Policy Research Working Paper 10203 Central America’s Deindustrialization Rishabh Sinha Development Economics Development Research Group October 2022 Policy Research Working Paper 10203 Abstract The paper assembles and harmonizes sectoral data from barriers can potentially deliver considerable industrial several sources to study the industrial trends in six Central expansion. But the economic impact of this policy is likely American economies. The industrial employment share to be marginal, with aggregate output increasing by 3 per- contracted by 2.5 percentage points on average over the cent or less if barriers are eliminated. At the same time, this past two decades. This deindustrialization was not trade- approach also carries several risks, and rather than reining driven in which economies substitute domestic production in inefficiency might introduce new distortions making the of industrial goods via cheaper imports. Instead, an increase economy more inefficient. Perhaps a more prudent growth in barriers restricting the efficient flow of labor across sec- strategy will be to concentrate on boosting productivity, tors drives this decline. Adopting policies that target such which, although challenging, has a direct effect on output. 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 rishabhsinha@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 Central America’s Deindustrialization Rishabh Sinha The World Bank JEL classification: F11; O11; O14 Keywords: Deindustrialization; Structural transformation; Central America; Growth; Trade; Labor market wedges 1 Introduction The advent of the Industrial Revolution was a defining moment in the economic history of the world. It ignited a period of Great Divergence in which the industrializing nations broke away from the rest (Clark, 2014). For the most part, this divergence that started centuries ago still defines the fault lines between the rich and the poor economies. It is not so surprising then that industrial expansion is a key component of many growth policies. 1 At the same time, deindustrialization remains a robust feature of economic development (Clark, 1957; Kuznets, 1957; Herrendorf et al., 2014), especially when considering the sector's employment share instead of its output contribution (Rodrik, 2016). Moreover, there is emerging evidence that the onset of deindustrialization has been premature in recent decades (Rodrik, 2016; Felipe et al., 2019). The peak industrial share in the developing economies is lower relative to the early industrializers and materializes at a lower income per capita. A growing literature is trying to uncover the underlying factors that drive deindustrialization. More crucially, it seeks to quantify the growth implications of deindustrialization and identify policies that can arrest the industry's slide. However, the country-level studies only cover a handful of economies. This analysis requires quality long-term sectoral data, which are not available for many countries. Small and low-income economies are particularly under-represented. Using learnings from other cases to devise policies is of limited use if factors driving deindustrialization vary across countries. This paper addresses this gap by conducting a country-level analysis of six Central American economies in the last two decades. I piece together data from several sources and harmonize them at the sector level. The raw data require cleaning, and I make several adjustments to ensure quality before using them in analysis. Though value-added shares are indeed a part of the examination, the study concentrates on the dynamics of sectoral employment shares. Two facts guide this choice. First, as already mentioned, deindustrializing trends are more pronounced in employment (Rodrik, 2016). Second, unlike its employment counterpart, the industrial value-added share is only a weak predictor of attaining economic prosperity (Felipe et al., 2019). 1 Examples include Make in India and Made in China 2025. However, these large-scale efforts transcend the boundaries of the developing world. The American Jobs Plan seeks to reinvigorate industrial production. The plan deems the sector the Arsenal of American Prosperity that can ‘fuel an economic recovery for working families (White House, 2021)’. 2 The data shows that deindustrialization has permeated Central America. Except for Panama, the industrial employment shares have contracted in each regional economy, with an average decline of 2.5 percentage points (pp). In the first step of my study, I uncover the fundamental drivers of industrial change in each country. Specifically, I use the sectoral data on employment, value-added, consumption, and trade to identify the importance of three channels in accounting for the dynamics. First, a decline in the industrial labor allocation can be demand-driven, i.e., due to a fall in the industrial share of final consumption. Second, a rise in industrial imports substituting domestic production may lead to trade-induced deindustrialization. Finally, an intensification in the barriers, or labor market wedges (Hsieh & Klenow, 2009), to the movement of resources into the industry can have the same qualitative effect on labor allocation. The accounting reveals that shifts in labor market wedges are essential in explaining the region's deindustrialization. The channel causes industrial employment to contract in each regional economy. Its impact on industrial share varies across countries, ranging from a more than 5pp decline in Panama and El Salvador to less than one pp decay in Costa Rica. These wedge-driven contractions are higher than the actual share loss in three of the five economies that have experienced deindustrialization. The channel plays a secondary role in Costa Rica, explaining only 15 percent of the country's massive 6.2 pp drop in industrial share. Global trade has flourished over the last two decades. Trade, therefore, can potentially explain deindustrialization in countries having a comparative advantage in non-industrial sectors. Instead of engaging in production to satisfy their industrial demand, it might be economical for these economies to import these goods. Surprisingly, I do not find evidence supporting this oft-cited trade-led deindustrialization hypothesis for Central America. Instead of creating headwinds, trade has usually fostered industrial expansion in the region. Nicaragua and El Salvador are the two economies where the channel has induced deindustrialization. Still, its actual impact is benign in both cases. Finally, changes in final consumption had a varied effect across the region. This channel aided industrialization in four economies, but an allocation of final consumption away from industrial goods led to a labor reallocation away from the sector in Costa Rica and Honduras. Quantitatively, this is the dominant factor that accounts for the deindustrialization in the two countries. The second part of the paper aims to quantify the aggregate impact of rising wedges. Their presence restricts the efficient flow of resources across sectors and may keep labor trapped in non-industrial 3 sectors. As industrial productivity usually lies considerably above others, wedge-driven deindustrialization might entail significant efficiency costs. I use a multi-sector equilibrium model to quantify these costs. Consistent with the structural transformation literature, the model features non-homothetic preferences (Kongsamut et al., 2001). The preferences also allow changes in relative prices to influence sectoral consumption allocation by utilizing a non-unitary elasticity of substitution across sectoral goods (Baumol, 1967; Ngai & Pissarides, 2007). Barring Costa Rica, labor market wedges in the industry have risen in all the economies, affecting deindustrialization. But wedges are not exclusive to the sector. Labor flow into services also faces barriers, and wedges in services are higher than in the industry in four of the six regional economies. Comparing across countries, I find the labor markets in Guatemala and Panama to be most distorted. I calibrate the model at the country level. This baseline matches the sectoral allocation of value-added and employment. Exogenously changing the wedges in this baseline alters the equilibrium allocation of labor resources, causing the output to change. Any such deviation is due to the variation in wedges. For each country, I measure the output change in the most recent year after I supplant the wedges from the earliest period. Output change is positive for every country except Costa Rica when current wedges in both sectors are replaced with their past values. Still, the implied loss is marginal and ranges from 0.4 percent of current output in El Salvador to around one percent in Panama. The above findings suggest that deindustrialization has not caused a significant drag on economic growth in Central America. Changes and labor market wedges are the principal driver of industrial contraction. Yet, the output impact they exert is marginal. But can a reindustrialization, led by a reduction in wedges, deliver meaningful efficiency gains in the future? I consider this scenario in another set of counterfactual exercises. The first exercise corresponds to a sector-neutral policy, i.e., it does not deem a special status to any sector and targets barriers present in both industry and services to drive growth. Eliminating wedges in industry and services causes the industrial employment share to expand by 3.5 pp on average, with modest gains in El Salvador and Honduras (1.3-1.7 pp). Extensions are much higher in Nicaragua (6.6 pp) and Panama (4.4 pp). More importantly, though, the efficiency gains reaped by the economies remain limited. The average output expansion stands a shade below 2 percent, with individual country figures ranging between 0.3-3.2 percent. 4 But industrial (and sectoral) policies, by their very nature, are sector-specific. They seek to address concerns plaguing the industry, with corresponding corrective measures remaining well within the boundaries of the sector. For instance, the Make in India program increased the maximum permissible FDI share in construction and mining to 100 percent while keeping the limits much lower in several other sectors (e.g., insurance, pension, etc.) (Government of India, 2020). Similarly, the American Jobs Plan called for a USD 50 billion investment in semiconductor manufacturing to reduce barriers to careers in high-innovation sectors (White House, 2021). Still, it will be naive to think that capital constraints in India are specific to only a subset of sectors. Or that the semiconductor industry is the only high- innovation sector in the US that experiences the barriers that the American Jobs Plan has under its lens. These sector-focused policies will likely lead to more labor flowing into the respective sectors. On the other hand, their impact on aggregate output is a separate issue altogether. Relative to a sector-neutral policy, eliminating industrial wedges while keeping services wedges intact generates a greater impetus for industrialization because resources face barriers reducing in a single sector. Correspondingly, a considerable expansion in industrial employment shares follows this selective elimination. The model predicts an average gain of 18 pp in industrial employment shares, multiples higher than the previous case. Each regional economy experiences a double-digit increase. This exercise produces a valuable and somewhat unexpected finding. This selective elimination of industrial wedges often leads an economy to a more distorted state. In other words, a sector-specific policy that targets industrial wedges can create massive reallocation of the workforce towards the sector. But it also runs the risks of increasing allocative inefficiency. The paper contributes to the deindustrialization literature by conducting a country-based analysis on six Central American economies. These countries are small, with populations ranging between 4–17 million people. Lack of quality data leaves a gap in knowing how deindustrialization manifests itself in such cases. Piecing together and analyzing data on the sectoral composition of employment, output, consumption, and trade from several sources, the investigation complements the previous attempts that focus on output alone (Caceres, 2017; Guisan & Aguayo, 2005, 2015). Crucially, it helps uncover the fundamental drivers of the region's deindustrialization. In the absence of country-specific studies, these economies can only follow the policy prescriptions from studies focusing on other countries, which may or may not apply to them. The inappropriateness of this approach becomes evident when one considers the trade-deindustrialization link. Recent studies have shown that trade can be a vital force driving deindustrialization (Pierce & Schott, 2016; 5 Rodrik, 2016). The paper adds to this literature by showing that the trade channel cannot explain Central America's deindustrialization. If anything, shifts in trading patterns have fostered industrial expansion in recent decades. Hence, protectionist policies to counter deindustrialization will most likely fail. Industrial activity might even diminish if the trading partners reciprocate these measures. The rest of the paper is organized as follows. In the next section, I outline the accounting methodology, followed by a discussion of the data used in the analysis. Then, I present the results of the accounting exercise, highlighting the relative importance of the different fundamental drivers of deindustrialization. The penultimate section introduces the resource allocation model, which I use to quantify the aggregate implications of labor market wedges. I conclude the paper by discussing some policy implications. 2 Accounting for deindustrialization I adopt an accounting approach that helps evaluate the relative importance of three channels– domestic consumption, trade, and labor market wedges. The framework follows from the open- economy methodology introduced in Uy et al. (2013) to study the sectoral composition of value-added, and later extended in Sinha (2019) to bring employment shares under the lens. The accounting framework includes an arbitrary number of countries. Let represent the labor endowment in country . The production follows a standard tiered structure in which three composite sectors (agriculture (), industry (), services ()) produce a sectoral good using tradeable varieties (Eaton & Kortum, 2002). The composite sectoral good produced by sector ( ∈ {, , }) in country is either used for consumption or utilized as an intermediate to produce a tradeable variety = + � (1) ∈{,,} where corresponds to the quantity of sector good used in the production of varieties of sector . The total receipt ℛ from sale of composite good is given by ℛ ≡ �� � (2) 6 where is the price of sector ’s good in country and �∑ = 1� is the bilateral trade share in sector . 2 Equations (1) and (2) can be used to express the nominal gross output of sector in country as = + ��1 − � � + �� − �� (3) ≠ where and represent the nominal value-added share of sector and its share in intermediate expenditure of sector , respectively. The term ∑≠ � − � corresponds to country ’s net exports in sector , which can be used to establish the following relationship between sectoral value-added ( ̃ �, and net exports � � ), consumption � � � in nominal terms. ̃ + � = � � � + ��1 − � � � � (4) The system of equations in (4) can be represented in matrix notation as follows 1 − �1 − � ⎡ −�1 − � −�1 − � ⎤ � ⎢ ⎥ �̃ + � � −�1 − � ⎥ � �̃ + � � = ⎢ � 1 − �1 − � � �� (5) ⎢ −�1 − � ⎥ � ⎢ −�1 − � −�1 − � ⎥ �̃ + � � ⎣ 1 − � 1 − � ⎦ ̃ and � , where the aggregate variables � are reduced to their respective small caps to represent them as ratios of total value added (GDP). An additional step is needed to link the value-added shares to employment shares. The wage rate that a unit of labor earns varies across sectors, which leads to differences in nominal value-added per worker. Under perfect competition, the workers are paid their marginal productivities. Hence, the wage rates represent sectoral marginal productivities, and the differences in wage rates translate into differences in marginal productivities across sectors. The employment shares can be recovered from the value-added shares using � � = ≡ (6) � � ∑ ∑ where ≡ is the nominal value-added per worker in sector . The parameter represents the labor market wedge (Hsieh & Klenow, 2009) which creates barriers to the movement of labor across 2 Country ’s expenditure on varieties from country as a fraction of its total expenditure on sector varieties. 7 sectors and restricts the equalization of labor productivity (nominal value-added per worker) across sectors. Introducing time-scripts, the labor allocation = [ , , ]′ can be written as a function of � , and productivity gaps � , trade shares final consumption shares � , = � � , � , � (7) where and denote the time-invariant matrices of labor shares and sectoral linkages . 3 Therefore, function (. ) which combines equations (5) and (6) helps solve for the resulting labor � and are counterfactually altered. � , allocation when one or more vectors among 3 Data The quantitative application of equation (7) requires time-series data on four variables– employment � and labor market wedges . Equation � , trade shares shares , final consumption shares � . Hence, in essence, the analysis hinges on the (6) allows the construction of using and � , and � , availability of , � . In addition, I also need the estimates of time-invariant labor share vector and intermediate share matrix . Equation (7) allows quantifying the impact of over-time changes in and on sectoral allocation. However, the unavailability of these data for Central American economies hinders this investigation. I source the required data from multiple sources. However, the raw data appear spurious for some country-years. Perhaps, for this reason, the literature has not undertaken a long-term sectoral analysis of the Central American economies, especially when it comes to labor allocation. I make several adjustments to the raw data before using them. The data appendix contains the details of these adjustments. Still, I note the nature of adjustments briefly as I provide a summary of each series. Employment shares ( ): The employment data are taken from the International Labor Organization’s (ILO) Key Indicators of the Labor Market database (ILO, 2018) that reports employment by sector. The database classifies total employment into six sectors – agriculture, manufacturing, construction, mining and utilities, trade and transportation (includes personal & 3 Sinha (2019) contains the detailed derivations of the accounting equations. 8 business services), and government and social services. 4 I aggregate the data into three broad sectors following the taxonomy of International Standard Industrial Classification of All Economic Activities, Revision 3 (ISIC Rev. 3.1, 2002). The ILO database often constructs the employment series using information from the labor force (LFS) and household surveys (HS) and reports two different series. 5 In many instances, these raw series rely on surveys that have grave limitations. For example, the surveys used to generate employment shares during 1986-1991 in El Salvador included only urban households. This loss of representation of rural households leads to massive jumps in the employment shares (Figure 1). After declining steadily from around 50 percent in 1975 to about 35 percent in the next ten years, the agricultural share from the raw series fell below 2 percent in 1986. It later recovers to 35 percent in 1992. I drop such observations from the analysis after identifying them. To maximize coverage, I merge the remaining data from the two raw series (LFS and HS) if the sectoral shares are similar across the two in overlapping years. If there are no overlaps, I merge the two series if the resulting trend after the pooling is consistent with the individual trends. Figure 1: Unadjusted agricultural and industrial employment share: El Salvador The figure shows the unadjusted employment shares of agriculture and industry as estimated using the raw data from ILO’s Key Indicators of the Labor Market database (ILO, 2018). 4 Often a small share of employment is not mapped to these six categories and is left unspecified. I drop this share from the analysis. 5 For Nicaragua, an employment series based on official estimates is available in addition to one estimated using LFS. 9 � ): The National Accounts Main Aggregates database of � , Consumption and trade shares ( the United Nations Statistics Division (UNSD) (UN, 2019) contains long-term data on final consumption, exports, and imports at the economy-wide level. The sectoral splits of any of the three variables are, however, not available. On the other hand, the UNSD database does report the sectoral composition of value-added. Hence, given the values of other variables in equation (5), it is possible � . to infer the sectoral composition of final consumption To execute this strategy, I first aggregate the value-added series to the three sector level using the earlier taxonomy. Like raw employment data, the value-added sectoral shares computed using the UNSD data also feature spurious jumps. Therefore, I refine the raw series before using it in the analysis (see data appendix for details). Next, I source data on the sectoral exports and imports shares from a different source, which I use to � . I exploit the Atlas of Economic Complexity (Center for estimate the net export shares International Development, 2018) database that reports a country’s exports and imports across 11 sectors. Agriculture and services trade data in the database is presented at the aggregate level and is consistent with ISIC Rev. 3.1. The database reports the industrial data disaggregated into nine sub- sectors, and I sum across them to construct sectoral exports and imports. � and Once I have the estimates of � (and the estimates of and ) from the above steps, I � . employ equation (5) to recover final consumption shares � and employment shares allows Labor market wedges ( ): Data on value-added shares expressing sector 's labor productivity relative to any arbitrary sector . As agriculture is almost always the sector with the lowest value-added per worker, I measure productivity in industry and services relative to agriculture and set = 0 ∀ , . Specifically, the wedge applicable to sector in period equals the ratio of its value-added per worker to agricultural labor productivity . Labor share of income ( ) and sectoral linkages ( ): I obtain the labor share of value-added at the sectoral level from the Global Trade Analysis Project (GTAP) database (Center for Global Trade Analysis, 2015). The GTAP data are available for three overlapping years over which I perform the quantitative analysis using the time-series on employment, consumption, and trade shares. I estimate the labor share of value-added for each of the three years, the average of which serves as the benchmark labor share for the entire period under analysis. 10 Unfortunately, the input-output data for the Central American economies are not usually available. Barring Costa Rica, for which the input-output tables are available in the OECD database (OECD, 2017), I couldn’t find such data for the other countries. To overcome this challenge, I use the average intermediate shares from eight regional countries (including Costa Rica). In the data appendix, I show that these shares of regional economies lie within close range. Therefore, their average provides a reasonable benchmark for the Central American countries. Table 1 reports the country-wise coverage of cleaned data obtained from various sources. Column (6) shows the effective period of analysis for each country, which corresponds to the overlapping years in columns (2)–(5). Table 1: Country-wise time coverage Value Trade Effective Series Employment added shares Sectoral trade period shares shares (aggregate) shares of coverage Atlas of Source ILO UNSD UNSD economic complexity (1) (2) (3) (4) (5) (6) Costa Rica 1980-2017 1976-2017 1976-2017 1995-2016 1995-2016 Guatemala 1981-2017 2001-2017 2001-2017 1995-2016 2001-2016 Honduras 1991-2017 1970-2017 1970-2017 1995-2016 1995-2016 Nicaragua 2003-2012 1987-2017 1987-2017 1995-2016 2003-2012 Panama 1970-2017 1970-2017 1970-2017 1995-2016 1999-2016* El Salvador 1978-2017 1990-2017 1990-2017 1995-2016 1995-2016 *1995-1998 is dropped for Panama as the imputed agricultural consumption shares using the linkages are negative. 4 Accounting for trends in industrial employment � , � in � , This section quantifies the relative importance of the three macro variables � accounting for the change in industrial shares. Analyzing inter-temporal changes is not only crucial for countries that observe quantitative shifts in the sectoral distribution of the workforce. Significant 11 but opposing forces generated by fundamental factors can operate in tandem, and the resulting stability of the distribution may hide the underlying shifts. I compare the industrial share in the final year (1 ) with that in the initial year (0 ) for each country. Column (3) in Table 2 reports this change �1 − �. There is clear evidence of deindustrialization in the region. Barring Panama, all economies experience a contraction, with industrial share shrinking by 2.5 pp on average. The fall is particularly severe in Costa Rica (6.5 pp). To estimate the effect of a channel (say ), I start with the balanced accounting relationship in the � , � , ��. To quantify how much labor marker wedge � , initial year � = � accounts for the aggregate change (Column (3), Table 2), I plug in its values from the final year . The levels of the other two factors remain fixed to their initial year values. The implied change in sectoral allocation , − captures the quantitative impact of changes in the labor market wedges ′ � , ,− ≡ �Δ,1 −0 , Δ,1 −0 , Δ,1 −0 � = � � , � , � − (8) An additional adjustment is needed before equation (8) can be applied to infer the relative � ( contribution of the other two channels. Replacing � ( � ) with � ) without any � ) may lead to an imbalance in the accounting relationship. Specifically, � ( adjustment to variation in net lending as a share of GDP over time prevents equality in equation (8) from holding. � ) by a unique factor, which exactly offsets any inter- � ( To balance the equation, I scale temporal change in net lending. Columns (4)–(7) in Table 2 report the implied changes in industrial shares. The first two and the last relate to the stand-alone impact of final consumption, trade, and wedges, respectively. Column (6) � are replaced with � and corresponds to the case where both � . 6 As the figures � and in columns (4)–(7) denote the stand-alone impact, they can be higher in magnitude or be of the opposite sign compared to the actual change in industrial shares (column (3)). For instance, a channel can have a contractionary effect even when an economy experiences a net expansion in industrial 6 The counterfactual exercise for Column (6) requires no adjustment for net lending. 12 share. Such a situation implies that the negative effect of that particular channel is more than offset by the extension caused by the others. 7 Let me first look at the regional averages reported in the last row. I find that changes in the patterns of final consumption and trade created grounds for an industrial expansion. Shifts in trade shares have supported an extension of the industrial employment allocation and imply an average increase of 2pp. However, the amplifying force generated by the two channels is more than offset by that affected by shifts in labor market wedges. The exercise shows that the overtime changes in wedges, on average, shaved off 4pp from the industrial employment share. Now, let's see how the inference drawn at the regional level compares with the experience of the constituent economies. Consistent with the regional finding, wedges generate contractionary pressure in each country. Still, its impact varies substantially. Implied shrinkage is comparatively much higher in Panama (8.7pp) and is also quantitatively above the regional average in El Salvador (5.2pp). On the other hand, the channel's impact is milder in Costa Rica and Honduras. Like for the region, the channel accounts for more than the total contraction in industrial share in three of the five economies that underwent deindustrialization and can explain 85 percent of the decline in Honduras. In contrast, though aiding the fall, it accounts for less than 15 percent of the 6.2pp decline in Costa Rica. Table 2: Actual and implied change in industrial shares Implied change (pp) Actual Final Labor Initial Final Final Country change Trade cons. & market year year cons. (pp) trade wedge (1) (2) (3) (4) (5) (6) (7) CRI 1995 2016 -6.52 -8.26 4.46 -6.17 -0.91 GTM 2001 2016 -3.18 0.48 1.79 0.41 -3.74 HND 1995 2016 -1.57 -2.69 1.17 -0.80 -1.34 NIC 2003 2012 -1.81 1.90 -0.51 2.79 -4.14 PAN 1999 2016 1.14 13.13 5.58 13.40 -8.73 SLV 1995 2016 -3.30 0.51 -0.18 2.05 -5.24 Average -2.54 0.85 2.05 1.95 -4.02 Figures in columns (4)-(7) indicate the change in industrial employment shares when the variables pertaining to a factor listed in the column is counterfactually changed to the level observed in the final year, keeping the value of all the other factors fixed at the initial year. All figures in percentage points (pp). 7 For example, changes in wedges caused considerable downward pressure on industrial shares in Panama. Yet, the tailwinds provided by other channels ensured that the country saw a modest net expansion. In other words, if there were no shifts in consumption and trade shares, Panama would have experienced significant deindustrialization. 13 The qualitative impact of shifts in the other two channels is not as consistent. Changes in trading patterns created headwinds in four economies while fostering deindustrialization in Nicaragua and El Salvador. Still, the contractionary pressure it generates is quantitatively benign, with an implied decline of 0.2-0.5pp. On the other hand, when assisting industrial expansion, its average impact stands at 3.2pp, ranging from 4.5pp in Costa Rica to 1.2pp in Honduras. Final consumption too has a varied qualitative impact across the regional countries. The channel aided industrialization in four economies, but an allocation of final consumption away from industrial goods led to a labor outflow from the sector in Costa Rica and Honduras. The change in consumption patterns implies a decline of 8.3 and 2.7pp for the two economies, respectively. Final consumption is the dominant factor in accounting for deindustrialization in these two countries. The channel's expansionary impact ranges from being relatively small in Guatemala and El Salvador (0.5pp) to being considerably massive in Panama (13.1pp). In summary, the above analysis yields the following three findings. First, shifts in labor market wedges are critical to understanding the deindustrialization process in Central America. The channel exerted a contractionary force in each regional economy in the sample. Second, the trade channel is not effective in explaining the decline in industrial employment share. Shifts in trading patterns have, in fact, aided expansion in four of the six countries. Even when providing headwinds, its impact is overshadowed by that of wedges. Third, final consumption bears a varied influence across countries. The channel is the principal factor in accounting for the deindustrialization in Costa Rica and Honduras. Still, its evolution over the years has fostered industrial share expansion in other regional economies. 5 Aggregate implications of labor market wedges The quantitative analysis in the previous section shows that changes in labor market wedges have been a critical force behind the region's deindustrialization. The presence of wedges indicates barriers to the efficient flow of labor, which restricts an equalization of nominal labor productivities across sectors. As these wedges are quantitatively considerable, it is often argued that the efficiency losses they induce are significant too. In this section, I take a deeper look into the evolution of these wedges and use a resource allocation model to quantify their impact on aggregate output. 14 The notations used in this section are consistent with those used previously. A stand-in firm in each employing labor inputs according to sector (in country during period ) produces output a decreasing returns-to-scale technology given by = (9) where ∈ (0,1] is the labor share of the value-added and is the sector-specific TFP. Consistent with the accounting framework, the sector-specific labor market wedges appear as a tax on per-unit labor employed by the stand-in firm (Hsieh & Klenow, 2009). The firm maximizes profit each period taking the wage rate as given. Therefore, the firm's problem is max − �1 + � ≥0 is the sectoral price. where 8 A representative household characterizes the consumption side. It possesses a unit of labor that it supplies inelastically to earn wages each period. The earned wages, together with firm profits, are spent to maximize utility. As the firm's, the household's utility maximization problem is also static. The household preferences are non-homothetic in sectoral goods and follow a CES preference structure (Comin et al., 2021) given below −1 � �Ω � � � =1 (10) where is the aggregate consumption or utility derived from consuming sectoral goods from each sector. The parameters Ω �Ω > 0, ∑ Ω = 1� reflect the relative preference for sectoral goods, which varies across countries. The parameter (> 0) controls the level of substitutability across sectoral goods. The sector-specific income elasticities capture the non-homothetic nature of household preferences. All else equal, an increase in income shifts consumption towards a sector with relatively higher value of . Therefore, the household preferences allow the operation of both price (Baumol, 1967; Ngai & Pissarides, 2007) and income effects (Kongsamut et al., 2001), which are standard in the structural transformation literature. 8 I add the super-script to differentiate the value-added price of a sector from the final output price , which I introduced in the accounting framework. 15 The aggregate output in this economy in any period is equal to the aggregate household consumption . Market clearing requires that the aggregate output in any period equals the aggregate household consumption . As the focus of the analysis is on resource allocation, I normalize the labor resource to unity ( = 1 ∀ , ). The two exogenous forces that affect aggregate output are sectoral TFP and labor market wedges . 5.1 Model calibration The preference function in equation (10) contains four elasticities together with country-specific sector-preferences weights Ω . In addition, the model entails three sets of production parameters– , , and . The first two vary over sectors, countries, and time, and like in the accounting exercise, the labor share of value-added is country-specific but time-invariant. Preference parameters (, , ) I borrow the estimates of elasticities from Comin et al. Comin et al. (2021), which uses panel data consisting of 39 countries. The substitution elasticity equals 0.50, and the sector-specific income- elasticities and are set to 0.11 and 1.21, respectively. The industrial sector serves as the reference sector for measuring income effects, and its income elasticity is normalized to unity ( = 1). 9 I estimate the country-specific sector weights by matching the sectoral allocation of value-added in the initial year for each country. The first-order conditions from the household optimization problem yield the following relationship 1− �0 Ω − = � 0 � �0 � (11) �0 Ω 0 �0 is sector ’s share of the nominal value-added in the initial period 0 . where Labor market wedges ( ) The calibration of labor market wedges follows the strategy employed in the accounting framework earlier, and the estimates of remain the same. The firm's optimization problem yields � � = �1 + � (12) 9Consistent with the value-added production functions employed in this paper, the elasticity estimates correspond to an estimation based on the value-added by sectors and not their final output. 16 Equating the wage rate across any two sectors gives 1 + = (13) 1 + There is one less independent equation in the system above than the number of distortions. Therefore, as in with the accounting exercises, I normalize wedges in the agricultural sector to zero ( = 0, ∀, ). The wedges experienced by non-agricultural sectors can then be written as � = −1 (14) � � is the nominal value-added per worker of sector . where Sector TFPs ( ) The sectoral TFPs evolve over time and each country follows an idiosyncratic process. Multiplying both sides of the production function in (9) by sectoral price for some base year yields = ≡ (15) � � � � denotes the real value-added of a sector. In the calibration, the year 2010 serves as the base where year for all regional economies. Labor share of value-added () The sectoral labor shares are taken from the GTAP database (Center for Global Trade Analysis, 2015) and are discussed in more detail previously. 5.2 Aggregate effects of changes in wedges over time Before quantifying the aggregate implications of wedges, I briefly show how they have evolved over the years. Have they increased, decreased, or been cyclical? Note that the estimates of take both positive and negative values, both of which imply inefficiency. Therefore, in principle, I need to transform the labor market wedges to make them monotonic with inefficiency. However, the 17 empirical estimates of in each Central American economy for the sample years never lie below zero, making such a modification unnecessary. 10 The duration over which I can estimate the labor market wedges is longer than the one considered earlier (column (6), Table 1). The reason is that the aggregate analysis abstracts from trade and thus no longer requires data on trade shares (columns (4) and (5), Table 1). As a result, the overlap in columns (2) and (3) of Table 1 provides the period for which I can conduct the aggregate analysis. Figure 2 shows the behavior of wedges in each regional economy over time. The circles in blue and the squares in red correspond to wedges in industry and services, respectively. Recall that the wedges in non-agricultural sectors are estimated relative to those in agriculture � = 0, ∀, �. The intertemporal evolution of labor market wedges offers several insights. First, as already noted, the estimates are always greater than zero, implying that labor faces obstacles in moving out of agriculture. But these barriers are not specific to the industrial sector which keeps it from expanding. The channel also constrains the extension of services. In four of the six Central American economies, wedges in the services sector are, in fact, higher than in the industry. Finally, there is no evidence that the economies are becoming more efficient over time. In four countries, wedges in both sectors in the most recent year are higher relative to the earliest year for which data are available. Industry in Costa Rica and services in Nicaragua offer the only bright spots, where wedges in the final year are below the initial year. Moreover, in many cases, evolution does not follow a secular trend. Panama appears the most striking specimen of this phenomenon. Beginning from the mid-1980s, wedges in both industry and services contracted sharply till the early 2000s. But they rose sharply thereafter to finish above the earlier peaks. These distinct reversals create pivot points that may help uncover the root causes of labor market inefficiencies, including those related to policy and regulations. 10 Chen et al. (2021) propose one such transformation. 18 Figure 2: Estimates of labor market wedges � � (a) Costa Rica (b) Guatemala (c) Honduras (d) Nicaragua (e) Panama (f) El Salvador The figures above show the labor market wedges estimated using equation (14) in the industry (blue circles) and services (red squares) for the six Central American economies. 19 Second, comparing countries, I find that wedges (in both sectors) are significantly higher in Guatemala and Panama. The labor markets in Costa Rica and Nicaragua are least distorted, even though the period covered for the latter is considerably restricted. Finally, there is no evidence that the economies are becoming efficient over time. In four countries, wedges in both sectors in the most recent year are higher relative to the earliest year for which data are available. Industry in Costa Rica and services in Nicaragua offers the only bright spot, where wedges in the final year are below the initial year. Moreover, in many cases, evolution does not follow a secular trend. Panama appears the most striking specimen of this phenomenon. Beginning from the mid-1980s, wedges in both industry and services contracted sharply till the early 2000s. But they rose sharply thereafter to finish above the earlier peaks. These distinct reversals create pivot points that may help uncover the root causes of labor market inefficiencies, including those related to policy and regulations. Now, let me return to the task of quantifying the aggregate impact of changes in labor wedges. The two exogenous elements– TFP � � and wedges � �, jointly determine the aggregate output of country in year . Let (. ) be the function that maps the two exogenous factors to . The quantitative exercise estimates the output variation Δ1 in the most recent year 1 when I replace the level of wedges from those in the earliest year 0 ��1 �, �0 �� Δ1 = −1 (16) ��1 �, �1 �� A negative Δ1 implies that aggregate output in the most recent year declines if the economy is subjected to past wedges, signifying an increase in allocative efficiency. Panel A in Table 3 reports the results of the exercise. Note that the most recent (column (1)) and the earliest year (column (2)) varies across countries due to differences in coverage. First, let's consider column (3), which shows the percent change in output when I replace wedges in both non-agricultural sectors. Except for Costa Rica, all other regional economies experience a positive Δ1 , i.e., a decline in allocative efficiency. This finding is not surprising given a general rise in wedges (Figure 2). Still, the aggregate cost of this rise in inefficiency is benign. The counterfactual output increase ranges from 0.44 percent in El Salvador to just above a percent in Panama. At the same time, the efficiency gains in Costa Rica are marginal too. Output declines by about a fifth of a percent when I use wedges from 1980. 20 Table 3: Impact of changes in wedges on aggregate output Change in output (%) Wedges changed in Initial Wedges Industry Industry Services Country wedges replaced & only only from from services (1) (2) (3) (4) (5) Panel A: Longest possible period at country level CRI 2017 1980 -0.21 -0.09 -0.07 GTM 2017 2001 0.71 0.12 0.48 HND 2017 1991 0.85 -0.02 0.72 NIC 2012 2003 0.27 0.44 -0.12 PAN 2017 1970 1.03 -0.02 0.25 SLV 2017 1990 0.44 -0.43 0.45 Average . . . Panel B: Longer uniform period (1991-2017) CRI 2017 1991 0.04 0.03 0.00 GTM 2017 1991 . . . HND 2017 1991 0.85 -0.02 0.72 NIC 2017 1991 . . . PAN 2017 1991 0.95 -0.35 0.25 SLV 2017 1991 0.28 -0.26 0.32 Average 0.53 -0.15 0.32 Panel C: Shorter uniform period (2003-2012) CRI 2012 2003 0.06 0.08 -0.05 GTM 2012 2003 0.64 0.15 0.43 HND 2012 2003 -0.48 -0.03 -0.46 NIC 2012 2003 0.27 0.44 -0.12 PAN 2012 2003 0.87 -1.97 0.45 SLV 2012 2003 -0.08 0.01 -0.09 Average 0.21 -0.22 0.03 Columns (3)-(5) report the output change Δ1 (in %) when wedges in the year 1 (column (1)) are replaced by those from year 0 (column (2)). The setup also allows me to quantify the efficiency implications of changes in the industrial and services wedges individually. Column (4) reports the output change when I alter the former, leaving the latter unchanged at the final year levels. Unlike the combined impact in column (3), the change in industrial wedges implies an output contraction in Honduras, Panama, and El Salvador. Hence, the shift in industrial wedges over time has contributed to output expansion. The individual component 21 is in line with the combined effect in Costa Rica, Guatemala, and Nicaragua. Industrial wedges account for around 15-45 percent of the aggregate change in column (3) for the former two regional economies. In contrast, the counterfactual rise in output in Nicaragua is 1.65 times larger. Column (5) sheds light on the economic consequences of changes in services wedges. Supplanting past wedges increases output in four of the six regional economies. These changes account for anywhere from 25 percent (in Panama) to almost the total effect (in El Salvador) reported in column (3). The intertemporal evolution of service wedges in Costa Rica and Nicaragua aid economic efficiency �Δ1 < 0� and accounts for a third of the combined effect in the former. The cross-country comparison in Panel A is somewhat inconsistent due to the variation in periods. The exercise considers changes in wedges over around half a century in Panama. In sharp contrast, it examines variations over less than a decade in Nicaragua. This enormous gap in duration can produce a corresponding immense gap in wedges across countries if the wedges evolve monotonically. Moreover, the factors that affect the wedges in any year might also have a regional or global component. Because of these two reasons, a cross-country comparison should utilize a uniform period. I consider two such uniform periods. In Panels B and C, I analyze the changes in labor market wedges during 1991-2017 and 2003-2012, respectively. Both panels have a drawback and an advantage. The former covers a longer horizon (>25 years) at the cost of ignoring Guatemala and Nicaragua, for which data are unavailable. Panel C presents the opposite scrutiny by including all countries but studying evolution over a much shorter period. Foremost, Panels B and C confirm that the aggregate consequences of changes in labor market wedges are marginal. The maximum output expansion, which signifies a rising inefficiency, is less than a percent (column (3) in Panel B, Panama). Similarly, the maximum contraction remains modest across the two panels and is around 2% of the total output (column (4) in Panel C, Panama). Next, like in the topmost panel, the overtime variation has usually led to a rise in inefficiency. Output change Delta is positive for all four and four of the six regional economies in Panels B and C, with aggregate costs being most pronounced in Panama. Finally, the wedges in industry and services often bear qualitatively differential effect, which also changes across the two panels. In Panel B, where I consider a longer horizon, the changes in services wedge always imply an expansion. On the other hand, the aggregate output increases in each country except Costa Rica when I supplant the past industrial wedges keeping those in services fixed at 2017 22 levels. These qualitative findings reverse when I focus on the shorter period in Panel C, which is subsumed by the one in the panel above it. 5.3 Aggregate implications of future industrialization via reduction in wedges The main finding from the above quantitative exercise is that the aggregate implications of changes in wedges are benign. Moreover, the accounting analysis in Section 4 showed that it is the dominant channel in accounting for Central America's deindustrialization. These two observations taken together suggest that deindustrialization has not generated a considerable drag on economic growth. But can a reindustrialization, induced via reductions in industrial wedges in the future, lead to meaningful efficiency gains? 11 To evaluate this scenario, I examine the aggregate consequences of eliminating the wedges in the most recent year. As the industrial wedges are above zero in each regional economy, lowering them will reduce barriers to the movement of labor into the industrial sector, thereby initiating industrialization. Specifically, for each country, I compute the implied output change Δ1 as ��1 �, � = 0�� Δ 1 = −1 (17) ��1 �, �1 �� and the corresponding change in industrial employment share as ℒ ��1 �, � = 0�� Δ 1 = −1 (18) ℒ ��1 �, �1 �� Like (. ), the function ℒ (. ) solves for the industrial employment share given the TFP and wedge sequence for a country . But will industrialization fostered through reductions in industrial wedges lead to meaningful efficiency gains in the future? To evaluate this scenario, I examine the aggregate consequences of eliminating the wedges in the most recent year. As the industrial wedges are above zero in each regional economy, lowering them will reduce barriers to the movement of labor into the industrial sector, thereby supporting industrialization. Table 4 reports the results from this counterfactual experiment. Figures in column (2) show the expansion in industrial employment share as a result of eliminating wedges in industry and services, and column (4) corresponds to the associated gain in aggregate output. The average regional increase 11 To some extent, the previous exercise shed light on this issue. Consider the case of Panama in Panel A of Table 4. The wedges in the country in 1970 were much smaller than in 2017, especially in the industrial sector (Figure 2). Yet, replacing them from the former year led to just about a percent of output expansion. Though not explicitly shown, the corresponding extension in the industrial employment share was 4.6 pp. Even this massive reallocation of labor resources towards industry brings only meager gains. 23 stands at 3.5 pp, ranging from 1.3 pp in El Salvador to 6.6 pp in Nicaragua. Still, the efficiency gains from this industrialization remain trivial. The average increase is below 2 percent, with the figures for individual countries lying between 0.3 to 3.2 percent. Table 4: Aggregate impact of wedge-led industrialization Change in industrial Change in output employment share (%) (pp) Wedges eliminated Wedges eliminated in in Initial Industry Industry Industry Industry Country wedges & & only only from services services (1) (2) (3) (4) (5) CRI 2017 3.24 10.88 0.30 -0.33 GTM 2017 3.84 19.25 3.16 -2.37 HND 2017 1.65 14.02 2.23 -1.64 NIC 2012 6.64 13.47 1.63 0.07 PAN 2017 4.42 32.34 2.48 -6.70 SLV 2017 1.33 18.31 1.65 -2.59 Average 3.52 18.04 1.91 -2.26 Columns (2) and (3) report the change in industrial employment share 1 (in pp) when wedges in the year 1 (column (1)) are eliminated. The corresponding change in output Δ 1 (in %) are reported in columns (4) and (5), respectively. As services wedges also lie above zero, eliminating them causes labor to move into the sector. Some of these resource gains occur at the cost of the industrial sector. Thus, industrial employment can expand more than in column (2) if I only eliminate the industrial wedges and keep those in services fixed. Column (4) reports the industrial share gain from this selective removal of wedges. Industrial employment share grows by about 18 pp on average, more than five times the expansion borne by a non-selective treatment. Each regional economy experiences robust double-digit gain, with Panama outperforming the others. But does this massive industrialization entail equally healthy economic growth? The figures in column (5) imply that it isn't so. Instead of generating efficiency gains, this selective elimination affects a contraction in five of the six economies. This finding contrasts with the previous result in column (4), where each country saw its output grow. The reason is that a selective elimination still leaves the economy distorted. Removing wedges in both industry and services leads to complete efficiency at which the output attains a maximum. But the wedges in either sector do not individually bear an 24 inverse relationship with economic efficiency. In other words, the distorted economy with no industrial wedge is not necessarily efficient compared to an economy with an industrial wedge with the same level of services wedge. As it turns out, in this case, the distorted regional economies with only services wedge are relatively more inefficient than the initial distorted states. In conclusion, eliminating labor market wedges can lead to quantitative expansion in industrial employment share. However, such industrialization will not generate substantial output growth. Furthermore, selective removal of industrial wedges can propagate inefficiency rather than curtailing it. 6 Discussion Both developed and developing countries are concerned about industrial progress, and the sector's expansion is often a policy target in many growth plans. The paper compiles data to study the industrial trends in six small Central American economies. It isolates labor market wedges as the dominant channel affecting deindustrialization. The shifts in wedges over the recent decades have increased allocative inefficiency. Therefore, a rollback in wedges to past levels can foster industrial and aggregate growth. So, does it make sense for Central America to adopt a policy that promotes industrialization through reductions in labor market wedges? Unfortunately, there are many risks associated with such an approach. First, the results in Section 5.3 outline the pitfalls of taking a sector-focused approach that targets industrial barriers to promote industrialization-led growth. While such a policy can cause a considerable expansion in industrial employment, it can also lead the economy to a more distorted state, causing a decline in output. Additionally, the results for Costa Rica and Panama show that this risk remains even when barriers in industry outweigh those in services. Therefore, only targeting barriers faced by the construction and semiconductor sectors in India and the US, respectively, even if they are most adversely affected, does not guarantee a rise in output. Second, the wedges can originate from several sources, making their identification extremely difficult. This identification problem can make even a sector-neutral approach a risky proposition. For example, consider a sectoral move that involves migrating to a new location. The resource costs of relocation are sometimes sizable (Poncet, 2006; Bayer & Juessen, 2012), and borrowing constraints can dissuade labor from moving (Bryan et al., 2014). Suspecting such a scenario, one can allocate funds for financial market interventions that make borrowing easier. However, it could also be the case that the borrowing constraints are not binding. Instead, some non-monetary factors are in fact responsible for 25 labor immobility. 12 For instance, a move might result in welfare loss stemming from differences in language, culture, laws, etc. (Belot & Everdeen, 2012). Therefore, a misidentification of actual cause might lead to interventions that not only do not reduce barriers but also waste scarce resources. On the other hand, even if the policy successfully carries out the difficult tasks of identifying the sources and implementing necessary interventions, the expected efficiency gains remain meager. Given the risks associated with the wedge-driven industrialization policies, perhaps a better strategy will be to concentrate on productivity growth, which certainly is not easy to achieve. But given the marginal gains from the barrier mitigating policies that promote industrial expansion, it might be prudent to pursue this difficult task as it has a direct and relatively more significant impact on growth. At the same time, achieving industrial productivity improvements may not always be the most straightforward option. Because productivity in the non-industrial sectors, especially services, lies far behind the global frontier in many countries, it might be easier to achieve productivity growth in such sectors. 13 Finally, marginal changes in aggregate output do not imply that the welfare implications are negligible too. Severe distributional consequences can accompany such benign aggregate effects (Fajgelbaum et al., 2020). Moreover, structural transformation bears a close relationship with regional convergence, income inequality, informality, and business cycles. Thus, a deeper evaluation of welfare implications requires exploring how deindustrialization interacts with these aspects of the economy. Additionally, an industrial decline can lead to political instability and a rise in crime and violence (Rodrik, 2016; Caceres, 2018). Such issues are fundamental for general well-being and need thorough consideration. 12 The above two factors are only among a handful of potential causes. A move often damages ties to local social networks, resulting in the loss of the associated benefits (Munshi & Rosenzweig, 2016). To the extent occupational choices extend to sectors, the wedges can arise because of gender-specific barriers faced by women in pursuing certain jobs (Sinha, 2020). A weak institutional environment can also build barriers to an efficient flow of resources (Sinha, 2021). Finally, differences in human capital and hours worked can partially explain these wedges Gollin et al., 2013; Lagakos & Waugh, 2013; Young, 2013). 13 Such productivity catch-up may not always lead to industrialization. For example, services productivity in Latin America lies far behind that of the industry, making faster productivity growth in the former relatively easier (Beylis et al., 2020). Though aiding growth, such a development will shift the comparative advantage away from the industry, creating grounds for deindustrialization. 26 References Baumol, William J. “Macroeconomics of unbalanced growth: The anatomy of urban crisis.” American Economic Review (1967): 415-426. Bayer, Christian, and Falko Juessen. “On the dynamics of interstate migration: Migration costs and self-selection.” Review of Economic Dynamics 15.3 (2012): 377-401. Belot, Michèle, and Sjef Ederveen. “Cultural barriers in migration between OECD countries.” Journal of Population Economics 25.3 (2012): 1077-1105. Beylis, Guillermo, Roberto Fattal Jaef, Rishabh Sinha, and Michael Morris. Going Viral: COVID-19 and the Accelerated Transformation of Jobs in Latin America and the Caribbean. World Bank Publications, 2020. Bryan, Gharad, Shyamal Chowdhury, and Ahmed Mushfiq Mobarak. “Underinvestment in a profitable technology: The case of seasonal migration in Bangladesh.” Econometrica 82.5 (2014): 1671- 1748. Cáceres, Luis René. “Deindustrialization and economic stagnation in El Salvador.” CEPAL Review (2017). Cáceres, Luis René. “Deindustrialization, labour and violence in El Salvador." CEPAL Review (2018). Center for Global Trade Analysis. “Global trade analysis project 9 database.” Department of Agricultural Economics, Purdue University (2015). Center for International Development. “The atlas of economic complexity”. Online database Atlas 2.5, Harvard University (2018). Chen, Chaoran, Diego Restuccia, and Raul Santaeulalia-Llopis. “The effects of land markets on resource allocation and agricultural productivity.” Review of Economic Dynamics, forthcoming, (2021). Clark, Colin. “The conditions of economic progress.” London: Macmillan (1957). Clark, Gregory. “The industrial revolution.” Handbook of Economic Growth. Vol. 2. Elsevier, 2014. 217- 262. Comin, Diego A, Danial Lashkari, and Marti Mestieri. “Structural change with long-run income and price effects.” Econometrica, 89.1, (2021): 311-374. Eaton, Jonathan, and Samuel Kortum. “Technology, geography, and trade.” Econometrica. 70.5 (2002): 1741-1779. Fajgelbaum, Pablo D., Pinelopi K. Goldberg, Patrick K. Kennedy, and Amit K. Khandelwal. “The return to protectionalism.” Quarterly Journal of Economics, 135.1, (2020): 1-55. Felipe, Jesus, Aashish Mehta, and Changyong Rhee. “Manufacturing matters… but it’s the jobs that count.” Cambridge Journal of Economics 43.1 (2019): 139-168. Gollin, Douglas, David Lagakos, and Michael E. Waugh. “The agricultural productivity gap.” Quarterly Journal of Economics 129.2 (2013): 939-993. Government of India. “Consolidated FDI policy.” Department for Promotion of Industry and Internal Trade, Ministry of Commerce and Industry, New Delhi (2020). 27 Guisan, Maria-Carmen, and Eva Aguayo. “Industry and Economic Development in Latin America.” Applied Econometrics and International Development 5.3 (2005). Guisan, Maria-Carmen, and Eva Aguayo. “Manufacturing and development in countries and areas of America, 2000-2010.” Regional and Sectoral Economic Studies 15.1 (2015): 187-192. Herrendorf, Berthold, Richard Rogerson, and Akos Valentinyi. “Growth and structural transformation.” Handbook of Economic Growth. Vol. 2. Elsevier, 2014. 855-941. Hsieh, Chang-Tai, and Peter J. Klenow. “Misallocation and manufacturing TFP in China and India.” Quarterly Journal of Economics 124.4 (2009): 1403-1448. International Labor Organization. “Key Indicators of the Labor Market.” Department of Statistics, International Labor Office, Geneva (2018). Kongsamut, Piyabha, Sergio Rebelo, and Danyang Xie. “Beyond balanced growth.” Review of Economic Studies 68.4 (2001): 869-882 Kuznets, Simon. “Modern economic growth.” New Haven: Yale University Press (1957). Lagakos, David, and Michael E. Waugh. “Selection, agriculture, and cross-country productivity differences.” American Economic Review 103.2 (2013): 948-80. Munshi, Kaivan, and Mark Rosenzweig. “Networks and misallocation: Insurance, migration, and the rural-urban wage gap.” American Economic Review 106.1 (2016): 46-98. Ngai, L. Rachel, and Christopher A. Pissarides. “Structural change in a multisector model of growth.” American Economic Review 97.1 (2007): 429-443. OECD. “Input-output tables.” OECD Publishing, Paris (2017). Pierce, Justin R., and Peter K. Schott. “The surprisingly swift decline of US manufacturing employment.” American Economic Review 106.7 (2016): 1632-62. Poncet, Sandra. "Provincial migration dynamics in China: Borders, costs and economic motivations." Regional Science and Urban Economics 36.3 (2006): 385-398. Rodrik, Dani. “Premature deindustrialization.” Journal of Economic Growth 21.1 (2016): 1-33. Sinha, Rishabh. “What explains Latin America's low share of industrial employment?” Policy Research Working Paper, WPS 8791, Washington, DC: World Bank Group (2019). Sinha, Rishabh. “Talent allocation and post-reform growth in Central America.” Policy Research Working Paper no. WPS 8791, The World Bank, Washington, D.C. (2020). Sinha, Rishabh. “Sectoral productivity gaps and aggregate productivity.” BE Journal of Macroeconomics, 21.1 (2021): 37-71. United Nations. “National accounts main aggregates database”. United Nations Statistics Division, New York, NY (2019). Uy, Timothy, Kei-Mu Yi, and Jing Zhang. “Structural change in an open economy.” Journal of Monetary Economics 60.6 (2013): 667-682. White House. “Fact sheet: The American Jobs Plan” Statements and Releases, The White House, Washington, DC (2021). Young, Alwyn. “Inequality, the urban-rural gap, and migration.” Quarterly Journal of Economics 128.4 (2013): 1727-1785. 28 A Data Appendix The paper uses data from many sources, and I have made several adjustments to the raw data. I discuss the adjustments in this appendix. Employment share: The employment data come from the International Labor Organization’s (ILO) Key Indicators of the Labor Market database (ILO, 2018). The database provides the sectoral split of the total employment across seven sectors. Table A1 lists the constituent sectors together with the aggregation scheme adopted to bring the data to three sectors. The database provides two employment series for each sector for some countries depending on whether the underlying data sourced from a labor force (LFS) or a household (HS) survey. For Nicaragua, an employment series based on official estimates (OE) is available in addition to one estimated using LFS. Aggregated Sectors Agriculture Industry Services ILO Employment by Economic Activity database Agriculture Mining and quarrying; electricity, gas, and Trade, transportation, water supply accommodation and food; business and administrative services Manufacturing Public administration, community, social and other services and activities Construction Other services United Nations National Accounts database Agriculture, hunting, forestry, Mining, utilities Wholesale, retail trade; restaurants fishing and hotels Manufacturing Transport, storage and communications Construction Other services activities Atlas of Economic Complexity database Vegetables, foodstuffs, and wood Chemicals and plastics Services Electronics Machinery Metals Minerals Stone and glass Textiles and furniture Transport vehicles Other manufacturing goods Table A1: Aggregation Scheme The raw data from the ILO database in some country-years are not suitable for analysis because of the nature of underlying surveys. For instance, figure A1 plots the agricultural and industrial share for El Salvador using the raw data. After declining steadily from just under 50 percent in 1975 to about 35 percent in the next 10 years, the reported agricultural share falls to below 2 percent in 1986 before 29 recovering to 35 percent again in 1992. Correspondingly, there are significantly large changes of opposite nature in industrial shares during the same period. The reason why such large changes are obtained using the data is because the surveys used to estimate sectoral employment are not representative. The surveys during 1986–1991 covered either main cities and metropolitan areas or urban areas. Not surprisingly, there was a sharp drop in the share of agriculture in total employment owing to its rural affiliation. The ILO database lists these flags in their notes which I use to identify such surveys and drop them from the analysis. Figure A1: Unadjusted agricultural and industrial employment share: El Salvador The figure shows the unadjusted employment shares of agriculture and industry as estimated from the raw data from ILO’s employment data. The next step entails merging the various employment series to maximize coverage (years) at the country level. There is only one series available for Honduras (HS) and Panama (LFS). In Costa Rica and Guatemala, there are periods when data are available from both series, but one series dominates the other in terms of time coverage. Hence, it is not possible to extend coverage. Moreover, the shares in overlapping periods are not very different across the available series. I choose the series that provides information for the longer period for each country. Merging data from two series in the case of El Salvador helps in extending coverage. The employment shares in overlapping years across the series also lie close to each other. For Nicaragua, the shares can be constructed using data from LFS or what the ILO mentions as official estimates. The two series covers very different time periods and there are large deviations in shares across the two over a span of just two years. From being around 45 percent of total employment in 2001 as reported using OE, the employment share drops to 30 percent in 2003 when estimated using the LFS data. Correspondingly, there are large jumps in the share of services as well. To focus on structural transformation that happened more recently, I pick the shares constructed using the LFS data. 30 Finally, as the interest of the study is to focus on long-term changes, I smoothen the annual series using the HP-filter with a smoothening factor of 40 to screen the cyclical components. I also smoothen the other annual series used in analysis for the same purpose. Value-added share: The National Accounts Main Aggregates database of the United Nations (UN, 2019) contains long time-series data on gross value-added by kind of economic activity at current prices in national currency which I use to arrive at the value-added shares for the three broad sectors. The UN data splits total value-added into 7 economic sectors and the aggregation scheme is listed in table A1. The aggregation is consistent with the aggregation adopted for employment shares. Before taking the estimated value-added shares for analysis, I check if there are larger changes in them over short duration that may imply inconsistencies in the data. While the series for Costa Rica, Honduras and Panama exhibit a steady behavior, I do notice large jumps in the case of Guatemala (from 2000 to 2001), Nicaragua (from 1986-1987) and El Salvador (from 1989 to 1990). For example, in Guatemala, the agriculture share drops by more than 9 percentage points while the industrial share gains 7.5 percentage points going from 2000 to 2001 (figure A2 (a, b)). To avoid these large changes which seem unlikely, I consider the data after the large jumps for each country. This limits the time coverage substantially for Guatemala. Figure A2: Value-added shares estimated from UN data (a) Agriculture (b) Industry Export and import shares of GDP: The data on exports and imports are taken from the National Accounts Main Aggregates database of the United Nations (UN, 2019). The database’s series on gross domestic product by expenditure (at current prices) reports the exports and imports for the entire economy at current prices in the local currency. Export and import shares at sectoral level: To find the export and import shares at the sectoral level, I split the aggregate trade shares obtained in previous step making use of the Atlas of Economic Complexity (Center for International Development, 2018) database that reports a country’s exports and import across 11 sectors. The aggregation scheme followed to bring the Atlas data to three sectors is listed in table A1. 31 Sectoral linkages: I use the OECD input-output database (OECD, 2017) to get the intermediate shares at sectoral level for Costa Rica taking the average of all the years for which the data are available. Unfortunately, the input-output tables of reasonable quality of other Central American economies are not available. For these countries, I use the average intermediate shares across eight countries including Costa Rica. Figure A3 plots the average across the eight countries together with the average values at the country level. In general, the shares across countries lie close to each other. I follow the aggregation scheme (Table A2) followed in Sinha (2019) which is consistent with the aggregation adopted for other series in the analysis (Table A1). Aggregated Sectors Agriculture Industry Services OECD Input-Output Database Agriculture and Fishing Mining Wholesale & Retail Trade Manufacturing: Food, Beverage & Tobacco Hotels & Restaurants Manufacturing: Textile, Dress, Leather Transport & Storage Manufacturing: Wood Post & Communication Manufacturing: Paper & Printing Financial Intermediation Manufacturing: Petrol Real Estate Manufacturing: Chemicals Renting of Machinery & Equipment Manufacturing: Rubber & Plastics Computer Related Activities Manufacturing: Non-metallic Minerals R&D & Other Business Activities Manufacturing: Basic Metals Public Administration & Defense Manufacturing: Fabricated Metals Education Manufacturing: Machinery Health & Social Work Manufacturing: Advanced Machinery Community & Personal Service Manufacturing: Electrical Machinery Private HHs with Employed Persons Manufacturing: Motor Vehicles Manufacturing: Transport Equipment Manufacturing: Furniture & Recycling Utilities Construction OECD National Accounts Database (table 5) Food & non-Alcoholic Beverages Durable Goods Services Alcoholic Beverages, Tobacco & Narcotics Semi-Durable Goods Non-Durable Goods (less agriculture) GGDC Productivity Level Database Agriculture, Forestry & Fishing Mining & Quarrying Wholesale & Retail Trade Manufacturing Hotels & Restaurants Utilities Transport & Communications Construction Financial & Business Services Community, Social & Personal Services Table A.2: Aggregation Scheme 32 Figure A3: Intermediate shares at sectoral level The figure shows the intermediate shares for eight countries and the average across them. The country-level shares are average across the years for which the data are available in the OECD input-output database. References Center for International Development. “The atlas of economic complexity.” Online database Atlas 2.5, Harvard University (2018). International Labor Organization. “Key Indicators of the Labor Market.” Department of Statistics, International Labor Office, Geneva (2018). OECD. “Input-output tables.” OECD Publishing, Paris (2017). Sinha, Rishabh. “Input substitutability and cross-country variation in sectoral linkages.” Economics Letters, 178(C), 121-124 (2019). United Nations. “National accounts main aggregates database”. United Nations Statistics Division, New York, NY (2019). 33