprehensively identifies the impact of imported intermediate goods throughout the supply chain. Its implementation is detailed in the next subsection. 5.3 Accounting for input usage To estimate domestic intermediate input usage, I proportionally allocate imports of inter- mediate inputs to industries using the input-output table. Imports of intermediate inputs are distributed across industries based on the industry’s input usage share in total interme- diate usage. This allocation assumes that industry patterns of input usage are the same for 27 imported goods and domestic goods. This approach allows me to estimate the impact on the textile industry when its domestic suppliers (i.e the metals or rubber industries) import from China. Chinese import shocks are allocated to industries as follows: Imports usageChina it = αji ∗ intermediate importsChina jt (5) j where i is input purchaser industry, j the input supplier industry, and αji the share of input j in total inputs of industry i. The aggregate intermediate imports for each industry correspond to the sum of industry imported intermediate goods. The input channel impact of Chinese intermediate is then estimated as follows: Yit = β1 IP P usageChina it + Xit β2 + θt + θi + ef it (6) where IP P usageChina it is the import penetration measure computed with China Imports usageit . I use the input-output table from the 2003 Kenyan Social Account Matrix (SAM) constructed by the International Food Policy Research Institute23 . One thing to keep in mind is that the above measure of input usage includes own indus- try input usage. Computationally, the adjusted measure of intermediate imports includes the main diagonal elements of the input-output table. For this reason, the unadjusted (di- rect intermediates exposure) and the adjusted (intermediate usage) measures of imported intermediates are not included in the same regression. However, as a robustness check, I run the regression including the direct impact while excluding the self-consumption when computing the I-O input shares. Results are presented in Table 6. The four columns present the same information as in previous tables. Estimating the impact through the entire supply chain reveals that the majority of the effect observed in the direct impact is driven by the inputs channel. The 2SLS estimate in column (3) can be interpreted as a 14.1% change in manufacturing employment in response to a unit increase in Chinese intermediate good import penetration. The corresponding standard deviation increase in employment is 0.34, which is a bigger impact than the direct impact of the unadjusted imports exposure measure estimated in the main analysis. To further support the inputs channel, I estimate this impact on two subsamples in Table 7. The first sample, in column (1), contains the firms that report using imported inputs in 23 Given the data limitations from the Ethiopia Input-Output table, I use an alternative country source. The major limitation from the Ethiopian I-O table is that it lacks the linkages. For example, some industries that are a natural source of inputs do not show up. 28 their production process. If the industries are benefiting from Chinese intermediate inputs, then one could hypothesize that the results should hold for input importers and not for others. As a placebo test, this regression is examined using a sample of domestic input users. The analysis aggregates only firms that have reported using imported inputs at least once, which is about 70% of firms. As shown in Table 2, at the industry level, the ratio of imported inputs cost to total inputs cost averages 60%. Values ranges from 20% to 100% with a median of 74%. For the placebo test, I aggregate only firms that never reported using imported inputs in their production function. Consistent with the inputs channel, results suggest that the impact on imported inputs users is positive, larger in magnitude than the full sample, and statistically significant. The estimates for the sample of firms with no imported inputs show no impact of imported intermediates on employment in these firms. Because the effect is concentrated in firms that report importing intermediate goods, it follows that the technology embodied in the imported inputs (Amiti & Konings, 2007) are beneficial to firms. In fact, imports of inputs from more developed economies are considered to be of higher quality (Bas & Strauss-Kahn, 2014; Feng et al., 2016). This result is consistent with the fact that in 2001, shortages of raw materials and other issues related to difficulty accessing inputs were listed among the top reasons that prevented Ethiopian manufacturing firms from operating at full capacity.24 The next section presents results on how the shortage of raw materials responded to the China shock. 6 Mechanisms The previous section demonstrated that exposure to Chinese imports is positively affecting employment through the inputs channel. These results can be explained by a few possible mechanisms. Three mechanisms through which imported inputs can lead to increases in em- ployment are tested: total factor productivity (TFP), skills upgrading, and within-industry reallocation. I also examine the heterogeneous impact of Chinese imports based on industry characteristics. Finally, supporting evidence suggests that firms are substituting away from traditional trade partners and towards cheaper Chinese imports. The analysis suggests that in Ethiopia, Chinese inputs increase employment through the positive impact on industry total factor productivity and increases in capacity utiliza- tion. The results indicate a positive impact on skills upgrading for the industries affected through the downstream shock. This suggests that, with the assumption of complementarity 24 Source: author’s calculation using the LMSM census 29 between labor and inputs, the use of higher quality inputs from China might require firms to employ more skilled workers. The findings do not indicate evidence of within-industry reallocation in response to Chinese imports. Finally, the heterogeneity analysis reveals that the employment impact of Chinese imports exposure is driven by large firms and more labor intensive industries. Whereas ownership type does not influence this impact. 6.1 Productivity There are several hypotheses of how trade impacts firm productivity. This section focuses on the impact on productivity driven by access to imported inputs. Several empirical studies provide evidence that access to cheaper and better intermediate inputs is a source of increased efficiency and productivity (Grossman & Helpman, 1991; Amiti & Konings, 2007; Topalova, 2007). Topalova & Khandelwal (2011) assert that the intermediate imports are particularly beneficial for developing countries facing limited access to technology and better inputs. Industry-level productivity is determined by computing the weighted average of firm- level productivity. To estimate firm-level productivity, I assume a Cobb-Douglas production function for the firm. The log-linear transformation of the production function is given by: yit = β0 + βk kit + βl lit + ωit + ηit (7) where yit is the logarithm of the firm’s value of production, kit is the logarithm of the capital input, and lit the logarithm of labor input. ωit and ηit are error terms representing firm productivity. The first error term captures the unobservable productivity shocks that are endogeneous to input choice. ηit is the productivity shocks that are uncorrelated with input choice. Firm productivity is obtained by estimating the production function and extracting the estimation residual. Olley & Pakes (1996) points to two challenges for estimating firm pro- ductivity. First is the simultaneity bias, which arises from the correlation between the firm’s productivity and their choice of inputs. For example, there could be a positive correlation between productivity and labor, such that high productivity firms will employ more workers. Second is the selection bias which arises from the negative correlation between the firms’ capital and their probability of exit. For example, low productivity firms endowed with larger capital will be less likely to exit the market. To address the simultaneity and selection bias in estimating the production function, this analysis follows Levinsohn & Petrin (2003). This estimation approach addresses the bias by 30 using material inputs as a proxy for the unobservable productivity shocks that are correlated with input choice. To support the validity of material inputs as a proxy, Levinsohn & Petrin (2003) show that the demand function of intermediate inputs is monotonically increasing in the materials demand function to be inverted as follows: ωit = ωit (kit , mit ) (8) Substituting this function in the production function, we can rewrite: yit = β0 + βk kit + βl lit + βm mit + ωit + ηit (9) = βl lit + φit (kit , mit ) + ηit (10) φit (kit , mit ) = β0 + βk kit + +βm mit + ωit (kit , mit ) (11) The estimation of the production function is done in two steps. In the first step, β ˆl is estimated using a third-order polynomial approximation of kit and mit . With β ˆl , φ can also be estimated as follows: φˆ ˆ it = yit − βl lit . Next, βk and βm are estimated using the Generalized Method of Moments approach. The details about these steps can be found in Levinsohn & Petrin (2003) and Petrin et al. (2004). The 2SLS results are reported in Table 12, Column (1). The impact of intermediate input usage is estimated by restricting the sample to industries aggregated across importing firms. The results indicate that a one-unit increase in Chinese import penetration increases productivity by 23.1%. This result is in line with a large body of empirical literature which provides evidence that a decline in input tariffs or imports of intermediate goods leads to productivity gains (Goldberg et al., 2010; Abreha, 2019; Goldberg et al., 2010; Redding et al., 2006; Nocke & Yeaple, 2006; Topalova & Khandelwal, 2011; Topalova, 2007; Amiti & Konings, 2007; Kasahara & Rodrigue, 2008). 6.2 Capacity utilization In early years of the survey, a greater share of firms reported lack of access to raw materials among the top reasons preventing them from operating at full scale. The results indicate that Chinese imports of intermediate inputs led to a decline in the share of firms reporting lack of access to raw materials as a constraint to operating at full capacity (Table 8), while also increasing firms capacity utilization (Table 12, Column (2)). 31 6.3 Skills upgrading Skills upgrading can be a result of pro-competitive or input effect. On the one hand, the literature reports that increased competition may lead to a decline in low-skilled workers, or to higher investments in innovation, which in turn will increase the relative share of skilled workers Mion & Zhu (2013); Grossman & Rossi-Hansberg (2008). On the other hand, trade can lead to skill-biased technological change (Bloom, 2011). The results indicate that Chinese imports of intermediate inputs led to an increase in skilled workers (Table 12, Column (3)). This could suggest that the imported inputs incorporate higher technology that require firms to hire skilled labor, under the assumption that labor and inputs are complementary. 6.4 Within-industry reallocation The trade literature underlines the role of between industry reallocation patterns of manu- facturing firms in response to trade shocks (Melitz, 2003; Bernard et al., 2006). This section discusses the within-industry entry and exit in response to increased Chinese imports 25 . Table 12, Columns (5) and (6) display the relationship between firm exit and import compe- tition. The 2SLS results suggest that Chinese import exposure had no effect on the average rate for firm exit or entry within the industry. 6.5 Heterogeneity Another way to learn about the underlying mechanisms for Chinese import exposure effects is by identifying how the industry employment effects differ across sub-groups of firms based on certain characteristics. I examine the impact across firms of different sizes, production factor intensity, and ownership status. Firms are classified as either large (employ at least 50 workers) or small (employ less than 20 workers); labor intensive (above the 50th percentile in the labor to capital expenses ratio); private or public. For each subgroup of firms, I ag- gregate across industries and estimate the impact of Chinese imports usage on employment, restricting to firms using imported inputs. The results are presented in Table 13. They provide evidence that the imported in- puts disproportionately benefited large firms (Columns 1 and 2). This is consistent with the evidence provided by Abreha (2019) on selection to importing among large Ethiopian firms. Similarly, labor intensive industries benefited more than capital intensive industries (Columns 3 and 4). Columns (5) and (6) indicate that there is no significant difference in how firms of different ownership were affected by Chinese imports of intermediate goods. 25 In the context of Ethiopia, intra-industry switching of firms is extremely rare. 32 6.6 Price effects What is so special about the Chinese imports? I provide supporting evidence of a strong inverse correlation between imported inputs in the same industry coming from China and the imported inputs coming from Europe, America and Asia (excluding China). In addition, the results indicate that firms are substituting towards cheaper Chinese imports. Figure 8 displays the estimates of equations 12 and 13 capturing the differences in quantities and prices of imports originating from China and those originating from the rest of the world respectively. Chinait =β0 + β1 Europeit + β2 Asiait + β3 Americait + θt + θi + it (12) P riceict =β0 + β1 China + θt + θi + θc + it (13) where Chinait from equation 12 is the Chinese imports (measured in metric tonnes) to Ethiopia for industry i and year t. Similarly, Europeit , Asiait and Americait are the imports to Ethiopia from Europe, Asia (excluding China) and America respectively. θt is the year fixed effect and θi the industry fixed effect. In equation 13, P riceit is the annual average price of imports from region c in industry i. θc is the region fixed effects. Regions include Europe, Asia and America. This substitution away from traditional partners suggests that the rise of China has been an opportunity for firms to access more affordable inputs. 7 Conclusion In Ethiopia, Chinese imports have risen rapidly over the past two decades. This paper evaluates the impact of increased exposure to Chinese imports on Ethiopian industry-level manufacturing employment. The analysis is conducted within the broader ”China shock” literature, which generally reports adverse effects of Chinese import competition on manu- facturing employment, particularly in high-income countries. To causally estimate this effect, Ethiopia’s imports from China are instrumented using imports from China by other African countries. The paper’s main results indicate that, during the period of analysis from 2002 to 2017, Ethiopian industries with higher exposure to Chinese imports experienced greater employment growth. On average, a one-unit increase in industry import penetration led to a 15.2 percent increase in industry employment. This effect is both economically and statistically significant, with the results remaining robust across multiple checks, including alternative instruments, measures of import exposure, and sample specifications. 33 The impact is disentangled into two competing channels: the competition channel and the inputs channel. This distinction is made by separating imports of final goods from im- ports of intermediate goods. The findings suggest that the inputs channel, rather than the competition channel, is the primary driver of the observed results. Specifically, employment gains are driven by the imports of intermediate goods rather than final goods. The key mechanisms through which these employment gains occur are increased industry productiv- ity and capacity utilization. Ethiopian industries that utilize Chinese intermediate inputs experience productivity improvements, enabling them to expand employment, particularly for higher-skilled workers. The analysis also provides evidence that firms are shifting away from traditional trade partners and increasing their reliance on cheaper Chinese imports. This shift may reflect firms’ strategic efforts to reduce production costs while enhancing competitiveness and capacity utilization. Compared to existing estimates in the literature, the positive and significant effect of imports on manufacturing employment is a novel result. While most of the empirical literature on Chinese import competition has focused on the competition effects, the broader trade literature emphasizes the important role for intermediate inputs. Decomposing the imports between the final and intermediate goods, and further accounting for input-output linkages across industries, reveals the inputs channel is critical in generating employment gains in Ethiopia. The findings from this paper have important implications for Ethiopia’s trade and in- dustrialization strategy. The evidence highlights the potential benefits of exposure to im- ported intermediate inputs, which enhance productivity, capacity utilization, and employ- ment. These results point to the value of access to high-technology inputs and productivity- enhancing materials that may be difficult to produce domestically. The findings are consis- tent with global evidence on the role of imported inputs in supporting firm-level productivity growth and employment. While these findings suggest potential benefits from imported inputs, they should be interpreted within the broader context of Ethiopia’s industrialization strategy, which empha- sizes import substitution (Ministry of Industry, 2013, 2023). Rather than viewing import substitution and access to imported inputs as opposing strategies, there may be scope to consider them as complementary approaches. One possible approach could be to support ongoing import substitution efforts while reducing barriers to importing essential inputs and materials that enhance domestic productivity. This could include measures such as facili- tating access to foreign exchange, which would enable firms to source critical inputs. At the same time, efforts could be directed toward fostering the domestic production of final goods 34 in key targeted industries. Given that Ethiopia is not at the frontier of technology, there may be efficiency gains from sourcing certain critical inputs and productivity-enhancing materials through imports, rather than attempting to produce them domestically. This could include high-technology inputs and capital goods, although the role of capital goods is not explored in detail in this paper. Such a strategy would allow Ethiopia to leverage the productivity benefits associated with high-quality inputs while simultaneously supporting local production, employment, and economic diversification. While these findings are specific to Ethiopia, they have broader relevance for other African economies, many of which share similar structural characteristics. Like Ethiopia, several African countries are still in the early stages of industrialization, with relatively low technological capacity for producing high-quality inputs domestically. The results suggest that access to imported intermediate inputs may serve as a key mechanism for enhancing employment and productivity in these contexts. This could be particularly useful for coun- tries pursuing regional industrial policies or those involved in regional trade agreements, such as the African Continental Free Trade Area (AfCFTA), which may present opportunities for sourcing high productivity inputs and materials from within Africa. For countries with a strong focus on import substitution, a complementary approach that allows for the import of critical inputs while supporting the domestic production of final goods could promote in- dustrial development. Nonetheless, the relevance of these findings may vary depending on a country’s industry structure, the availability of domestic inputs, and the scope of its import substitution strategy. Further research could explore the extent to which these findings hold in other African economies. 35 References Abebe, G., McMillan, M., & Serafinelli, M. (2022). Foreign direct investment and knowledge diffusion in poor locations. Journal of Development Economics , 158 , 102926. Abreha, K. G. (2019). Importing and firm productivity in ethiopian manufacturing. The World Bank Economic Review , 33 (3), 772–792. Acemoglu, D., Autor, D., Dorn, D., Hanson, G. H., & Price, B. (2016). Import competition and the great us employment sag of the 2000s. Journal of Labor Economics , 34 (S1), S141–S198. Aghion, P., Bergeaud, A., Lequien, M., Melitz, M. J., & Zuber, T. (2024). Opposing firm- level responses to the china shock: Output competition versus input supply. American Economic Journal: Economic Policy , 16 (2), 249–269. Aghion, P., Bloom, N., Blundell, R., Griffith, R., & Howitt, P. (2005). Competition and innovation: An inverted-u relationship. The quarterly journal of economics , 120 (2), 701– 728. Amiti, M., & Konings, J. (2007). Trade liberalization, intermediate inputs, and productivity: Evidence from indonesia. American Economic Review , 97 (5), 1611–1638. Atalay, E. (2017). How important are sectoral shocks? American Economic Journal: Macroeconomics , 9 (4), 254–80. Autor, D. H., Dorn, D., & Hanson, G. H. (2013). The china syndrome: Local labor market effects of import competition in the united states. American Economic Review , 103 (6), 2121–68. Bas, M., & Strauss-Kahn, V. (2014). Does importing more inputs raise exports? firm-level evidence from france. Review of World Economics , 150 (2), 241–275. Bernard, A. B., Jensen, J. B., & Schott, P. K. (2006). Survival of the best fit: Exposure to low-wage countries and the (uneven) growth of us manufacturing plants. Journal of international Economics , 68 (1), 219–237. Bigsten, A., Gebreeyesus, M., & S¨oderbom, M. (2016). Tariffs and firm performance in ethiopia. The Journal of Development Studies , 52 (7), 986–1001. Biscourp, P., & Kramarz, F. (2007). Employment, skill structure and international trade: Firm-level evidence for france. Journal of International Economics , 72 (1), 22–51. 36 Bloom, N. (2011). Cep discussion paper no 1000 january 2011 trade induced technical change? the impact of chinese imports on innovation, it and productivity nicholas bloom, mirko draca and john van reenen. Bloom, N., Draca, M., & Van Reenen, J. (2016). Trade induced technical change? the impact of chinese imports on innovation, it and productivity. The review of economic studies , 83 (1), 87–117. Blyde, J. S., & Fentanes, O. (2019). The heterogeneous impacts of import competition on mexican manufacturing plants (Tech. Rep.). IDB Working Paper Series. Bowsher, C. G. (2002). On testing overidentifying restrictions in dynamic panel data models. Economics letters , 77 (2), 211–220. Diao, X., Ellis, M., McMillan, M. S., & Rodrik, D. (2021). Africa’s manufacturing puzzle: Evidence from tanzanian and ethiopian firms (Tech. Rep.). National Bureau of Economic Research. Ebenstein, A., Harrison, A., McMillan, M., & Phillips, S. (2014). Estimating the impact of trade and offshoring on american workers using the current population surveys. Review of Economics and Statistics , 96 (4), 581–595. Edwards, L., & Jenkins, R. (2015). The impact of chinese import penetration on the south african manufacturing sector. The Journal of Development Studies , 51 (4), 447–463. Erten, B., Leight, J., & Tregenna, F. (2019). Trade liberalization and local labor market adjustment in south africa. Journal of International Economics , 118 , 448–467. Feenstra, R. C. (2003). A homothetic utility function for monopolistic competition models, without constant price elasticity. Economics Letters , 78 (1), 79–86. Feng, L., Li, Z., & Swenson, D. L. (2016). The connection between imported intermediate inputs and exports: Evidence from chinese firms. Journal of International Economics , 101 , 86–101. Fenta, K. (2014). Industry and industrialization in ethiopia: Policy dynamics and spatial distributions. European Journal of Business and Management , 6 (34), 326–344. Gaulier, G., & Zignago, S. (2010). Baci: International trade database at the product-level. the 1994-2007 version (Working Papers No. 2010-23). CEPII. Retrieved from http:// www.cepii.fr/CEPII/fr/publications/wp/abstract.asp?NoDoc=2726 37 Gebreeyesus, M. (2013). Industrial policy and development in ethiopia: Evolution and present experimentation (Tech. Rep.). WIDER working paper. Goldberg, P. K., Khandelwal, A. K., Pavcnik, N., & Topalova, P. (2010). Imported interme- diate inputs and domestic product growth: Evidence from india. The Quarterly journal of economics , 125 (4), 1727–1767. Grossman, G. M., & Helpman, E. (1991). Trade, knowledge spillovers, and growth. European economic review , 35 (2-3), 517–526. Grossman, G. M., & Rossi-Hansberg, E. (2008). Trading tasks: A simple theory of offshoring. American Economic Review , 98 (5), 1978–97. Halpern, L., Koren, M., & Szeidl, A. (2015). Imported inputs and productivity. American Economic Review , 105 (12), 3660–3703. Iacovone, L., Rauch, F., & Winters, L. A. (2013). Trade as an engine of creative destruction: Mexican experience with chinese competition. Journal of International economics , 89 (2), 379–392. Ianchovichina, E., & Martin, W. (2001). Trade liberalization in china’s accession to wto. Journal of Economic Integration , 421–445. IMF, I. M. F. (2016). 2016 article iv consultation - press release; staff report; and statement by the executive director for the federal democratic republic of ethiopia (Tech. Rep.). IMF Country Report No. 16/322. Kasahara, H., & Rodrigue, J. (2008). Does the use of imported intermediates increase productivity? plant-level evidence. Journal of development economics , 87 (1), 106–118. Leamer, E. E., et al. (1995). The heckscher-ohlin model in theory and practice. International Finance Section, Department of Economics, Princeton University . . . . Lee, D. S., McCrary, J., Moreira, M. J., & Porter, J. (2022). Valid t-ratio inference for iv. American Economic Review , 112 (10), 3260–90. Levinsohn, J., & Petrin, A. (2003). Estimating production functions using inputs to control for unobservables. The review of economic studies , 70 (2), 317–341. Medina, P. (2022). Import competition, quality upgrading and exporting: evidence from the peruvian apparel industry. The Review of Economics and Statistics . doi: https://doi.org/ 10.1162/rest a 01221 38 Melitz, M. J. (2003). The impact of trade on intra-industry reallocations and aggregate industry productivity. econometrica , 71 (6), 1695–1725. Ministry of Industry. (2013). Ethiopian industrial development strategic plan. (Accessed: 12/02/2024) Ministry of Industry. (2023). National import substitution strategy for selected manufacturing industry subsectors in ethiopia. https://www.moi.gov.et/sites/ default/files/resource/Import%20Substitution%20Strategy Final VF.pdf. (Ac- cessed: 12/02/2024) Mion, G., & Zhu, L. (2013). Import competition from and offshoring to china: A curse or blessing for firms? Journal of International Economics , 89 (1), 202–215. Nocke, V., & Yeaple, S. (2006). Globalization and endogenous firm scope. National Bureau of Economic Research Cambridge, Mass., USA. Olley, S., & Pakes, A. (1996). The dynamics of productivity in the telecommunications equipment industry. econometrica , 64 (6), 1263-98. Petrin, A., Poi, B. P., & Levinsohn, J. (2004). Production function estimation in stata using inputs to control for unobservables. The Stata Journal , 4 (2), 113–123. Pierce, J. R., & Schott, P. K. (2016). The surprisingly swift decline of us manufacturing employment. American Economic Review , 106 (7), 1632–62. Raith, M. (2003). Competition, risk, and managerial incentives. American Economic Review , 93 (4), 1425–1436. Redding, S. J., Bernard, A. B., & Schott, P. K. (2006). Multi-product firms and product switching. Revenga, A. L. (1992). Exporting jobs? the impact of import competition on employment and wages in us manufacturing. The Quarterly Journal of Economics , 107 (1), 255–284. Rodrik, D. (2018). An african growth miracle? Journal of African Economies , 27 (1), 10–27. Schmidt, K. M. (1997). Managerial incentives and product market competition. The review of economic studies , 64 (2), 191–213. Taniguchi, M. (2019). The effect of an increase in imports from china on local labor markets in japan. Journal of the Japanese and International Economies , 51 , 1–18. 39 Topalova, P. (2007). Trade liberalization, poverty and inequality: Evidence from indian districts. In Globalization and poverty (pp. 291–336). University of Chicago Press. Topalova, P., & Khandelwal, A. (2011). Trade liberalization and firm productivity: The case of india. Review of economics and statistics , 93 (3), 995–1009. World-Bank. (2021). Manufacturing, value added (% of gdp) - ethiopia. World Development Indicators , The World Bank Group. Yang, M.-J., Li, N., & Lorenz, K. (2021). The impact of emerging market competition on innovation and business strategy: Evidence from canada. Journal of Economic Behavior & Organization , 181 , 117–134. 40 Figures and Tables Figures Figure 1: Chinese manufacturing imports and FDI to Ethiopia Notes: The value of Chinese FDI inflows includes all sectors, not just manufacturing. The value of manu- facturing FDI in Ethiopia includes all sources of FDI, not just Chinese. Sources: (1) BACI database 1996-2017, CEPII; (2) The Statistical Bulletin of China’s Outward Foreign Di- rect Investment covering 2003-2019 published by China’s Ministry of Commerce (MOFCOM); (3) Ethiopia Central Statistical Agency’s annual survey of Large and Medium Scale Manufacturing (LMSM) 1996-2017 41 Figure 2: Imports from China and the ROW in USD (2017 millions) Notes: Each bar represents the total value of imported goods by origin and year. Green bars refer to year 1998, Orange bars to year 2002, and blue bars to year 2016. Graph constructed using data from the BACI database. Imports share by origin and years Year Africa America Asia China Europe 1998 2.70 11.77 26.09 6.48 52.21 2002 8.93 5.46 38.54 7.89 38.85 2016 5.90 9.56 34.23 27.96 22.26 Notes: Each row represents the share of imports from each origin on total imports of the year. Table constructed using data from the BACI database. 42 Figure 3: Manufacturing employment and Chinese import penetration over time Notes: Green line is manufacturing employment, measured as the yearly permanent employment across all manufacturing industries. It uses the left axis and is measured in thousands 2017 USD. Orange line is the average Chinese import penetration by year. It uses the right axis. 43 Figure 4: First stage Notes: Graph plots the fitted line of the first-stage regression where the dependent variable is the predicted Chinese import penetration in Ethiopia and the explanatory variable is the Chinese import penetration in African economies. The corresponding Kleibergen-Paap F-statistic is 150.86 44 Figure 5: Decomposition of Chinese imports across intermediate and consumption goods Notes: Each row represents the share of imports from each origin on total imports of the year. Table constructed using data from the BACI database. 45 Figure 6: Textile industry I-O input shares Source: ETH 2005/2006 input-output table 46 Figure 7: Industry share of own input usage Food & beverages Tobacco products Textiles Wearing apparel Leather & footwear Wood Paper Chemicals Non-metallic Minerals Basic metals Machinery Electrical machinery Vehicles Furniture and others 0 .2 .4 .6 .8 1 Share of own input usage Source: ETH 2005/2006 input-output table 47 Figure 8: Evidence that firms are substituting away from traditional trade partners and towards cheaper Chinese imports 0 0 -50 -.2 Point estimates Point estimates -100 -.4 -150 -.6 -200 -.8 Europe Asia (excl. China) America Inputs Final Inputs Final (a) Imports (qty) from China and ROW (b) Price differences 48 Tables Table 1: Industry composition of employment and Chinese imports (2002-2017) Industry Employment Imports share Mean ∆ Emp. Mean ∆ Imp. share Food & beverages 0.276 0.004 4.86 44.12 Tobacco products 0.007 0.001 –4.14 200.84 Textiles 0.146 0.100 1.53 10.34 Wearing apparel 0.061 0.080 21.40 29.54 Leather & footwear 0.080 0.019 6.38 16.72 Wood 0.016 0.008 12.24 35.98 Paper 0.017 0.006 7.92 18.98 Printing 0.044 0.005 2.44 74.73 Petroleum & fuel 0.000 0.010 . 51.03 Chemicals 0.057 0.054 7.19 18.94 Rubber & plastics 0.069 0.044 9.38 23.38 Non-metallic Minerals 0.093 0.018 11.14 25.34 Basic metals 0.027 0.060 19.13 44.40 Fabricated metals 0.040 0.081 15.95 38.22 Machinery 0.003 0.154 739.39 32.80 Computing machinery 0.000 0.010 –100.00 65.48 Electrical machinery 0.001 0.124 86.60 29.21 Communic. equipment 0.001 0.117 39.21 230.19 Medical equipment 0.000 0.015 206.65 34.24 Vehicles 0.017 0.058 21.73 62.08 Transport equipment 0.000 0.017 . 148.68 Furniture and others 0.045 0.015 7.61 17.02 Notes : Employment average is the average industry employment from 2002 to 2017. Employment share is the average value of industry employment in total manufacturing employment in each year. Imports average is the average industry imports from 2002 to 2017. Imports share is the average value of industry Chinese imports in total manufacturing imports in each year. Imports are expressed in USD (2017 millions), employment values are in thousands. Table constructed using data from BACI and LMSM 49 Table 2: Summary statistics (1) (2) (3) VARIABLES Mean Sd N Panel A: Trade variables Chinese imports 86,164 12.06 278 Chine import penetration 0.925 1.485 278 Initial absorption 142,177 138,823 278 Initial sales 86,656 140,139 278 Initial total imports 61,101 65,886 278 Initial total exports 5,580 12,156 278 Imports from America 0.0923 0.150 278 Imports from Asia 0.835 0.768 278 Imports from Europe 0.644 0.718 278 Panel B: Manufacturing variables Permanent workers 7,783 9,518 278 Output 243,304 413,897 278 Share of imported inputs 0.591 0.287 278 Capital intensity 5.660 3.894 278 Labor intensity 0.509 2.899 278 Skill intensity 0.310 0.113 278 Ratio permanent workers/total workers 0.935 0.0710 278 Average number of firms 93.70 131.3 278 Average share of large firms 0.397 0.270 278 Entry rate 0.150 0.223 278 Exit rate 0.192 0.300 278 Notes : Monetary values in thousands real USD (base year = 2017). Covers years 2002-2017. Deflator from the WDI MFG value added. Initial refers to year 1998. Per worker measures are computed using permanent workers. Skill intensity is measured as the ratio of non-production workers over production workers (excludes unpaid, apprentice and seasonal). Capital intensity is the ratio of capital to production workers. 50 Table 3: Impact of total imports on industry employment (1) (2) (3) (4) OLS First stage 2SLS RF Chinese IPP in Ethiopia 0.120∗∗∗ 0.152∗∗∗ (0.034) (0.050) Chinese IPP in SSA countries 0.302∗∗∗ 0.046∗∗∗ (0.025) (0.017) Year Fixed Effects Yes Yes Yes Yes Industry Fixed effects Yes Yes Yes Yes Controls Yes Yes Yes Yes R2 0.94 0.82 0.94 0.94 Observations 278 278 278 278 Kleibergen-Paap F -statistic 150.85 Dep. var Mean 7,783 SD 0.23 Notes : Table displays results on Log(employment). The unit of observation is the 2-digit industry by year. The sample includes years from 2002 to 2017. In the 2SLS results, the change in Ethiopian import exposure is instrumented with Chinese imports in other Sub-Sahara African economies. RF in column(4) stands for reduced form. Standard errors in parentheses are clustered at 2-digit industries in all specifications. All regressions include years, industry fixed effects and the industry controls. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 51 Table 4: Industry composition of Chinese imports by end use category (2002-2017) Industry Imports average Average Average intermediate consumption imports share imports share Food & beverages 2.92 0.55 0.45 Tobacco products 1.41 0.00 1.00 Textiles 56.02 0.73 0.27 Wearing apparel 80.72 0.00 1.00 Leather & footwear 13.15 0.03 0.97 Wood 7.02 0.98 0.02 Paper 4.27 0.90 0.10 Printing 6.28 0.31 0.69 Chemicals 45.82 0.76 0.24 Rubber & plastics 32.70 0.91 0.09 Non-metallic Minerals 15.21 0.84 0.16 Basic metals 131.14 1.00 0.00 Fabricated metals 71.35 0.97 0.03 Machinery 112.23 0.94 0.04 Computing machinery 8.95 0.99 0.01 Electrical machinery 110.93 0.84 0.16 Communications equipment 98.06 0.93 0.07 Medical equipment 13.18 0.94 0.06 Vehicles 43.93 0.95 0.00 Transport equipment 18.66 0.89 0.11 Furniture and others 9.61 0.23 0.77 Notes : Employment average is the average industry employment from 2002 to 2017. Employment share is the average value of industry employment in total manufacturing employment in each year. Imports average is the average industry imports from 2002 to 2017. Imports share is the average value of industry Chinese imports in total manufacturing imports in each year. Imports are expressed in USD (2017 millions), employment values are in thousands. Table constructed using data from BACI and LMSM 52 Table 5: Impact of final and intermediate imports on employment (1) (2) (3) (4) OLS First stage-interm First-stage-cons 2SLS Chinese intermediate IPP in Ethiopia 0.184∗∗∗ 0.197∗∗ (0.052) (0.082) Chinese consumption IPP in Ethiopia 0.016 -0.029 (0.013) (0.051) Chinese consumption IPP in SSA countries -0.047∗∗ 0.978∗∗∗ (0.019) (0.067) Chinese intermediate IPP in SSA countries 0.355∗∗∗ -0.217∗ (0.034) (0.118) Year Fixed Effects Yes Yes Yes Yes Industry Fixed effects Yes Yes Yes Yes Controls Yes Yes Yes Yes R2 0.94 0.82 0.89 0.94 Observations 278 278 278 278 Kleibergen-Paap F -statistic 109.80 212.85 Dep. var Mean 7,783 SD-Intermediate 0.20 SD-Consumption -0.12 Notes :Table displays results on Log(employment). The unit of observation is the 2-digits industry by year. The sample includes years from 2002 to 2017. Standard errors in parentheses are clustered at 2-digit industries in all specifications. All regressions include years, industry fixed effects and the industry controls. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. Table 6: Impact of final and intermediate imports usage on employment (1) (2) (3) (4) OLS First stage-interm First-stage-cons 2SLS Chinese interm. IPP usage in Ethiopia 0.099∗∗ 0.141∗∗ (0.048) (0.064) Chinese consumption IPP in Ethiopia 0.007 -0.079 (0.013) (0.049) Chinese consumption IPP in SSA countries -0.005 1.005∗∗∗ (0.017) (0.066) Chinese interm. IPP usage in SSA 0.615∗∗∗ 0.109 (0.035) (0.138) Year Fixed Effects Yes Yes Yes Yes Industry Fixed effects Yes Yes Yes Yes Controls Yes Yes Yes Yes R2 0.94 0.95 0.89 0.93 Observations 278 278 278 278 Kleibergen-Paap F -statistic 310.82 231.39 Dep. var Mean 7,783 SD-Intermediate 0.34 SD-Consumption -0.31 Notes :Table displays results on Log(employment). The unit of observation is the 2-digits industry by year. The sample includes years from 2002 to 2017. Standard errors in parentheses are clustered at 2-digit industries in all specifications. All regressions include years, industry fixed effects and the industry controls. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 53 Table 7: Impact of intermediate imports usage on employment for firms using domestic vs imported inputs (1) (2) (3) Imported Domestic input input Interaction Chinese interm. IPP usage in Ethiopia 0.143∗∗ 0.091 -0.036 (0.058) (0.234) (0.094) Chinese interm. IPP usage in SSA Share imported inputs * Chinese interm. IPP in SSA 0.345∗∗∗ (0.111) Share imported inputs 0.436∗∗ (0.186) Year Fixed Effects Yes Yes Yes Industry Fixed effects Yes Yes Yes Controls Yes Yes Yes R2 0.93 0.82 0.95 Observations 275 181 278 Joint p-value 0.00 Notes :Table displays the 2SLS results on Log(employment). The unit of observation is the 2-digits industry by year. The sample includes years from 2002 to 2017. column(1) reports the reduced form results on the sample restricted to the firms using imported inputs. column(2) reports the reduced form results on the sample restricted to firms that do not use imported inputs. column(3) reports the reduced form results from the entire sample of firms aggregated at the industry level. The joint p-value corresponds to the p-value of the test that the coefficients on the interaction and Chinese IPP usage are jointly zero. Standard errors in parentheses are clustered at 2-digit industries in all specifications. All regressions include years, industry fixed effects and the industry controls. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 54 Table 8: Impact of intermediate imports usage on shortage of inputs (1) (2) (3) (4) OLS First stage 2SLS RF Chinese interm. IPP usage in Ethiopia -9.473∗∗ -13.845∗∗∗ (3.789) (4.293) Chinese interm. IPP usage in SSA 0.667∗∗∗ -9.237∗∗∗ (0.030) (3.065) Year Fixed Effects Yes Yes Yes Yes Industry Fixed effects Yes Yes Yes Yes Controls Yes Yes Yes Yes R2 0.91 0.94 0.91 0.91 Observations 275 275 275 275 Kleibergen-Paap F -statistic 484.47 Dep. var Mean 7,350 Notes : Table displays results on the share of firms within an industry reporting shortage of raw materials as main reason preventing them from operating at full scale. The unit of is the 2-digits industry by year. The sample includes years from 2002 to 2017. In the 2SLS results, the change in Ethiopian import exposure is instrumented with Chinese imports in other Sub-Sahara African economies. Standard errors in parentheses are clustered at 2-digit industries in all specifications. All regressions include years, industry fixed effects and the industry controls. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. Table 9: Correlation between Chinese imports and imports from the ROW (1) (2) (3) Share imports from Europe -0.132 -0.315∗∗∗ -0.340∗∗∗ (0.091) (0.095) (0.101) Share imports from Asia (excl. China) -0.447∗∗∗ -0.492∗∗∗ (0.074) (0.080) Share imports from America -0.336∗∗ (0.123) Year Fixed Effects Yes Yes Yes Industry Fixed effects Yes Yes Yes R2 0.30 0.53 0.55 Observations 599 598 572 Dep. var Mean 0.24 Notes : Table displays the OLS results on industry imports from China. The unit of observation is the 2-digits industry by year. The sample includes years from 2002 to 2017. Standard errors in parentheses are clustered at 2-digit industries in all specifications. All regressions include years, industry fixed effects and the industry controls. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 55 Table 10: Correlation between Chinese imports and imports from the ROW - by end use category (1) (2) (3) Total Intermediates Consumption Share imports from Europe -0.340∗∗∗ -0.645∗∗∗ -0.455∗∗ (0.101) (0.048) (0.164) Share imports from Asia (excl. China) -0.492∗∗∗ -0.697∗∗∗ -0.576∗∗∗ (0.080) (0.047) (0.091) Share imports from America -0.336∗∗ -0.611∗∗∗ -0.348∗∗∗ (0.123) (0.067) (0.106) Year Fixed Effects Yes Yes Yes Industry Fixed effects Yes Yes Yes R2 0.55 0.69 0.57 Observations 572 1,765 410 Dep. var Mean 0.24 0.26 0.33 Notes : Table displays the OLS results on industry imports from China, separately for intermediate and final imports. The unit of observation is the 2-digits industry by year. The sample includes years from 2002 to 2017. Standard errors in parentheses are clustered at 2-digit industries in all specifications. All regressions include years and industry fixed effects and the industry contros. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. Table 11: Correlation between the price of Chinese imports and the price of imports from the ROW - by end use category (1) (2) (3) Total Intermediates Consumption china -104.791∗∗∗ -139.014∗∗∗ -71.513∗∗∗ (23.463) (36.762) (24.085) Year Fixed Effects Yes Yes Yes Industry Fixed effects Yes Yes Yes Controls Yes Yes Yes R2 0.03 0.05 0.09 Observations 6,249 3,327 2,529 Notes : Table displays the OLS results on the price differential between imports from China and those from the ROW, separately for intermediate and final imports. The unit of observation is the 2-digits industry by year and region. The sample includes years from 2002 to 2017. Standard errors in parentheses are clustered at 2-digit industries in all specifications. All regressions include years, industry and regional fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 56 Table 12: Mechanisms of the Impact of Chinese imports - input usage (1) (2) (3) (4) (5) Capacity Skills TFP utilization intensity Entry Exit Chinese interm. IPP usage in Ethiopia 0.231∗∗∗ 0.191∗ 0.038∗∗∗ -0.030 -0.044 (0.040) (0.109) (0.009) (0.024) (0.037) Year Fixed Effects Yes Yes Yes Yes Yes Industry Fixed effects Yes Yes Yes Yes No Controls Yes Yes Yes Yes Yes Observations 272 262 275 259 257 Dep. var Mean 1.94 0.73 0.31 0.30 0.26 Notes : Table displays the 2SLS results on industry outcomes presented in each column. The unit of obser- vation is the 2-digits industry by year. The sample includes years from 2002 to 2017. Standard errors in parentheses are clustered at 2-digit industries in all specifications. All regressions include years, industry fixed effects and the industry controls. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. Table 13: Heterogeneous impact of Chinese imports by industry characteristics (1) (2) (3) (4) Labor Large intensive Private Public Chinese interm. IPP usage in Ethiopia 0.137∗ 0.232∗∗∗ -0.113 0.059 (0.072) (0.073) (0.169) (0.144) Year Fixed Effects Yes Yes Yes Yes Industry Fixed effects Yes Yes Yes Yes Controls Yes Yes Yes Yes R2 0.90 0.90 0.91 0.71 Observations 258 252 261 215 Dep. var Mean 6,334 3,722 5,263 2,431 Joint p-value 0.00 0.00 Notes : Table displays the 2SLS results on industry Log(employment) across different sub-samples. The unit of observation is the 2-digits industry by year. The sample includes years from 2002 to 2017. column(1) is restricted on industry large firms (firms employing at least 20 workers). column(2) runs on the sample of firms above the 50th percentile of labor intensity, aggregated across industries. column(3) restricts on private firms, aggregated across industries. column(4) restricts on public firms, aggregated across industries. Standard errors in parentheses are clustered at 2-digit industries in all specifications. All regressions include years, industry fixed effects and the industry controls. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 57 A Appendix A.1 Figures Figure A.1: Composition of Chinese imports in Ethiopia (2016) Source: The Observatory of Economic Complexity (OEC) oec.world/en/profile/bilateral-country/chn/partner/eth 58 Permanent employment 18 40 61 83 104 20 2 3 4 5 37 44 51 58 64 56 837 618 399 180 19 92 30 68 06 44 19 2 3 4 5 6 96 19 87 65 43 21 99 96 96 20 20 20 02 02 02 20 20 20 08 08 08 20 20 20 14 Printing 14 14 0 .5 1 1.5 2 2.5 Fabricated metals Food \& beverages 0 1 2 3 4 5 0 .005 .01 .015 .02 34 68 101 135 24 52 79 107 135 10 1 1 2 2 19 6 3 0 7 4 19 06 00 94 88 82 19 53 519 984 450 915 9 3 7 1 5 96 96 96 20 20 20 02 02 02 20 20 20 08 08 08 20 20 20 14 14 14 Textiles Machinery Chemicals 0 .5 1 1.5 2 2.5 0 .2 .4 .6 .8 1 0 .5 1 1.5 49 98 147 196 18 61 104 146 189 25 63 101 140 178 19 0 1 2 3 4 19 52 32 12 92 72 14 47 80 13 46 96 19 96 96 20 20 20 02 59 02 02 20 20 20 08 08 08 20 20 20 14 14 14 Wearing apparel Rubber \& plastics 0 2 4 6 8 0 .2 .4 .6 .8 1 0 2 4 6 8 Electrical machinery Permanent employment 44 88 131 175 219 62 90 119 147 176 4 20 35 51 67 19 67 26 85 44 03 20 67 14 61 08 19 19 48 22 96 70 44 96 96 96 20 20 20 02 02 02 20 20 20 08 08 08 20 20 20 14 Vehicles 14 14 0 .5 1 1.5 2 Import penetration (in thousands USD) 0 .1 .2 .3 .4 .5 0 .2 .4 .6 Leather \& footwear Non-metallic Minerals 22 3 5 7 9 19 32 951 670 389 108 87 279 471 663 855 11 18 25 32 39 96 19 5 5 5 5 5 19 41 36 31 26 21 96 96 20 20 02 20 02 02 20 20 20 08 08 08 20 20 Paper 20 14 14 14 Basic metals Figure A.2: Industry total manufacturing employment and Chinese import penetration 0 1 2 3 0 .1 .2 .3 .4 0 .5 1 Furniture and others Figure A.3: Intermediate imports share from China and the Rest of the World Notes: Figure graphs the share of intermediate imports by origin. Dash dots line shows the share of intermediate good imports out of total imports from Asia (excluding China). Dash line shows the share of intermediate good imports out of total imports from Europe. Solid line shows the share of intermediate good imports out of total imports from China. 60 Figure A.4: Intermediate Chinese imports share to Ethiopia and USA Notes: Figure graphs the share of intermediate imports by destination. Dash line shows the share of intermediate good imports out of total imports from China to USA. Solid line shows the share of intermediate good imports out of total imports from China to Ethiopia. 61 Figure A.5: Average effective tariff in Ethiopia over time 40 Average effective tariff rate 25 3020 35 1992 1996 2000 2004 2008 2012 2016 Year Source: Bigsten et al. (2016) using data from the Ethiopia’s Ministry of Finance and Economic Development 62 A.2 Tables Table B.1: Industry control variables (1) (2) (3) (4) Chinese IPP in Ethiopia 0.164∗∗∗ 0.165∗∗∗ 0.153∗∗∗ 0.152∗∗∗ (0.048) (0.048) (0.051) (0.050) Log(Imports from America) -0.019 -0.014 -0.029 (0.042) (0.042) (0.042) Log(Imports from Asia-C) -0.065 -0.083 (0.063) (0.062) Log(Imports from Europe) 0.182∗∗∗ (0.070) Number of Obsevations 279 278 278 278 Year Fixed effects Yes Yes Yes Yes Industry dummies Yes Yes Yes Yes Note: Standard errors clustered at the 2-digits industry level. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. Table B.2: Industry control variables (1) (2) (3) (4) (5) (6) Chinese IPP in Ethiopia 0.164∗∗∗ 0.165∗∗∗ 0.153∗∗∗ 0.152∗∗∗ 0.160∗∗∗ 0.160∗∗∗ (0.048) (0.048) (0.051) (0.050) (0.056) (0.056) Log(Imports from America) -0.019 -0.014 -0.029 -0.037 -0.037 (0.042) (0.042) (0.042) (0.042) (0.042) Log(Imports from Asia-C) -0.065 -0.083 -0.090 -0.090 (0.063) (0.062) (0.063) (0.063) Log(Imports from Europe) 0.182∗∗∗ 0.194∗∗∗ 0.194∗∗∗ (0.070) (0.072) (0.072) Log(Initial Employment) 1.165∗∗∗ 1.012∗∗∗ (0.096) (0.075) Log(Initial capital intensity) 0.190∗∗ (0.090) Number of Obsevations 279 278 278 278 267 267 Year Fixed effects Yes Yes Yes Yes Yes Yes Industry dummies Yes Yes Yes Yes Yes Yes Note: Standard errors clustered at the 2-digits industry level. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 63 Table B.3: Correlation between initial employment and future Chinese imports (1) (2) (3) ∆IP P ∆IP P 2002-2017 2002-2017 Log(Employment) Initial employment -0.000∗∗∗ (0.000) 1996-2002 Employment growth -0.004∗∗∗ 0.023∗∗∗ (0.000) (0.001) IPP 0.165∗∗∗ (0.058) Initial employment*trend -0.008∗ (0.004) Log(Imports from America) -0.034 (0.060) Log(Imports from Asia-C) -0.080 (0.050) Log(Imports from Europe) 0.161∗∗∗ (0.042) Number of Obsevations 267 462 267 Year Fixed effects Yes Yes Yes Industry fixed effects Yes Yes Yes Note: Clustered standard errors in parentheses (at the industry level). All regressions include years and industry fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 64 Table B.4: Overidentification test of instruments (1) (2) (3) First stage 2SLS 2SLS Overident IPP(IV) 0.302∗∗∗ (0.025) IPP 0.152∗∗∗ 0.143∗∗∗ (0.050) (0.043) Year Fixed Effects Yes Yes Yes Industry Fixed effects Yes Yes Yes Controls Yes Yes Yes R2 0.82 0.94 0.94 Observations 278 278 226 Kleibergen-Paap F -statistic 150.85 Overidentification tests p-values J-Statistic 0.69 Sargan test 0.10 Basmann test 0.18 Note: Clustered standard errors in parentheses (at the industry level). All regressions include years and industry fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. Table B.5: Impact of total Chinese imports using AADHP’s instrument (1) (2) (3) (4) OLS First stage 2SLS Reduced form ∗∗ IPP 0.120 0.197 (0.047) (0.136) IPP (IV) 3.461 0.680 (3.524) (0.881) Number of Obsevations 278 278 278 278 Year FE Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Controls Yes Yes Yes Yes F-statistic 0.96 Mean Dep.var 8.09 Note: Clustered standard errors in parentheses (at the industry level). All regressions include years and industry fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 65 Table B.6: Impact of consumption good imports using AADHP’s instrument (1) (2) (3) (4) OLS First stage 2SLS Reduced form IPP 0.056 -0.083 (0.114) (0.244) IPP (IV) 2.890∗∗∗ -0.240 (0.582) (0.739) Number of Obsevations 278 278 278 278 Year FE Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Controls Yes Yes Yes Yes F-statistic 24.67 Mean Dep.var 8.09 Note: Clustered standard errors in parentheses (at the industry level). All regressions include years and industry fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. Table B.7: Impact of intermediate good imports usage using AADHP’s instrument (1) (2) (3) (4) (5) (6) (7) OLS First stage 2SLS RF Imported input Domestic input Interac IPP interm - down 0.099 0.276∗∗∗ (0.113) (0.093) IPP interm - down (IV) 24.611∗∗∗ 6.792∗∗ 6.021∗ 7.671 3.986 (5.659) (2.550) (3.353) (6.279) (4.683) share imported 0.330 (0.595) interac interm 7.579∗ (4.009) Number of Obsevations 278 278 278 278 272 231 278 Year FE Yes Yes Yes Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes Yes F-statistic 18.92 Mean Dep.var 8.09 Note: Clustered standard errors in parentheses (at the industry level). All regressions include years and industry fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 66 Table B.8: Impact of total Chinese imports using New Zealand, Spain, and Iceland (1) (2) (3) (4) OLS First stage 2SLS Reduced form ∗∗ IPP 0.120 0.117∗∗ (0.047) (0.058) ∗∗∗ IPP (IV) 43.085 5.022 (12.903) (3.624) Number of Obsevations 278 278 278 278 Year FE Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Controls Yes Yes Yes Yes F-statistic 11.15 Mean Dep.var 8.09 Note: Clustered standard errors in parentheses (at the industry level). All regressions include years and industry fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. Table B.9: Impact of consumption goods imports using New Zealand, Spain, and Iceland (1) (2) (3) (4) OLS First stage 2SLS Reduced form IPP 0.056 0.033 (0.114) (0.148) ∗∗∗ IPP (IV) 24.385 0.801 (1.676) (4.020) Number of Obsevations 278 278 278 278 Year FE Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Controls Yes Yes Yes Yes F-statistic 211.78 Mean Dep.var 8.09 Note: Clustered standard errors in parentheses (at the industry level). All regressions include years and industry fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 67 Table B.10: Impact of intermediate goods imports usage using New Zealand, Spain, and Iceland (1) (2) (3) (4) (5) (6) (7) OLS First stage 2SLS RF Imported input Domestic input Interac IPP interm - down 0.099 0.276∗∗∗ (0.113) (0.093) IPP interm - down (IV) 24.611∗∗∗ 6.792∗∗ 6.021∗ 7.671 3.986 (5.659) (2.550) (3.353) (6.279) (4.683) share imported 0.330 (0.595) interac interm 7.579∗ (4.009) Number of Obsevations 278 278 278 278 272 231 278 Year FE Yes Yes Yes Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes Yes F-statistic 18.92 Mean Dep.var 8.09 Note: Clustered standard errors in parentheses (at the industry level). All regressions include years and industry fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. A.3 Further robustness tests results Table C.11: Exclude one industry each time (1) (2) (3) (4) (5) (6) (7) (8) Textile Leather Machinery Chemicals Coke Computing Wood Food Chinese IPP in Ethiopia 0.145∗∗∗ 0.149∗∗∗ 0.136∗∗∗ 0.154∗∗∗ 0.152∗∗∗ 0.152∗∗∗ 0.156∗∗∗ 0.150∗∗∗ (0.050) (0.052) (0.039) (0.053) (0.050) (0.050) (0.052) (0.053) Log(Imports from America) -0.022 -0.034 0.005 -0.030 -0.029 -0.029 -0.035 -0.031 (0.043) (0.045) (0.033) (0.044) (0.042) (0.042) (0.044) (0.044) Log(Imports from Asia-C) -0.103 -0.090 -0.065 -0.088 -0.083 -0.083 -0.072 -0.088 (0.063) (0.066) (0.048) (0.065) (0.062) (0.062) (0.069) (0.066) Log(Imports from Europe) 0.151∗∗ 0.197∗∗∗ 0.168∗∗∗ 0.180∗∗ 0.182∗∗∗ 0.182∗∗∗ 0.187∗∗ 0.190∗∗ (0.076) (0.073) (0.055) (0.074) (0.070) (0.070) (0.077) (0.074) Number of Obsevations 262 262 262 262 278 277 262 262 Year FE Yes Yes Yes Yes Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Yes Yes Yes Yes Controls Yes Yes Yes Yes Yes Yes Yes Yes Mean Dep.var 8.09 8.09 8.09 8.09 8.09 8.09 8.09 8.09 Note: Standard errors clustered at the 2-digits industry level. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 68 Table C.12: Results using different data samples (1) (2) (3) (4) (5) (6) Post 2002 All years ISIC excl Restrict1 Restrict2 Restrict3 Chinese IPP in Ethiopia 0.152∗∗∗ 0.088∗∗ 0.165∗∗∗ 0.138∗∗∗ 0.137∗∗∗ 0.147∗∗∗ (0.039) (0.040) (0.055) (0.034) (0.048) (0.044) Log(Imports from America) -0.029 0.021 -0.035 -0.005 -0.053 -0.058 (0.057) (0.044) (0.073) (0.046) (0.052) (0.067) Log(Imports from Asia-C) -0.083∗ -0.068 -0.040 -0.079 -0.080∗ -0.070 (0.050) (0.064) (0.076) (0.056) (0.048) (0.058) Log(Imports from Europe) 0.182∗∗∗ 0.173∗∗ 0.271∗∗∗ 0.173∗∗∗ 0.158∗∗∗ 0.197∗∗∗ (0.053) (0.071) (0.063) (0.050) (0.052) (0.066) Number of Obsevations 278 356 237 277 272 278 Year FE Yes Yes Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Yes Yes Note: Column(1) excludes observations before 2002. Column(2) includes all years (1996-2017). Column(3) excludes the following industries due to lack or negligible imports and/or employment data: Tobacco prod- ucts, Wood, Petroleum and nuclear fuel, Computing machinery, Communication equipment, and Medical equipment. Column(4) excludes firms where employment varies inconsistently (by more than 5*average over time). Column(5) Excludes firms that only show up once in the data. Employment and Imports are win- sorized at 1%. Clustered standard errors in parentheses (at the industry level). All regressions include years and industry fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. Table C.13: Alternative definitions of imports exposure (1) (2) (3) (4) IPP absorption 0.152∗∗∗ (0.050) IPP employment 0.021∗∗∗ (0.006) IPP sales 0.045∗∗∗ (0.013) Total imports (10,000 USD) 0.016∗∗ (0.007) Number of Obsevations 278 267 267 278 Year FE Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Controls Yes Yes Yes Yes Note: Clustered standard errors in parentheses (at the industry level). All regressions include years and industry fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 69 Table C.14: Using lagged values of imports (1) (2) (3) (4) IPP absorption 0.094∗∗∗ (0.033) IPP employment 2.060∗∗∗ (0.579) IPP sales 4.534∗∗∗ (1.334) Total imports (10,000 USD) 0.014∗∗ (0.007) Number of Obsevations 278 267 267 278 Year FE Yes Yes Yes Yes Industry FE Yes Yes Yes Yes Controls Yes Yes Yes Yes Note: Clustered standard errors in parentheses (at the industry level). All regressions include years and industry fixed effects. ***, **, and * indicate significance at the 1, 5, and 10 percent critical level. 70