Policy Research Working Paper 9478 Ex-Ante Evaluation of Sub-National Labor Market Impacts of Trade Reforms Maryla Maliszewska Israel Osorio-Rodarte Rakesh Gupta Macroeconomics, Trade and Investment Global Practice November 2020 Policy Research Working Paper 9478 Abstract A macro-micro simulation framework that links a com- sectors employing a higher proportion of women. In the putable general equilibrium model with the survey-based absence of additional policies, growth is not equally distrib- global income distribution dynamics model can be used uted. In all the scenarios in which the Sri Lankan economy to assess the economic and distributional effects of macro- grows, the distribution of gains is regressive. Increasing economic shocks and policies. The methodology is used to labor demand for skilled workers translates into a larger assess the economic and subnational labor market impacts skilled wage premium—by as much as 1.1 percent with of a series of stylized trade policy options for the Sri Lankan respect to the baseline. Implementation of full trade reform economy over a 10-year time period. The analysis focuses accelerates the concentration of economic activity in the on the impact of unilateral para-tariff liberalization, free- western regions of Colombo, Gampaha, and Kalutara. Net trade agreements with China or India, and a full-reform employment gains in the western regions would increase scenario. The simulation results show that more ambitious from 111,000 to 136,000 in the full reform scenario by trade reform can result in larger gains in gross domestic 2028 and with respect to baseline conditions. product, poverty reduction, and exports, particularly in This paper is a product of the Macroeconomics, Trade and Investment Global Practice. 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 authors may be contacted at mmaliszewska@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 Ex-Ante Evaluation of Sub-National Labor Market Impacts of Trade Reforms Maryla Maliszewska, Israel Osorio-Rodarte and Rakesh Gupta1,2 Keywords: economic growth of open economies; measurement and analysis of poverty; computable general equilibrium modeling; distributional impacts of trade; trade agreements; Sri Lanka JEL Classification: F16, I32, F61, F63, O53, C68 1 Maryla Maliszewska, Israel Osorio-Rodarte, and Rakesh Gupta: The Word Bank. The paper is part of the background analysis for the World Bank Report, Distributional Impacts of Trade (2020). It was generated under the guidance of Caroline Freund, Antonio Nucifora and Carolina Sanchez-Paramo. We received comments and suggestions from Erhan Artuc, Paulo Bastos, Paul Brenton, Maurizio Bussolo, Erik William Churchill, Jakob Engel, Carmen Estrades, Michael Ferrantino, Sanjay Kathuria, Deeksha Kokas, Tae Hyun Lee, Gladys Lopez-Acevedo, Mariem Malouche, Ambar Narayan, David Newhouse, Bob Rijkers, Raymond Robertson, Carlos Rodriguez-Castelan, Sebastian Saez, Hans Timmer and participants of the various World Bank seminars and trainings where this paper was presented. 2 This paper is a product of the staff of the International Bank for Reconstruction and Development/the World Bank. The findings, interpretations, and conclusions expressed in this paper do not necessarily reflect the views of the World Bank, the Executive Directors of the World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. This material should not be reproduced or distributed without the World Bank's prior consent. 1. Introduction Trade reform inevitably produces winners and losers. Most economists accept that, in the long run, economies open to international trade perform better than do closed ones, and that relatively open trade policies contribute to sustained economic development. The long-term gains from trade openness have been discussed extensively. 3 By providing incentives to reallocate resources, openness to trade aligns a country’s economic activity with its comparative advantage. Of concern, however, is that trade liberalization could harm disadvantaged groups and that even in the long run, successful open economies may leave some vulnerable people permanently behind. International trade operates in complex economic and political contexts. By creating winners, losers and a potentially large pool of resources to reallocate, trade reform has inherent political economy repercussions. During recent decades, the regional integration agenda was motivated by the need to modernize economies and integrate into global value chains as well as a tool to dismantle monopolies, rent- seeking protectionist schemes, or obsolete labor arrangements. The political economy consequences of trade were recognized early in the academic literature (Hillman 1982; Cassing and Hillman 1985; Baldwin 1989; Rodrik 1995). It has been understood that assessing the economic and distributional consequences of trade policies is key in designing adequate policy implementation and balancing possible options for advancing reform. The assessment of the distributional aspects of trade reform requires a systematic approach. Studying the poverty and distributional effects of trade reform or large trade shocks requires a methodology that accounts on the one hand for the aggregate nature of the policy or shock being studied and, on the other hand, the heterogeneity of the overall effects among individuals or households. The traditional randomized trial offers only limited scope for measuring the impact. As noted by Bourguignon and Bussolo (2013) in the context of macroeconomic policies, ex-post evaluation methodologies based on the comparison between individuals exposed and not exposed to the policy or shock are not applicable since, by definition, all individuals are affected by a policy with macroeconomic dimensions. The evaluation methodology thus needs to rely not only on micro data but also on the creation of a macro counterfactual within a general equilibrium setting. This paper discusses a simulation framework used for the ex-ante assessment of the distributional aspects of trade. The methodology is applied to Sri Lanka for the evaluation of relevant trade policy simulations, such as potential impacts of the unilateral tariff and para-tariff liberalization and implementing a series of free-trade agreements (FTAs) with China and India. This methodology combines a computable general equilibrium (CGE) model linked to a microsimulation in a top-down approach. This modeling framework allows for the incorporation of the complex interactions across agents (consumers, producers and government) at the country, sector and factor levels. It captures changes in comparative advantage and trade flows following trade liberalization and shifts in demand as income rises. The analysis includes 17 sectors and 35 trading partners (see Table A1) and simulates the impacts of policy changes up to 2028. This paper is organized as follows. Section 2 presents the methodology used in this paper for the ex- ante evaluation of distributional impacts of trade. Section 3 elaborates on a set of policy-relevant scenarios for the case of Sri Lanka. Section 5 presents the simulation results at the macro, sectoral, and district levels. Lastly, section 6 concludes with recommendations for further development. 3 Initially, the openness to trade discussion circled around three pillars: does trade liberalization stimulate growth and relieve poverty? does it boost productivity? are open economies less stable? An early review of the literature is presented in Winters, McCulloch, and McKay (2004). For a recent review of the literature, see (Kokas and Engel, 2020.). 2 2. Macro-micro methodologies for ex-ante evaluation of the distributional impacts of trade The importance of assessing the distributional gains from trade has been long recognized in economic thinking. The origin of dynamic microsimulation can be traced back to the 1950s seminal work of Orcutt (1957) whose contributions aimed at overcoming the limitations of models available at that time. Orcutt observed that those earlier models could be used to predict the aggregate impact but could not describe the distributional impacts of policy reforms nor the effects on inequality of long-term policy options or shocks. Data availability and modeling have significantly advanced since then, yet dynamic microsimulations remain the main tool to study distributional change and to provide the unique perspective of projecting samples of population forward in time. Traditionally, tools for policy analysis concerned with distributional issues tend to be either exclusively micro or macro. Microsimulation models based on household samples are used for studying reforms of a variety of policy choices, such as tax-benefit systems, impact of delivering public goods like education or health care, or changes in labor market regulations. Macro modeling, on the other hand, is typically used to measure the impact of reforms in tax, trade, finance or monetary policies on the sectoral structure of the economy, its level of employment and wages. From the aggregate information, it is possible to inform which groups in the population will benefit or lose from the reform being analyzed. Aggregate outcomes from macro models are insufficient to assess how individual households might be affected by the policy being analyzed. Likewise, seemingly micro-oriented policies like a reform in cash transfers or in the pension system may have important macro effects through labor supply and savings, in turn, these macro effects of micro-oriented policies will impact wages, employment and returns to wealth affecting the income distribution within a population. Hence, the need to combine macro and micro approaches to policy analysis is obvious. Several approaches have been proposed to integrate macro-micro frameworks that allow for the simultaneous study of the distributional and macro impact of policy reforms (Bourguignon and Bussolo 2013). The usual problem of data availability and quality is, in the case of a macro-micro model, compounded because data from national accounts and from household surveys – the two most common sources of macro and micro data – need to be reconciled, let alone the economic treatment of their respective sources. In this context, the modeling framework used in this paper represents a departure from the static simple top-down approach. It incorporates a macro-micro framework operating in a dynamic setting. This paper discusses the CGE-GIDD macro-micro methodology that can be used for the ex-ante assessment of trade policy and shocks. This paper uses the Global Income Distribution Dynamics (GIDD) microsimulation framework in combination with the global computable general equilibrium (CGE) model called LINKAGE/ENVISAGE. CGE-GIDD is a “top-down” macro-micro simulation framework that exploits heterogeneity observed in household surveys to distribute macroeconomic shocks. CGE-GIDD was developed by the World Bank (Bussolo, De Hoyos, and Medvedev 2010), inspired by previous efforts involving simulation exercises (Bourguignon, Ferreira, and Leite 2008; Bourguignon, Bussolo, and Pereira da Silva 2008; Davies 2009). The full technical documentation of the LINKAGE/ENVISAGE model is provided by van der Mensbrugghe (2013; 2020); the following sub-section briefly describes the main structure of the models. 3 3. The CGE-GIDD microsimulation model4 The CGE-GIDD models are connected through changes in labor supply (demographic change), skill formation (educational attainment), and real earnings. In terms of labor supply, the macro and micro models incorporate projections for skilled and unskilled labor over time. These projections are based on standard population projections provided by the United Nations. The GIDD framework, in addition, reallocates workers across sectors5 based on aggregate general equilibrium conditions. Lastly, to estimate the impacts on real earnings, the GIDD framework incorporates the CGE-based simulated results in skilled wage premia, income growth, and changes in relative prices for food and nonfood expenditures. The general notion of the model is depicted in Figure 1 below. Figure 1: GIDD methodological framework Source: Bourguignon and Bussolo (2013) 3.1. The LINKAGE Computable General Equilibrium Model The microsimulation can portray the behavior of consumers and workers, but it does not have a market-clearing mechanism because it does not include the suppliers of goods and demanders of labor, i.e., the firms. This creates the need for having a source of information on macro variables such as the level of wages, the change in relative prices, and overall income growth. At its core, the macroeconomic model is essentially a neo-classical growth model, with aggregate growth predicated on assumptions regarding the growth of the labor force by skill level, savings/investment decisions (and therefore capital accumulation) and productivity. 4 These methodological sections are based on Ahmed et al. (2020) and Balistreri et al. (2018). 5 Rice, vegetables and fruits, other crops, fishing, other agriculture, energy, food, milk production, other food, textiles, wearing apparel & leather, chemicals, margins, motor vehicles, other machinery, other manufacturing, utilities, construction, sea transportation, other transportation, communication, business services, and social & other (unspecified) services. 4 The CGE model relies on neoclassical economic theory to model macroeconomic behavior. Production is modeled using nested Constant Elasticity of Substitution (CES) functions that combine at various levels, with different substitution elasticities, intermediates and primary factors. 6 Households’ consumption demand is derived from the maximization of household-specific utility function following the Linear Expenditure System (LES). International trade is modeled assuming imperfect substitution among goods originating in different geographical areas. Demand for foreign good result from a CES aggregation function of domestic and imported goods. Export supply is symmetrically modeled as a Constant Elasticity of Transformation (CET) function. Producers decide to allocate their output to domestic or foreign markets responding to relative prices. The assumptions of imperfect substitution and imperfect transformability grant a certain degree of autonomy of domestic prices with respect to foreign prices and prevent the model from generating corner solutions. 7 The labor market specification is a key element of LINKAGE and an important driver of the distributional results. Therefore, its specification calls for some clarification and justification. Two types of labor are distinguished, skilled and unskilled. 8 These categories are considered imperfectly substitutable inputs in the production process. Moreover, some degree of factor market segmentation is assumed: skilled workers are perfectly mobile across sectors, whereas, in developing countries, the labor market for the unskilled is divided into agriculture and non-agriculture segments. The labor market segmentation by skill level has become a standard assumption in CGE modeling. The imperfect substitution in the production process for workers with different skills is likely to persist for the medium-term time horizon, as unskilled workers cannot be ‘transformed’ into skilled ones, even with increased on-the-job training. The CGE can incorporate several assumptions on the structure of labor markets and complementarities between workers and technology. In the standard setting, the market for unskilled labor can be segmented into agricultural and non-agricultural activities. This can be controversial. However, econometric analysis indicates that a gap in remunerations between these two segments remains even after controlling for education, gender, experience and other variables including cost of living differentials (between rural areas, where agricultural activities are predominantly located, and urban ones). Some barriers to mobility – land ownership providing economic security to farmers, specificity to human capital acquired in agriculture, or others – must exist and hinder equalization of wages across the two segments. In the model, this segmentation is implemented with some flexibility. Using a Harris-Todaro specification, 6 At the top of the nest, a value-added bundle is combined with an intermediate-inputs bundle under the Leontief technology assumption. The intermediate inputs bundle is combined with different inputs, with an Armington assumption applied to specific inputs. That is, for a given type of intermediate input, there is substitutability between domestic and imported inputs, and then again between imported inputs from different source countries. The value-added bundle is made up of unskilled labor being slightly substitutable with a capital and skilled labor bundle. The choice of elasticity often depends on the structure of the model and the level of aggregation. We find that an elasticity of 2 between capital and skilled labor and the same value between skilled and unskilled labor work well for the level of disaggregation in the paper and from previous testing and applications by other researchers (such as Ahmed et al. (2016), Devarajan et al. (2015), and Francois Bourguignon, Bussolo, and Pereira da Silva (2008)). A lower elasticity would be appropriate the more aggregated the structure of the economy and its factor markets. At the extreme, when the economy is one sector with one labor and capital (a Solow growth model), the elasticity between labor and capital traditionally takes the value of one (a Cobb-Douglas function in Solow (1956)) or less than one (a CES function) in more recent findings (e.g. León-Ledesma, McAdam, and Willman (2010) or La Grandville (2017). The same tendency appears to be the case with trade related elasticities and the level of commodity specificity (e.g., Hillberry and Hummels (2013) and Devarajan, Go, and Li (1999)). 7 However, even if these assumptions attenuate factor prices equalization when countries trade with each other, trade is still an important channel through which factor prices are influenced. 8 Recent applications distinguish work by gender. 5 a certain number of unskilled workers migrate from one segment to the other in response to changes of the wage differentials across the segments. 9 This rich set-up can capture the changes in wages for different types of workers. Since skilled-unskilled and agriculture–non-agriculture wage gaps represent important drivers of inequality, the set-up enables the delineation of changes in the income distribution of each economy. The evolution of the labor force is in line with the change of working age population. 10 The growth of supply of skilled and unskilled workers is consistent with this overall growth of the labor force present in the microsimulation model. Capital accumulation, labor expansion, and productivity changes determine the growth dynamics in the model. In each period, capital stock is equal to the depreciated capital stock from the previous period augmented with the investment. Investment is driven by savings which, in turn, depend on per capita income growth and the population youth and elderly dependency ratios. In contrast to national savings, foreign savings are exogenous. The model assumes heterogeneity of productivity trends across sectors. Productivity in agriculture is assumed to be factor-neutral and exogenous and is set to estimates from empirical studies (e.g., Martin and Mitra (1999)). Productivity in manufacturing and services is labor-augmenting; it is skill-neutral but sector-biased. In the case of agriculture, productivity growth averages 2.5% per annum for all countries. In manufacturing and services changes in productivity are country-specific and based on past trends. Following the broad findings of Bosworth and Collins (2003) productivity growth in manufacturing is assumed to grow about three percentage points faster than in services. 11 This model calculates wages of skilled and unskilled workers by combining firms’ demand for labor with the aggregate change in supply of these two types of workers. In addition to the changes in rewards, the computable general equilibrium model also calculates the overall economic growth and sectoral reallocation of labor. 12 As already described above, changes of these variables – the Linkage Aggregate Variables (LAVs) – are used as inputs in the final step of the microsimulation. 9 We do not distinguish agricultural sector and rural regions, which may neglect the presence of rural non-farm activities. See, for example, Reardon, Stamoulis, and Pingali (2007). 10 This means that participation rates are constant. This is a conservative assumption given that the overall labor force participation rates are likely to increase as more women enter the labor force in developing countries or due to the changes in the retirement age in high income countries. The United Nations population projections are disaggregated by age groups and gender. They include net migrations per 1,000 people by country/region. We do not provide alternative migration scenarios, which can be complex and beyond the scope of the paper. However, migration can certainly be handled by the methodology, see for example, Ahmed et al. (2016). 11 The dynamic calibration allows the model to reproduce, in its baseline scenario, GDP growth rate trends that are in line with available long-term projections such as, for example, the growth projections under the Shared Socio-Economic Pathway Scenario 2 of Dellink et al. (2017). In fact, during the dynamic calibration the model targets these exogenous growth rates of GDP by endogenously determining the labor augmenting productivity factor described in the text. In the counterfactual scenarios, the real GDP growth rate is endogenized, and the uniform productivity variable is exogenized, and simulated to grow at the rate determined in the baseline scenario. The real growth rate can thus change depending on the other shocks considered in the scenario. 12 In the GIDD microsimulations, workers move across sectors to achieve the proportions of employment in agriculture and non-agriculture calculated by the CGE model. Note that these inter-sectoral movements are net of the sectoral shifts already generated by the reweighting procedure. The microsimulation procedure to select which specific worker moves is based on a probabilistic model described in detail in (Bussolo, De Hoyos, and Medvedev 2010). 6 3.2. The Global Income Distribution Dynamics Model The first step in the microsimulation exercise is to implement a set of changes in the household surveys’ demographic structure. The population growth adjustment is particularly important in countries with rapid demographic changes. In practical terms, the adjustment for population growth allows the analysis to explicitly take into account the changes in the size of the working-age population. 13 We perform population and education projections during the first stage of the microsimulation model and in creating the baseline scenario for the CGE model. For each country, we construct the demographic profile in two steps. First, the age and gender composition are exogenously determined following medium variant estimates from the World Population Prospects (UN DESA 2019). In the second step, following Bourguignon and Bussolo (2013), country-specific educational profiles are constructed using initial educational achievement levels observed in the household surveys with some conservative yet simple assumptions about educational progress. The microsimulations’ second step adjusts individual factor returns by skill and sector in accordance with the results of the CGE model. The GIDD imposes an entirely new vector of earnings on each worker, conditional on each worker’s individual characteristics. In a basic setup, the CGE and GIDD can be linked with two sectors (agricultural and non-agricultural) and two types of workers (skilled and unskilled) but depending on the quality of the data and the scope of the study, this restriction can be relaxed. 14 The GIDD reallocates workers moving them out from shrinking into expanding sectors. The sectoral reallocation process estimates the probability of each worker to be reallocated into new sectors, based on individual characteristics. Once workers are reallocated, a new vector of earnings is generated using estimates from a set of Mincer equations. To account for unobserved characteristics, each individual residual is brought into the new sectors and scaled accordingly. This latter process assures that an individual switching sectors can carry his unobservable personal characteristics - beyond age, experience and years of schooling. The third step adjusts average wages between groups of workers and sectors. While the second step operates at the individual level, the third step operates at the group level scaling the average wages for each type of worker and sector. In practical terms, one group is selected as numeraire, i.e. unskilled agricultural, and average wages for each group are scaled relative to the numeraire. Within a group, all earnings are scaled with respect to the numeraire group. Operating through changes in relative wages guarantees internal consistency between macro and micro results considering that in the CGE-GIDD linkage, initial relative wages were obtained from the micro data and feed into the global CGE model. It is important to highlight that until this point the microsimulation has operated only in relative terms. Lastly, GIDD adjusts the average income/consumption per capita to guarantee that it changes exactly in line with the CGE results. After creating new earnings for workers, a vector of per capita household income is constructed considering new earnings and household size. When information about the relationship between incomes and savings exists, if not, a one-to-one passthrough from per capita household income to consumption is assumed. In this regard, GIDD constructs a household-specific deflator to adjust for changes in relative prices. The price deflator is constructed using initial and final price indexes of food versus non-food expenditure from the macro model and household-specific budget consumption shares for food and non-food expenditure observed in the micro data. Food and non-food shares can be adjusted estimating a Lorenz’s curve of budget share expenditure on food and total per capita household consumption. These steps are explained in detail in the annex of this paper. 13For the case of Sri Lanka, as will be shown later, working age population is projected to diminish by approximately 100,000 individuals by 2030. 14 See for instance, World Bank (n.d.). 7 4. Applying the Macro-Micro Simulations to Potential Trade Policy Reforms in Sri Lanka To illustrate the use of this methodology, the CGE-GIDD approach is used to assess a set of ex-ante trade policy simulations for the case of Sri Lanka. Sri Lanka is a country that has some key characteristics that make the country a good candidate to showcase the methodology. First, it is a small open economy and effectively a price taker in international markets. Second, as other developing countries, its export basket is dominated by a primary commodity – in this case exports of tea. Lastly, after a period of political instability that ended in 2009, the economy grew at an average 5.8 percent during the period of 2010-2017. This reflects a determined policy momentum towards reconstruction and rapid structural transformation, despite signs of a slowdown in the last few years. This paper studies the impacts of stylized policy scenarios including assessments of unilateral para-tariff liberalization, free trade agreements with China and India, and a full reform scenario (para-tariffs + FTAs). Section 3.1 discusses the current structure of the labor market in Sri Lanka while Section 3.2 describes the details of the baseline and trade policy scenarios. Simulation results are discussed in Section 4, first focusing on the sectoral level and then on the sub-national distribution of those effects. The model can be used to assess the sub-national, localized effects of trade. The impact of reforms is differentiated across types of households and workers. Such heterogeneity is key in determining the welfare (consumption, poverty, inequality, etc.) and distributional impacts of any reforms to be undertaken. The GIDD distributes the macroeconomic results of the CGE model to households in line with the Sri Lanka Household Income and Expenditure Survey (HIES 2016). The microeconomic model distributes shocks while keeping the consistency with the aggregate results obtained from the macro CGE model. 4.1. Employment and skills profile The Sri Lanka Household Income and Expenditure Survey (HIES) 2016 was utilized to examine the characteristics of workers at the sectoral and district levels. 15 The HIES 2016 data set consists of approximately 21,755 households and 82,935 individuals found within these households. We focus on profiling workers in each sector according to the localized Sri Lanka official version of the 1-digit and 4-digit International Standard Industrial Classification (ISIC), 16,17 the province in which they live, median wage and skill level proxied by educational attainment. Women account for one-third of employment of 7.7 million workers and they are primarily employed in the agriculture (incl. vegetables and fruits) sector. Male workers comprise 5.07 million (66 percent) and female workers comprise of 2.63 million (34 percent). Agriculture (25 percent), manufacturing (18 percent), trade and commerce (18 percent), and social services and others (12 percent) sectors are the four largest employers in Sri Lanka according to estimates from HIES 2016 (Table 1). 18 15The survey was conducted by the Department of Census and Statistics (DCS) under the National Household Survey Programme (NHSP) of Sri Lanka. The survey is nationally representative, and the unit of analysis is the household with details on all the individuals comprising the household. The sampling frame consists of two-stage stratified sampling – area of residence (urban and rural) and the estate sectors in each district of the country were the selection domains. This makes our results representative at the district level too, which is essential for the results of the CGE-GIDD at the regional level. 16 http://www.statistics.gov.lk/industry/ASI_2016_Report.pdf. 17 Sri Lanka HIES 2016 is utilized for the descriptive analysis of workers section, despite that wage information is more reliable in the Sri Lanka Labor Force Survey (LFS 2015). 18 Analysis presented in this section relies on the 1-digit ISIC industry classification (10 sectors). 8 Women in Sri Lanka are mainly employed in the agriculture sector. Around 28 percent (approximately 740,000) of women are employed in agriculture, compared with 25 percent in the manufacturing sector – approximately 650,000, social services and other sectors – 18 percent (approximately 470,000), and commerce sector – 12 percent (approximately 325,000) of the total jobs for women. These sectors are also the four largest employers of women (Figure 2 and Table 1). On the other hand, more than 50 percent of men are employed either in agriculture, manufacturing or trade-related activities. Table 1: Estimated number of workers, by sector and gender Total Men Women Estimated Estimated Estimated Sector Obs. % of total Obs. % of total workers workers workers Agriculture 1,913,073 5,066 1,175,861 61.5 2,986 737,212 38.5 Mining 51,798 186 48,475 93.5 15 3322.562 6.4 Manufacturing 1,404,570 2,879 754,379 53.7 2,547 650,191 46.3 Public utilities 39,633 133 32,441 81.9 31 7191.704 18.2 Construction 607,792 2,378 586,746 96.5 83 21,046 3.5 Commerce 1,063,299 2,869 738,379 69.4 1,247 324,920 30.6 Transports & communications 737,296 2,532 648,494 88.0 350 88,802 12.0 Financial & business services 356,071 906 237,835 66.8 449 118,236 33.2 Public administration 575,378 1,463 360,764 62.7 872 214,613 37.3 Other services, Unspecified 960,634 1,891 492,266 51.2 1,804 468,368 48.8 Total 7,709,543 20,303 5,075,641 65.8 10,384 2,633,903 34.2 Source: Staff estimates based on Sri Lanka HIES 2016; confidence intervals of estimated workers can be requested from the authors. 9 Figure 2: Sector-level concentration of jobs, by gender Sector composition (%) by sex Agriculture 38.5 61.5 Mining 6.4 93.6 Manufacturing 46.3 53.7 Public 18.1 utilities 81.9 Construction 3.5 96.5 Commerce 30.6 69.4 Transport & 12.0 communications 88.0 Finance & 33.2 Business 66.8 Public 37.3 administration 62.7 Other 48.7 services 51.3 0 20 40 60 80 100 Percent Female Male Source: Staff estimates based on Sri Lanka HIES 2016; confidence intervals of estimated workers can be requested from the authors. Manufacturing jobs are concentrated in some western and northern provinces. Manufacturing jobs are concentrated in the Colombo, Gampaha and Kalutara in the Western Province; Kurunegala and Puttalam in the North Western province and Kandy in Central province (Figure 3). The jobs in the textiles and wearing apparel sectors are similarly distributed as the manufacturing sector, which tend to demand similar infrastructure. The level of urbanization is higher in these districts compared to the other districts or provinces in Sri Lanka. Trade and transportation related jobs are rather evenly spread out by districts, except for Colombo and Gampaha districts that employ a disproportionately higher share of jobs in this sector. A similar profile with the social services sector (education, health care and domestic personnel) is observable with jobs evenly distributed across Sri Lanka. However, most jobs in the agriculture sector are found in districts outside the Western, North Western and Central provinces (Figure A32-Figure A35 in Annex). 10 Figure 3: Geographic (district-level) concentration of jobs in Sri Lanka by sectors Source: Staff estimates based on HIES 2016 Public administration, public utilities and finance and business sectors have the largest average monthly wages (over LKR 25,000); while the agricultural sector has the lowest estimated monthly wages with a little less than LKR 15,000 nationally on average (Figure 6). The average total monthly wage in Sri Lanka stands at approximately LKR 25,780.5 (mean) and LKR 21,000 (median) in 2015 nominal currency units. The distribution of the monthly wages also displays a usual positive skewness (right-skewed) that is often observed in income or wage distributions. In terms of wages disaggregated by gender - the mean is greater than the median wage for men and women. Women earn lower median monthly wages compared to men across most sectors. Male workers earn more than female workers (Figure 4 and Figure 5) on average by 26 percent. Wage gaps are also the largest among sectors where women are most employed – women employed in the agriculture sector face the largest proportional wage gaps (Table 1 and Figure 6). This is true except for higher average wages in one sector – social services and others sector. This can be explained by the fact that 7 percent of the total employed workforce consists of non-paid employees. As in many developing countries, women tend to work significantly more in jobs without pay. And 15 percent of all employed women fall under this category, vis-à-vis 2 percent for the case of men. This also implies that the wage analysis with descriptive statistics presented here does not include this significant group of workers classified as “non-paid employee”.19 19 Employment estimates presented here from HIES 2016 consider only the main/primary jobs the individual or respondent has held in the last reference period to assign the individual to a type of occupation and do not consider the second job. However, wage analysis, to compute median wages, includes total of wages from main and secondary jobs. 11 Figure 4: Average (median and mean) wages by Figure 5: Density of log monthly wages by gender gender Source: Staff estimates based on HIES 2016 Source: Staff estimates based on HIES 2016 Women, on average, have higher educational attainment (in terms of mean years of schooling) in recent years or decade-cohorts compared to previous generations. Furthermore, rural women now have (1990s cohort) quite similar mean years of education compared to men in urban areas (Figure 7). This trend will aid to close the gender wage gaps if the descriptive evidence will hold true in the projections for the future, baseline or otherwise. However, wage gaps continue to persist in favor of men after controlling for educational attainments. The elements portrayed in this section inform the CGE model, particularly on the intensity of skilled and unskilled labor use. In most recent applications, the split by gender can be incorporated with very detailed sectoral information (World Bank, n.d.). 20 20 The Gender Disaggregated Labor Database (GDLD) is a global micro labor force database based on the World Bank’s household survey collection and other public resources. This database includes harmonized economic activities and occupation categories from local classification to international comparable classifications in detailed levels. It fills an important information gap in global gender statistics by providing detailed accounts on education, employment levels, wages, labor income, and employment status at a very disaggregated (2-digit ISIC level) economic activity level and occupation category than is commonly available. 12 Figure 6: Median wages by gender & sectors Average (median) total monthly wage Public administration Public utilities Finance & Business Other services Construction Transport & Communications Manufacturing Commerce Mining Agriculture 0 10,000 20,000 30,000 40,000 Female Male Source: Staff estimates based on HIES 2016 13 Figure 7: Years of education, by gender and location 12 10 Years of education 8 6 1940 1950 1960 1970 1980 1990 Women/Rural Women/Urban Men/Rural Men/Urban Source: Staff estimates based on HIES 2016. Note: 1990s cohort refer to individuals born between 1990 and 1995. Average mean years of education by decade-cohort of individual’s year of birth; individuals in 1990 cohort may not have fully completed education at the time of survey (2015), and hence these individuals could attain higher mean years of education. 4.2. Defining Trade Policy Scenarios21 4.2.1. Baseline As mentioned above, an ex-ante evaluation requires the creation of a counterfactual simulation that will serve as a business-as-usual scenario with no policy change. The model relies on the GTAP version 9 database benchmarked to 2011. The CGE model runs up to 2028, replicating the key historical macroeconomic aggregates from the World Bank World Development Indicators (World Bank, n.d.). Productivity growth in the baseline scenario is calibrated to achieve the historical and forecasted GDP growth rates up to 2021. Productivity growth is adjusted thereafter to be consistent with historical trends. 22 The baseline scenario also incorporates tariff reductions under existing FTAs. These are based on the data set provided by the International Trade Center, including all TPP members’ FTA commitments up to 2027 (International Trade Centre 2016). We also assume that Sri Lanka enjoys the Generalized System of Preferences (GSP) preferences on its exports to EU markets. The baseline contemplates continuation of recent trends using standard closure rules for macroeconomic accounting. The scenarios incorporate three closure rules. First, government expenditures are held constant as a share of GDP, fiscal balance is exogenous while direct taxes adjust to cover any 21 This section was largely was developed by Guillermo Arenas (Economist, World Bank Group) using the Tariff Reform Impact Simulations Tool. 22 These productivity assumptions are not altered in in the counterfactual scenarios. 14 changes in the revenues to keep the fiscal balance at the exogenous level. The second closure rule determines the investment-savings balance. Households save a portion of their income, with the average propensity to save influenced by elderly and youth dependency rates, as well as GDP per capita growth rates. The savings function specification follows Loayza, Schmidt-Hebbel, and Servén (2000) with different coefficients for developed and developing countries. Since government and foreign savings are exogenous, investment is savings driven. The last closure determines the external balance. We fix the foreign savings and therefore the trade balance, hence changes in trade flows result in shifts in the real exchange rate. 4.2.2. Para-tariff unilateral liberalization Over the last two decades, Sri Lanka introduced several additional taxes on imports, commonly referred to as “para-tariffs”. Ad hoc introduction of these taxes and frequent revisions to their rates make the import tax structure complex and unpredictable. Although custom duty rates have largely been kept intact and at a lower rate, the introduction of para-tariffs reversed previous efforts aimed to further simplify and reduce taxes on imports. While the number of para-tariffs has been reduced from four to two over the last decade, 23 the remaining para-tariffs affect a significant percentage of imports and, in some cases, grant high protection levels to selected products. Currently, the Ports and Airports Development Levy (PAL)24 and the Export Development Import Cess (EIC) 25 are the two para-tariffs active in Sri Lanka. A third of the revenue obtained from the Ports and Airports Development Levy will be lost due to exemptions. Since January 1, 2016, the PAL rate is 7.5 percent of the CIF value (5 percent from January 1, 2011 to December 31, 2015). In theory, the levy should be imposed on every article imported into Sri Lanka, but the law allows for several exemptions (e.g. goods used for manufacturing of exports, staple food items, etc.). The PAL Act also grants the Director-General of Customs and the Minister of Finance the power to partially or totally exempt any article from this levy. In practice, about a quarter of tariff lines benefit from reduced or zero rates in 2016 (1,138 tariff lines paying zero and 638 tariff lines paying 2.5% PAL). About a third of PAL revenue is lost due to exemptions. The Export Development Import Cess protects a wide range of sectors, particularly footwear, animal and agricultural products. The EIC was applied on 1,937 tariff lines (TLs) in 2016 (roughly 30% of TLs). Ad valorem rates were applied on 598 TLs, specific rates on 787 TLs, and mixed rates on 552 TLs. As with the PAL, there are several exemptions granted to either products or specific industries that reduce or eliminate Cess liabilities. About half of EIC revenue was lost due to exemptions in 2015. But unlike PAL, which has a low rate and is uniformly applied, EIC significantly increases the protection rate for products in some sectors. Most ad-valorem equivalents of EIC specific rates are in the double digits with a sizeable share of them above 50 percent - especially those applied on animal and agricultural products. The EIC is also applied to more than half of tariff lines in several sectors like footwear, vegetables, and apparel and textiles. To illustrate, Figure 8 shows the percentage of tariff lines under Cess in each import sector and the weighted average protection granted by the Cess in the y-axis. 23 Four para-tariffs were active in the 2000s: Import surcharge; Regional Infrastructure Development Levy (RIDL); Port and Airport Development Levy (PAL); and Import Cess. The Import surcharge and the RIDL were eliminated in 2010 and 2011, respectively. 24 PAL was imposed by the Ports and Airports Development Levy Act, No. 18 of 2011. 25 EIC was imposed by special orders under Section 14 (Import Cess) of the Sri Lanka Export Development Act No. 40 of 1979. 15 Figure 8: Prevalence and Average Protection Granted by Cess (2015) Source: Authors’ estimates The unilateral liberalization scenario in Sri Lanka phases out para-tariffs unilaterally over 10 years. We assume that the protection rates are the same as in 2016 data. We assume that PAL with the rate of 2.5 percent is removed in year 1 and PAL with the rate of 7.5 percent is removed in year 6 starting in 2018. In the case of the CESS we start by formulating a negative list of products that will remain protected over a longer-term horizon. The list has been compiled based on several criteria. The list of products excluded from para-tariff liberalization was compiled based on three product lists provided by The Ministry of Development Strategies (MoDSIT). These lists included products that were classified as sensitive in the industrial and agricultural sectors as well as products that generated the highest tariff and para-tariff revenues. We dropped products with a combined tariff, PAL, and CESS rate below 10 percent from the original lists to reach a combined list of 10 percent of tariff lines. The final list of excluded products contains 902 tariff lines coming from the original sensitive lists: 396 from industry, 371 from agriculture, and 226 from revenues. The CESS on non-sensitive products is assumed to be removed gradually over the years 1- 5, and gradually over the years 6-10 for the products on the sensitive list. Sri Lanka imposes very high duties on imports of agricultural and food products (Figure 9). The highest protection is assigned to imports of rice and milk products. For example, total duties (customs, CESS and PAL combined) on imports of rice from India or the Middle East amount to 100 percent or 93 percent respectively. The import duties in manufacturing are much lower with the highest rate recorded on imports of wearing apparel from China of 27 percent. The reform of CESS and PAL would reduce the duties on imports of all products with the highest absolute reductions recorded in milk, other food products and wearing apparel. Despite significant liberalization, several sectors such as rice, food or milk products remain highly protected. 16 Figure 9: Total import duties imposed by Sri Lanka (percent) in 2018 and 2028 Note: Trade weighted average of customs duties, CESS and PAL. Source: staff estimates based on customs data from 2016. 4.2.3. Free-Trade Agreement with China This section considers a potential FTA with China. The stylized FTA assumes that Sri Lanka retains a 10 percent negative list for customs duties but eliminates customs duties on the remaining 90 percent tariff lines over 20 years. Sri Lanka eliminates para-tariffs over the years 6-10 on most-favored-nation (MFN) basis. We assume that China eliminates its custom duties over 5 years starting in 2018. Our assumptions result in a significant reduction of Sri Lankan duties across all agricultural and manufacturing products with the exception of energy, which remains relatively protected (Figure 10 and Figure 11). Sri Lanka gains duty free access to the Chinese market with wearing apparel, chemicals, rubber and plastic as well other manufacturing products seeing the biggest absolute decline in duties paid. Key exports of Sri Lanka to China fall under the category of other manufacturing (dredgers HS890510 – 40 percent and uppers HS640610 – 7 percent of total exports), other crops (fermented black tea 9.5 percent), other agriculture (coconut 7 percent) and wearing apparel (brassieres 3 percent). 17 Figure 10: Duties imposed by Sri Lanka on imports from China before and after FTA (percent) Note: Trade weighted average of customs duties, CESS and PAL. Source: staff estimates based on customs data from 2016. Figure 11: Duties imposed by China on exports from Sri Lanka before and after FTA (percent) Note: Trade weighted average of customs duties, CESS and PAL. Source: staff estimates based on customs data from 2016. 4.2.4. Free-Trade Agreement with India Further, we consider an expansion of the current FTA with India. 26 We assume that Sri Lanka retains a 10 percent negative list for customs duties but eliminates them on the remaining 90 percent tariff lines over 10 years. Sri Lanka eliminates para-tariffs over the years 6-10 on MFN basis. We assume that India eliminates its custom duties over 5 years. Despite the fact that the FTA was not implemented in 2018, the 26 This analysis was done before the Sri Lanka – India FTA was signed in 2019. 18 analysis focuses on deviations from the baseline scenario, hence even if the FTA was implemented a couple of years later, the structure of the Sri Lankan economy is unlikely to change to the extent that the results would be impacted by the later implementation of this illustrative experiment. Our assumptions result in an asymmetric liberalization with Sri Lanka reducing its import duties significantly, but keeping them at still relatively high levels, while enjoying much better access to the Indian market (Figure 12 and Figure 13). The biggest absolute reduction of duties on Sri Lankan side is recorded in milk products, other foods, chemicals, other machinery and other manufacturing products. India liberalizes its access to Sri Lankan other crops (the key export under this category is tea), food (key exports are mineral waters), textiles as well as chemicals, rubber and plastics. However, exports under most of these categories other than tea are very small. Figure 12: Duties imposed by Sri Lanka on imports from India before and after FTA (percent) Note: Trade weighted average of customs duties, CESS and PAL. Source: staff estimates based on customs data from 2016. 19 Figure 13: Duties imposed by India on exports from Sri Lanka before and after FTA (percent) Note: Trade weighted average of customs duties, CESS and PAL. Source: staff estimates based on customs data from 2016. 4.2.5. Full reform scenario In a full reform scenario, we analyze the economic impacts of Sri Lanka i) reducing the para-tariffs; ii) implementing FTAs both with China and with India; and iii) full implementation of the WTO Trade Facilitation Agreement (TFA). In the FTA with China, Sri Lanka carries out a linear cut of custom duties in 90 percent of its product lines over a period of 20 years, starting in 2018, with the rest of its 10 percent being included in a negative list and therefore maintaining their initial custom duties. Sri Lanka also plans to eliminate its para-tariffs against China on MFN basis within a period between years 6 and 10. Relatively to the FTA with India, Sri Lanka will eliminate custom duties of 90 percent of products lines over 10 years, a shorter period than China FTA, and will also establish a 10 percent negative list for customs duties. Sri Lanka para-tariffs against China will also be eliminated on MFN basis between years 6 and 10. Both China and India, on the other hand, will eliminate their custom duties against Sri Lanka over a period of 5 years, without a negative list of protected products (Figure 14 and Table 2). Like the para-tariff unilateral liberalization scenario, we introduce a gradual reduction of iceberg trade costs due to streamlining of non- tariff measures (NTMs) in goods and services of 10 percent over 10 years. The full implementation of the TFA could reduce trade barriers by 17.4 percent over 10 years (Moïsé and Sorescu 2013). 20 Figure 14: Sri Lanka tariffs (percentage) in the full reform scenario with FTAs to China and India (difference between tariffs in 2028 and 2017) Sri Lanka tariffs on China Sri Lanka tariffs on India Vegetables and… Rice Other Crops Other Crops Other Agric Energy Other Agric Food Other Food Food Textiles Other Food Wearing Apparel Chemicals Rubber… Wearing Apparel Motor Vehicles Other Machinery Other Machinery Other… Other Manufacturing 0 50 100 150 0 20 40 60 2017 2028 2017 2028 Source: Staff estimates based on LINKAGE CGE Simulations Table 2: Sri Lanka tariffs reductions (in percentage) in the combined FTAs to China and India (difference between tariffs in 2028 and 2017) Item China India Item China India Rice -7 Other Food -5 -18 Vegetables and fruits -2 -2 Textiles -4 -4 Other Crops -1 -1 Wearing Apparel -27 -12 Fishing -3 Chem., Rubber, Plastic -8 -10 Other Agriculture -7 -80 Motor Vehicles -3 -2 Energy -7 -6 Other Machinery -5 -7 Food -7 -3 Other Manufacturing -8 -9 Milk Products -61 Source: Staff estimates based on LINKAGE simulations 5. Simulation results With the reduction of barriers to trade, all liberalization scenarios are expected to result in faster expansion of GDP and international trade compared to the baseline. Our results indicate that 10 years after liberalization, the volume of GDP would be expected to be around 2 percent higher than in the baseline (Table 3). Exports expand under all scenarios, even when only import duties are reduced with no better access to other markets under the unilateral liberalization scenario. This is due to the availability of cheaper imported inputs, but also because of our assumption of fixed trade balance (as a share of GDP) that leads to adjustments in real exchange rate. When import duties decline and imports increase, the real exchange rate depreciates, and exports expand to keep the trade balance a fixed share of GDP. Overall, the 21 higher the degree of own liberalization and improved market access the higher the gains in GDP, exports and imports, but also the higher the sectoral reallocation and adjustment costs. Table 3: Key macro implications of simulated scenarios, percentage deviations from the baseline Para-tariff China FTA India FTA Full reform liberalization 2023 2028 2023 2028 2023 2028 2023 2028 GDP 0.77 1.97 0.83 1.37 0.94 2.18 1.37 2.81 Exports 7.6 18.3 3.3 7.52 3.8 19.3 7.52 24.23 Imports 5.4 9.8 4.2 11.11 5.1 11.6 11.11 18.86 Source: Staff estimates based on LINKAGE CGE simulations. The sectoral impacts of scenarios are driven by the relative liberalization of imports and exports and initial trade shares with integrating countries. Sectors where products are exposed to tougher competition as a result of the reduction of imports duties, might be crowded out by imports. At the same time if they constitute important inputs to domestic production of the same or other sectors, the domestic output of sectors which use those inputs intensively might expand. With better access to foreign markets certain domestic products become more competitive abroad and enjoy higher sales on foreign markets. Under the unilateral para-tariffs liberalization scenario the key sectors that benefit from the lower import duties are wearing apparel and textiles, services, trade and transport. For some these sectors, demand is being stimulated either by their use in trade (trade and transport) or by higher spending power of consumers due to faster growth of income. As resources are attracted into sectors which become more competitive, selected sectors are bound to see their output declining due to finite availability of capital and labor. A few sectors experience small declines of domestic output, while they are being replaced by more competitive imports. Those negatively affected include other agricultural products, milk and food products. Under unilateral liberalization scenario exports of wearing apparel expand mostly to the United States and rest of the world, while additional exports of machinery and other manufacturing products are mostly directed to India, rest of Asia, the United States and rest of the world (Figure 15). Under the FTA with China the results are also largely driven by Sri Lanka’s unilateral liberalization of para-tariffs with the decline of the same sensitive sectors as in the previous scenarios (Figure 11). In this scenario, the main sectors experiencing expansion of output include textiles and wearing apparel, whose output gains are driven solely by exports to China (Figure 12). Contrary to previous scenarios sectors experiencing gains in output and exports to China include also other crops, fishing products, chemicals, rubber and plastic products and other manufacturing goods. Under the FTA with India the output of wearing apparel and textiles as well as other crops expand the most (Figure 9). The expansion of those sectors is mostly driven by a boost to exports to India, with exports of other machinery and other manufacturing goods also expand to the rest of the world and the rest of Asia (Figure 10). Similarly, to unilateral scenario, several sectors would be expected to see their output decrease slightly relative to the baseline 2028 level (their level is still above their current values). Sectors experiencing negative impacts include vegetables and fruits, other agricultural goods, food and milk products. Those results are mostly driven by the unilateral liberalization of para-tariffs in those sectors and not by specific commitments in their FTA with India. In the full reform scenario, the results are more pronounced, with an increase of exports of 24.2 percent relative to the baseline, compared with 18.3 percent under para-tariff liberalization alone. Sectors like wearing apparel, which has one of the highest increases of production, are benefiting from cheaper imported inputs derived by an increase of imports due to lower tariffs, and therefore becoming 22 more competitive. Trade, transportation, services and construction see even higher increases in terms of output compared to the unilateral liberalization scenario, driven by higher welfare gains obtained from the FTAs, and as well the extra stimulus derived from Sri Lanka higher volumes of trade. Due to the redistribution of resources away from less competitive sectors, such as food, milk products and agriculture, the results show that Sri Lanka’s economy is becoming more industrialized and more services oriented (Figure 16). Thanks to the FTAs, trade creation is particularly evident in the sectors of other crops, energy and other manufacturing for India, and textiles and chemicals, rubber and plastics for China (Figure 10). In the following section, only the para-tariff and the full reform scenario will be analyzed. Given the complementarities in the simulation results, the discussion about the scenarios of China and India FTAs are not discussed in the following section, but they can be requested from the authors. Figure 15: Exports by Sector and Destination under Unilateral Liberalization scenario – deviations from the baseline in million USD (2028) Source: Staff estimates based on LINKAGE CGE Simulations 23 Figure 16: Exports by Sector and Destination under full reform FTA – deviations from the baseline in million USD (2028) Source: Staff estimates based on LINKAGE CGE Simulations 5.1. Labor market effects by sector In line with historical trends, the baseline contemplates a reduction in the agricultural sector and increases in manufacturing and services. In the baseline, agriculture and related subsectors across the board are expected to shrink significantly and reduce the number of workers employed as well as workers move to sectors with higher productivity. The expected cumulative decline over the 12 years period is between 30 and 35 percent for agriculture and fruits and vegetables subsectors, respectively; while mining is expected to decline its share of employment by 28 percent. In fact, employment in agriculture as a percentage of total employment in Sri Lanka declined from 40.6 percent in 2000 to 27 percent in 2016 (International Labor Organization, n.d.). Industrial and manufacturing sectors and related subsectors like transportation equipment, machinery equipment, chemicals and other manufacturing are expected to expand the most in the baseline. The expected cumulative growth for the period in question is between 12 and 21 percent depending on the subsector (Figure 17). Services related sectors are also expected to marginally grow - financial and business-oriented services, trade and transport and social services and others expand in the baseline between 1 and 5 percent. It is important to note that our simulations contemplate the same level of unemployment across years and the size of the working age population grows according to medium-variant UN population projections. Due to expected change in population structure, the pool of employment in Sri Lanka is projected to decline, marginally by 100,000 workers; mainly because of a decline in the working age population (ages 15-64). The following paragraphs cover changes with respect to baseline conditions. 24 Figure 17: Aggregate cumulative labor demand impacts (2018-2030) in the baseline scenario by sectors Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016 Note: Excluded sector: communications and electronic equipment. Approximately 98% of sectors/workers represented in HIES 2016. Bubbles represent total estimated workers in each sector. Para-tariff liberalization scenario and full scenario switches jobs, out of agriculture and towards textile and wearing apparel sectors (as compared to the baseline). Agriculture subsectors and mining are expected to further decline. Agriculture and vegetables are large employers of women. Textiles and wearing apparel sectors of manufacturing are expected to benefit the most due to para-tariff liberalization and expand by 12 percent and 11 percent, respectively. Similarly, these sectors are expected to expand by 14 percent and 12 percent each in the full reform scenario, marginally larger than the para-tariff liberalization scenario (Figure 18). Social services sector expands modestly in both scenarios (Figure 18). In both scenarios, wearing apparel sector is projected to observe a net increase in employment in Sri Lanka, vis-à-vis a decline under baseline conditions. As a result, the trade policy scenarios can bring a more robust value chain integration between local producers of i) textile and ii) wearing apparel. Skilled workers will benefit relatively more from the trade liberalization scenarios. According to simulation results, the wages of skilled workers are expected to grow at as slightly faster rate in all scenarios and skill premia increases by 10.1 percent in the baseline, and a further 0.5 percent in the unilateral liberalization scenario and 2.1 percent in the case of the full reform scenario. A faster increase in skilled workers wages is driven by expansion of skilled-labor intensive services and manufacturing sectors such as machinery or other manufacturing products. It is important to highlight that the pace of the sectoral changes in employment is rather nuanced in the short-term versus the long-term, as can be observed by the time trends of employment change over time in Figure 20 to Figure 22. The speed of adjustment is attributed to the phase of tariff elimination and implementation of trade protocols, which highlights the importance of allowing some period of adjustment for labor mobility and sectoral reallocation of resources. In this regard, agriculture and other related sectors (except food and milk production), in the baseline, 25 shrink slower in the near-term horizon (5 years after implementation), and shrink faster in the long-term horizon from (6 to 10 years after implementation). Contrarily, manufacturing and services related sectors expand faster in the near-term and slower in the long-term (Figure 20). Figure 18: Aggregate cumulative labor demand effects by sectors in 2028 by sectors in the para-tariffs liberalization scenario, w.r.t. baseline Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016 Note: Excluded sector: communications and electronic equipment. Approximately 98% of sectors/workers represented in HIES 2016. Bubbles represent total estimated workers in each sector. 26 Figure 19: Aggregate cumulative labor demand effects by sectors in 2028 by sectors in the full scenario, w.r.t. baseline Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016 Note: Excluded sector: communications and electronic equipment. Approximately 98% of sectors/workers represented in HIES 2016. Bubbles represent total estimated workers in each sector. 27 Figure 20: Aggregate cumulative labor demand percentage changes by sectors in the baseline scenario by 2023 and 2028 Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016 Note: Excluded sector: communications and electronic equipment. Approximately 98% of sectors/workers represented in HIES 2016. Bubbles represent total estimated workers in each sector. 28 Figure 21: Aggregate cumulative labor demand percentage changes by sectors in the para-tariffs liberalization scenario, w.r.t. baseline in 2023 and 2028 Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016 Note: Excluded sector: communications and electronic equipment. Approximately 98% of sectors/workers represented in HIES 2016. Bubbles represent total estimated workers in each sector. Figure 22: Aggregate cumulative labor demand percentage changes by sectors in the full scenario, w.r.t. baseline in 2023 and 2028 Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016 Note: Excluded sector: communications and electronic equipment. Approximately 98% of sectors/workers represented in HIES 2016. Bubbles represent total estimated workers in each sector. 29 5.2. Sub-national labor market effects In the baseline scenario, employment in agriculture and related sectors is shrinking over time across all districts in Sri Lanka. The job losses range from 15 percent to 30 percent in different districts. However, since some districts are predominantly agriculture dependent, the job losses range from approximately 2,000 jobs (Jaffna, Mannar, Vavuniya, Mullaitivu, and Kilinochchi) to approximately over 20,000 jobs (Nuwara-eliya, Kurunegala, Badulla, and Ratnapura). Manufacturing and services related sectors are generally expected to grow and benefit already industrialized and wealthier districts – between 5,000 and 25,000 jobs approximately – in districts like Colombo, Gampaha, Kalutara, Kandy, Galle, Kurunegala, and Kegalle. These districts are also expected to attract and absorb skilled workers from agricultural districts which experience job losses (See Table A2 in Annex). Some districts can expect protracted effects reflecting a large internal migration towards more urbanized metropolitan areas. For example, in the district of Nuwara-eliya job losses under baseline conditions are expected across all sectors. On the other hand, Kurunegala district is estimated to experience job losses in agriculture, but job creation across almost all other sectors in the long-term horizon. These differentiated results by sectors also aggregate to net job losses and gains respectively in each district. This is something to consider when designing policy reforms – similar to a stock and flow concept - some sectors may perform significantly better (worse), but the size of the sector employment and the associated changes in labor demand is crucial to understand the absolute numbers of job losses (gains) in each district. Relatedly, the net change in employment by district or by sector are other metrics to consider for a policy maker (See Table A2 in Annex). The sectoral reconfiguration observed in para-tariffs unilateral liberalization and the full-reform scenario, especially in wearing apparel, leads to larger concentration in urban areas. In the full reform scenario, a similar trend in employment changes is estimated across all districts as in the baseline scenario. Textiles, wearing apparel, and trade and transportation sectors expand with respect to baseline, from 9 to 17 percent. Concurrently, the higher wages in textiles and wearing apparel would attract workers from chemicals and some manufacturing sectors (in Annex). Most employment gains are projected in the western regions of Colombo, Gampaha and Kalutara with a net increase in employment of 111,000 from 2018 to 2028 – amid the rest of the country experiencing a net employment decline of 216,000 jobs. Implementation of full-trade reform accelerates the concentration of employment in the western regions, going from 111,000 to 134,000, or 20 percent increase in net employment gains by 2028. 30 Figure 23: Labor demand changes, by district in the baseline scenario Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016 Note: Green areas experience the highest labor demand, and pink areas experience negative labor demand. Excluded sector: communications and electronic equipment. Approximately 98% of sectors/workers represented in HIES 2016. Bubbles represent total estimated workers in each sector. Figure 24: Labor demand changes, by districts in the full scenario and w.r.t. baseline Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016 Note: Green areas experience the highest labor demand, and pink areas experience negative labor demand. Excluded sector: communications and electronic equipment. Approximately 98% of sectors/workers represented in HIES 2016. Bubbles represent total estimated workers in each sector. 31 Figure 25 Labor demand changes, by districts in the para-tariffs scenario and w.r.t. baseline Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016 Note: Green areas experience the highest labor demand, and pink areas experience negative labor demand. Excluded sector: communications and electronic equipment. Approximately 98% of sectors/workers represented in HIES 2016. Bubbles represent total estimated workers in each sector. 5.3. Poverty and distribution In 2018, approximately 8.6 million people, or 40.4 percent of the population, lived with less than PPP$5.50/day. Typically, the World Bank uses the PPP$1.90/day for the measurement of global extreme poverty, but it recognizes that this global floor might not be appropriate to measure the evolution of poverty for countries at higher levels of development. Additional poverty lines of PPP$3.20 and PPP$5.50/day have been used for lower-middle and upper-middle income countries, respectively. For the case of Sri Lanka, an upper middle-income country, this paper tracks the evolution of poverty using the PPP$5.50/day poverty line. Poverty measured in this way has steadily declined since the mid-1980s. Based on household survey data available in World Bank’s PovcalNet, more than 80 percent of Sri Lankans lived with less than PPP$5.50/day in 1985. By 2016, this figure has been reduced in half. The baseline projects poverty at PPP$5.50/day further declining to 13.4% by 2028. Under baseline conditions and without the effect of policy, the Sri Lankan economy will grow steadily above past historical trends in line with expected recovery from the civil war period. Per capita household consumption would grow 4.9% on annual basis from 2016-2030, compared to an increase of 2.0 percent from 1985 to 2009 (civil war period), and a more recent 4.0 percent from 2009-2016. Under these assumptions and by 2028, the poverty headcount ratio (%) at PPP$5.50/day will be 13.4 percent (down from 40.4% in 201627), which is equivalent to 2.9 million considering a projected population of 21.4 million. Figure 26 shows that this difference is larger in 2022, or 5 years after implementation, when the poverty headcount ratio would be 1.4 percentage points lower than baseline. Poverty reduction in the China and India FTAs, and the para- tariff liberalization are smaller, between 0.37 and 0.31 percentage points by 2028. Limited skill formation and in the absence of redistributive policy, the economic gains of trade reform are regressive. The growth incidence curves shown in Figure 27 depict the net economic gains of each trade 27 Latest estimate based on World Bank’s PovcalNet (http://iresearch.worldbank.org/PovcalNet/). 32 reform scenario, with respect to baseline and by 2028. Accelerated growth spur by trade reform will cause an increase in demand for skilled labor. An increase in economic activity, considering an exogenous supply of skilled and unskilled labor causes skilled wage premia to grow. In the CGE-GIDD framework (Figure 1), it is implicitly assumed that trade reform does not alter the creation (supply) of skilled labor. From 2016 to 2028, baseline conditions contemplate a wage premium increase of 10.2 percentage points, from 5.6 percent in 2016 to 15.9 by 2028. In the case of the full reform scenario, wage premium would further increase in 1.1 percentage points. Figure 26 Projected evolution of poverty in Sri Figure 27 Distributional impact of trade reform, Lanka 2016-2028, baseline by scenario and w.r.t baseline Source: Staff estimates based on LINKAGE CGE Simulations Source: Staff estimates based on LINKAGE CGE and HIES 2016 Simulations and HIES 2016 High poverty incidence districts register the highest decline in poverty rates, particularly in the full- reform scenario. As shown in Figure 28, poverty rates at the district level suggests that except for urbanized districts around the capital (Colombo, Gampaha, and Kalutara), essentially rural districts experience high incidence of poverty. As noticed in the previous section on labor effects, these same districts are also predominantly agricultural. All districts are estimated to reduce poverty levels in the baseline between 2018 and 2028 – from 6 percentage points in urban districts with low-initial incidence of poverty to roughly 30 percentage points in districts with high-initial incidence, i.e. Moneragala and Ratnapura. In the full-reform scenario, most districts experience declines in the poverty headcount ratio (at PPP$5.50/day) - with respect to the baseline and by 2028, but high poverty incidence districts register the highest decline in poverty rates. Only Vavuniya and Puttalam experience marginal positive changes. Irrespectively, the full reform scenario yields the largest results in poverty reduction, as shown before, due to much larger labor reallocation, increases in wages, and economic growth. 33 Figure 28 Evolution of poverty headcount ratio (PPP$5.50/day), by district Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016 5.4. Caveats This modeling framework allows to incorporate the complex interactions but lacks a response on cumulative effects due to absorption of technology and foreign direct investment (FDI). The applied multi- regional dynamic CGE model accounts simultaneously for interactions among producers, households and governments in multiple product markets and across several countries and regions of the world. Although incorporating well-developed dynamic features such as accumulation of capital through changes in savings and investment, the model, however, lacks positive dynamic feedback loops concerning the accumulation of knowledge and the absorption of foreign technology through FDI, it also does not allow for modeling of extensive margins in exports. Since Sri Lanka is likely to become more competitive following implementation of trade reforms and FTAs it might be able to access new markets or start exporting new products, attracting new FDI. Therefore, the gains illustrated here may underestimate the eventual impact and represent the lower bound of potential benefits of trade reforms. In the short-term, the results might be overestimated due to the assumption of a fixed trade balance. In fact, imports might increase faster than exports, so there might be a transitional trade balance deterioration. However, in the medium-term, this could mean an underestimation of gains, if productivity improves and leads to positive changes in the trade balance. Further, we assume that the rate of unemployment remains at the initial level and all new entrants to the labor market find gainful employment. This could be short-term, and again, there may be a more positive change in the longer-term. 34 Limitations associated with the use of microdata and the reconciliation with macroeconomic statistics should be considered. Nationally representative household surveys are incorporated in the CGE modeling framework to provide information related to the contribution of labor to value added, disaggregated by sector and type of worker. To incorporate this information, one must reconcile macro and microdata sources. This reconciliation must deal with the fact that (1) the aggregates obtained from microeconomic data do not add up to the aggregate statistics in national statistics; and (2) microeconomic data may not provide accurate information about some very small sectors. 6. Concluding remarks This paper discussed a simulation framework used for the ex-ante assessment of the distributional aspects of trade, at the sectoral and sub-national district levels. In this regard, the assessment of the distributional aspects needs to operate within a systematic framework. The CGE-GIDD is a macro and microeconomic model linked through a set of linked aggregate variables, including labor supply, skill formation, and real earnings. The model operates in a top-down fashion. The CGE model relies heavily on economic theory and provides aggregate and complex interactions within consistent market-clearing mechanisms. The microsimulation can project the behavior of consumers and workers at a very disaggregated level. The CGE-GIDD allows us to track patterns from policy shocks, to macro and sectoral outputs, and finally to disaggregated effects on labor at a sub-national sectoral level. On a technical level, the labor market specification is a key element that drives the distributional results. In such circumstances, the assumptions used to model and calibrate the labor market and the complementarities between workers and capital have a great influence on the results. A set of relevant trade policy simulations were assessed for the case of Sri Lanka. An ex-ante evaluation requires the creation of a counterfactual simulation that serves as a comparison. The baseline follows historical trends and is key in understanding the net effect of the alternative policy scenarios. The alternative scenarios contemplate i. a unilateral para-tariff liberalization, ii. FTA with China, iii. FTA with India, and lastly, iv. a full-reform scenario that incorporates the previous three elements plus implementation of the WTO Trade Facilitation Agreement. In practice, the level of protection with tariffs is small for open economies and most of the effective gains from trade would come from reduction in NTMs and extra forms of taxation, such as para-tariffs. Macro simulation results show strong complimentary between the policy choices. As a result, and for brevity, only the para-tariff unilateral liberalization and the full-reform scenario were discussed in the section dealing with the distributional assessment. Reductions of trade barriers result in faster expansion of GDP and trade, accompanied by greater economic activity in urban areas, shifts toward export competitive sectors and increases in wage inequality. With respect to baseline and by 2028, implementation of full trade reform would help in reducing poverty (PPP$5.50/day) 0.66 additional percentage points. Our analysis shows that despite these positive effects on poverty reduction and in the absence of additional policies, the economic gains would be in favor of skilled workers – which are located in the top 40 percent of the income distribution. In the realm of scenario analysis, it is important to highlight that these results are not forecasts; rather, these results can be best understood as mechanical responses only occurring when the economy operates under a narrow set of assumptions. Policy creates a new set of incentives for economic actors and the framework presented here can assess the magnitude and direction of those incentives within a complex, systematic, and detailed framework. The distributional gains from trade have long been recognized as an important component for economic theory and subsequently for trade policy design and implementation. Recent advances in computational power, better standards for collecting macro and micro data and increases in the technical capacity of governmental bodies (particularly national statistical offices, and finance ministries) increase the accessibility these methodologies in developing countries. This paper aims to demonstrate how this 35 method can be leveraged in the context of a small, open economy. There are a set of possible extensions that have been implemented in other contexts, particularly promising is incorporating a higher level of granularity in the “down” part of the framework relying on firm-level, census, or new databases that use high-frequency data. Policy implications should aim to distribute the gains with more equity. By understanding the varied impacts across regions and sectors, policy makers can design policies to mitigate negative impacts on selected groups of workers and facilitate the transition of workers across regions and sectors. In doing so, the gains from trade could be distributed more equally. Possible specific policies might include: • Skills for migrant workers: Increasing economic activity in urban areas is expected with a major reallocation of labor from districts with large agriculture and vegetables/fruit sectors into urban areas that host emerging sectors (chemicals, textile and apparel, and other manufacturing). Policies that promote the acquisition of job-related skills will benefit both workers and firms. While workers’ technical knowledge and expertise are critically important, the emphasis on managerial and administrative capacities should be underscored, particularly for firms engaging in exporting activities. • Better mobility and communication. Some non-tradable sectors, such as social services, construction, trade and transport sectors which are essentially in highly urbanized districts experience positive effects on the extensive and intensive margins of labor demand. Still, geographic labor mobility in Sri Lanka is very limited. Infrastructure and communication investment policies can help to reduce the gap between the less connected, isolated and job loss prone locations with the expanding urbanized districts. Domestic labor mobility could also be promoted with security and socio-political integration, which is crucial in the aftermath of a prolonged civil war. • Enforce progressive labor standards and policies that protect vulnerable workers. Textiles and wearing apparel experience the highest rate of labor demand expansion with large increases in employment of women. Policies that prevent the exploitation of vulnerable workers, particularly migrant unskilled women, can help to guarantee the long-term viability of the expanding sectors and expanding urban communities. Policies that protect the safety in the workplace and that promote that workers obtain a fair share from their labor guarantee the long-term sustainability of the sector, and the economic viability of the new urban areas that they will populate. 36 References Ahmed, S. Amer, Maurizio Bussolo, Marcio Cruz, Delfin Go, and Israel Osorio-Rodate. 2020. “Global Inequality in a More Educated World.” Journal of Economic Inequality. Ahmed, S. Amer, Marcio Cruz, Delfin S. Go, Maryla Maliszewska, and Israel Osorio-Rodarte. 2016. “How Significant Is Sub-Saharan Africa’s Demographic Dividend for Its Future Growth and Poverty Reduction?” Review of Development Economics, May. https://doi.org/10.1111/rode.12227. Baldwin, Robert E. 1989. “The Political Economy of Trade Policy.” Journal of Economic Perspectives 3 (4): 119–35. https://doi.org/10.1257/jep.3.4.119. Balistreri, E.J., M. Maliszewska, I. Osorio-Rodarte, D.G. Tarr, and H. Yonezawa. 2018. “Poverty, Welfare and Income Distribution Implications of Reducing Trade Costs through Deep Integration in Eastern and Southern Africa.” Journal of African Economies 27 (2). https://doi.org/10.1093/jae/ejx024. Bosworth, Barry P., and Susan M. Collins. 2003. “The Empirics of Growth: An Update.” 2003(2). Brookin. Pap. Econ. Act. Bourguignon, François, and Maurizio Bussolo. 2013. “Income Distribution in Computable General Equilibrium Modeling.” In , 1:1383–1437. Handbook of Computable General Equilibrium Modeling. Elsevier. http://ideas.repec.org/h/eee/hacchp/v1y2013icp1383-1437.html. Bourguignon, François, Maurizio Bussolo, and Luiz A Pereira da Silva. 2008. The Impact of Macroeconomic Policies on Poverty and Income Distribution : Macro-Micro Evaluation Techniques and Tools. Houndmills, Basingstoke, Hampshire ; New York Washington, DC: Palgrave Macmillan ; World Bank. http://www.loc.gov/catdir/toc/ecip081/2007040478.html. Bourguignon, François, Francisco H.G. Ferreira, and Phillippe G. Leite. 2008. “Beyond Oaxaca-Blinder: Accounting for Differences in Household Income Distributions.” Journal of Economic Inequality 6: 117–48. https://doi.org/10.1007/s10888-007-9063-y. Bourguignon, François, Fransisco H.G. Ferreira, and Nora Lustig. 2005. The Microeconomics of Income Distribution Dynamics in East Asia and Latin America. Washington D.C.: World Bank. http://documents.worldbank.org/curated/en/984291468049815108/The-microeconomics-of-income- distribution-dynamics-in-East-Asia-and-Latin-America. Bussolo, Maurizio, Rafael E De Hoyos, and Denis Medvedev. 2010. “Economic Growth and Income Distribution: Linking Macro-Economic Models with Household Survey Data at the Global Level.” International Journal of Microsimulation 3 (1): 92–103. http://ideas.repec.org/a/ijm/journl/v3y2010i1p92-103.html. Cassing, James H., and Arye L. Hillman. 1985. “Political Influence Motives and the Choice between Tariffs and Quotas.” Journal of International Economics 19 (3–4): 279–90. https://doi.org/10.1016/0022- 1996(85)90036-4. Davies, James B. 2009. “Combining Microsimulation with CGE and Macro Modelling for Distributional Analysis in Developing and Transition Countries.” International Journal of Microsimulation 2 (1): 49–65. Dellink, Rob, Jean Chateau, Elisa Lanzi, and Bertrand Magné. 2017. “Long-Term Economic Growth Projections in the Shared Socioeconomic Pathways.” Global Environmental Change 42 (January): 200–214. https://doi.org/10.1016/J.GLOENVCHA.2015.06.004. Devarajan, Shantayanan, Delfin S. Go, and Hongyi Li. 1999. “Quantifying the Fiscal Effects of Trade Reform.” Policy Research Working Papers. World Bank. https://doi-org.libproxy-wb.imf.org/10.1596/1813-9450-2162. Devarajan, Shantayanan, Delfin S. Go, Maryla Maliszewska, Israel Osorio Rodarte, Hans Timmer, Israel Osorio- Rodarte, and Hans Timmer. 2015. “Stress-Testing Africa’s Recent Growth and Poverty Performance.” Journal of Policy Modeling, Policy Research Working Paper, 37 (4): 521–47. https://doi.org/10.1016/j.jpolmod.2015.04.006. 37 Grandville, Olivier de La. 2017. Economic Growth : A Unified Approach. Edited by Robert M Solow. Second edi. Cambridge, United Kingdom : Cambridge University Press. Hillberry, Russell, and David Hummels. 2013. “Trade Elasticity Parameters for a Computable General Equilibrium Model.” In Handbook of Computable General Equilibrium Modeling. https://doi.org/10.1016/B978-0-444- 59568-3.00018-3. Hillman, Arye L. 1982. “Declining Industries and Political-Support Protectionist Motives.” The American Economic Review 72 (5): 1180–87. http://www.jstor.org/stable/1812033. International Labor Organization. n.d. “ILOSTAT.” https://ilostat.ilo.org/. International Trade Centre. 2016. “Market Access Map (MAcMap) Tariff Rates for 2016–2046 between TPP Member Countries under the TPP Agreement.” Kokas, Deeksha, and Jakob Engel. 2020 “Distributional Impact of Trade: A Survey of Theory, Evidence and Policy.” Washington, D.C.: The World Bank. León-Ledesma, Miguel A., Peter McAdam, and Alpo Willman. 2010. “Identifying the Elasticity of Substitution with Biased Technical Change.” American Economic Review. https://doi.org/10.1257/aer.100.4.1330. Loayza, Norman, Klaus Schmidt-Hebbel, and Luis Servén. 2000. “What Drives Private Saving across the World?” The Review of Economics and Statistics 83 (2): 165–81. https://doi.org/10.1162/003465300558678. Martin, Will, and Devashish Mitra. 1999. “Productivity Growth and Convergence in Agriculture and Manufacturing.” WPS 2172. Policy Research Working Paper. Washington, DC. http://documents.worldbank.org/curated/en/921921468765622994/Productivity-growth-and- convergence-in-agriculture-and-manufacturing. Mensbrugghe, Dominique van der. 2013. “LINKAGE Technical Reference Document Version 7.1.” http://siteresources.worldbank.org/INTPROSPECTS/Resources/334934- 1314986341738/TechRef7.1_01Mar2011.pdf. ———. 2020. “The Environmental Impact and Sustainability Applied General Equilibrium (ENVISAGE) Model, Version 10.1.” TP/19/01. GTAP Technical Paper. Moïsé, Evdokia, and Silvia Sorescu. 2013. “Trade Facilitation Indicators: The Potential Impact of Trade Facilitation on Developing Countries’ Trade.” No. 144. OECD Trade Policy Papers. Paris. https://doi.org/10.1787/5k4bw6kg6ws2-en. Olivieri, Sergio, Sergiy Radyakin, Stanislav Kolenikov, Michael Lokshin, Ambar Narayan, and Carolina Sanchez- Paramo. 2014. Simulating Distributional Impacts of Macro-Dynamics: Theory and Practical Applications. The World Bank. https://doi.org/10.1596/978-1-4648-0384-0. Orcutt, Guy H. 1957. “A New Type of Socio-Economic System.” The Review of Economics and Statistics 39 (2): 116. https://doi.org/10.2307/1928528. Reardon, Thomas, Kostas Stamoulis, and Prabhu Pingali. 2007. “Rural Nonfarm Employment in Developing Countries in an Era of Globalization.” In Agricultural Economics. https://doi.org/10.1111/j.1574- 0862.2007.00243.x. Rodrik, Dani. 1995. “Political Economy of Trade Policy.” In Handbook of International Economics, 3:1457–94. Elsevier. https://doi.org/10.1016/S1573-4404(05)80008-5. Solow, Robert M. 1956. “A Contribution to the Theory of Economic Growth.” The Quarterly Journal of Economics 70 (1): 65. https://doi.org/10.2307/1884513. UN DESA. 2019. World Population Prospects 2019. United Nations. Department of Economic and Social Affairs. World Population Prospects 2019. http://www.ncbi.nlm.nih.gov/pubmed/12283219. Winters, L. Alan, Neil McCulloch, and Andrew McKay. 2004. “Trade Liberalization and Poverty: The Evidence so 38 Far.” Journal of Economic Literature 42 (1): 72–115. https://doi.org/10.1257/002205104773558056. Wittenberg, Martin. 2010. “An Introduction to Maximum Entropy and Minimum Cross-Entropy Estimation Using Sta.” The Stata Journal 10 (3): 315–30. World Bank. n.d. “The African Continental Free Trade Area: Economic and Distributional Effects.” Washington, D.C. ———. n.d. “The African Continental Free Trade Area: Economic and Distributional Effects.” Washington D.C. ———. n.d. “World Development Indicators.” Washington, D.C. 39 Annex 1: Additional Simulation Results Figure A29: Impact on output, exports, and imports of para-tariff liberalization scenario, % change w.r.t. baseline in 2028 Source: Staff estimates based on LINKAGE CGE Simulations Figure A30: Impact on output, exports and imports of FTA with India scenario, % change w.r.t. baseline in 2028 Source: Staff estimates based on LINKAGE CGE Simulations 40 Figure A31: Impact on output, exports and imports of FTA with China scenario, % change w.r.t. baseline in 2028 Source: Staff estimates based on LINKAGE CGE Simulations 41 Table A1: Distribution of workers by sector and gender, percent Description Total Males Females Description Total Males Females Paddy rice 23.26 20.89 27.82 Motor vehicles and parts 0.07 0.09 0.02 Wheat 0.07 0.09 0.03 Transport Cereal grains 4.48 3.33 6.69 equipment 0.06 0.08 0.02 Fishing 1.58 2.30 0.20 Electronic Minerals 0.67 0.95 0.12 equipment 0.02 0.03 0.00 Textiles 1.31 0.58 2.71 Machinery and equipment 0.37 0.39 0.34 Wearing apparel 5.46 2.54 11.07 Manufactures 1.91 2.42 0.94 Leather Electricity 0.25 0.33 0.10 products 0.22 0.14 0.37 Gas manufacture Wood products 0.99 1.48 0.06 distribution 0.02 0.03 0.00 Paper products Water 0.21 0.24 0.15 publishing 0.64 0.63 0.66 Construction 7.92 11.62 0.80 Petroleum coal Trade 16.68 17.74 14.65 products 0.05 0.07 0.00 Transport 6.71 9.79 0.78 Chemical rubber plastic Sea transport 0.01 0.02 0.00 products 0.86 0.83 0.93 Air transport 0.01 0.00 0.03 Mineral Communication 0.14 0.14 0.14 products 1.14 1.36 0.71 Financial services 1.69 1.44 2.16 Ferrous metals 0.10 0.15 0.00 Insurance 0.34 0.33 0.37 Metals 0.00 0.00 0.00 Business services 2.48 2.74 1.99 Metal products 0.54 0.73 0.17 Recreation and other services 5.89 6.05 5.59 Social services 13.86 10.47 20.38 Source: Staff estimates based on HIES 2016 Note: CGE classification 42 Figure A32: District-level concentration of agricultural jobs Source: Staff estimates based on HIES 2016 Figure A33: District-level concentration of manufacturing jobs Source: Staff estimates based on HIES 2016 43 Figure A34: District-level concentration of textiles and wearing apparel jobs Source: Staff estimates based on HIES 2016 Figure A35: District-level concentration of trade, transport, finance and social services jobs Source: Staff estimates based on HIES 2016 44 Table A2: Expected change in number of workers in Sri Lanka disaggregated by districts and sectors – baseline scenario Districts Agri Mining Textil WearAp Chemic TranspEq Machin OthManu Utiliti Construc Trade&Tr Finance SocialServ VegFruit Total Colombo (4,215) (227) 382 236 2,416 161 555 5,766 105 1,851 12,839 2,605 12,156 (205) 34,424 Gampaha (6,689) (334) 1,160 5,213 4,765 961 760 6,705 508 7,064 24,223 2,749 17,783 (290) 64,579 Kalutara (8,931) (332) 440 716 1,487 231 401 2,961 39 1,694 7,080 924 6,313 (313) 12,709 Kandy (12,058) (303) 176 108 650 37 90 2,038 140 1,676 5,691 647 5,064 (601) 3,353 Matale (7,774) (198) 34 (259) 237 23 35 249 7 53 507 71 658 (1,572) (7,930) Nuwara-eliya (24,867) (93) (159) (867) (70) - - (100) (135) (1,238) (3,596) (139) (2,349) (14,819) (48,431) Galle (13,245) - 372 703 690 45 197 2,236 39 2,110 5,749 578 5,508 (169) 4,814 Matara (14,969) - 107 (53) 328 75 215 935 78 639 2,119 318 2,304 (165) (8,069) Hambantota (10,550) (318) 62 (180) 729 - 44 1,163 20 179 885 153 1,103 (1,599) (8,310) Jaffna (5,003) (64) (26) (10) 42 - 7 565 (20) (533) (469) 35 (266) (1,780) (7,523) Mannar (2,105) (14) (32) (6) (2) - - 2 (9) (243) (499) (35) (313) (75) (3,331) Vavuniya (2,816) (22) (140) - 2 - - 19 (33) (349) (684) (53) (467) (802) (5,347) Mullaitivu (2,714) (28) (64) (19) (0) - - 13 - (193) (250) (15) (262) (223) (3,755) Kilinochchi (1,365) (64) (74) (76) (1) - - 2 (12) (328) (390) (42) (412) (204) (2,967) Batticaloa (7,464) (244) (448) (79) (7) - - 39 - (1,191) (1,941) (97) (1,722) (307) (13,461) Ampara (10,220) (267) (235) (165) 36 10 2 204 (49) (841) (1,568) (33) (1,186) (209) (14,522) Tricomalee (5,272) (155) (78) (173) 1 - - 50 (25) (950) (1,468) (31) (1,181) (235) (9,518) Kurunegala (19,803) (310) 839 794 2,376 157 250 3,670 93 2,590 8,939 1,080 9,412 (879) 9,206 Puttlam (12,702) (457) (111) (633) 271 - 15 567 (52) (695) (1,564) (13) (655) (3,248) (19,276) Anuradhapura (16,294) (37) 29 (142) 461 - 23 510 22 248 1,531 275 2,248 (626) (11,752) Polonnaruwa (7,155) (265) 17 (94) 359 - - 327 33 210 716 199 906 (153) (4,899) Badulla (18,369) - (1) (209) 137 - 24 217 (16) (295) (494) 44 (152) (8,007) (27,123) Moneragala (15,473) (38) (6) (387) 25 11 - 66 (32) (294) (1,033) (49) (686) (878) (18,774) Ratnapura (23,541) (3,591) 47 (569) 403 - 132 1,423 21 178 1,189 197 1,625 (1,163) (23,648) Kegalle (9,929) (160) 279 758 958 23 251 1,454 70 1,719 4,229 489 4,487 (241) 4,387 Total (263,522) (7,524) 2,571 4,606 16,292 1,734 3,002 31,082 791 13,060 61,740 9,855 59,915 (38,765) (105,162) Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016 Note: Green areas experience the highest labor demand, and pink areas experience negative labor demand. Table A3: Expected change in number of workers in Sri Lanka disaggregated by districts and sectors – para-tariff scenario w.r.t baseline Districts Agri Mining Textiles WearingApp Chemicals TransportEq Machinery OtherManuf Utilities Construction Trade&Trans Finance SocialServ VegFruit Total Colombo (1,282) (139) 1,129 7,060 (1,454) 57 666 (1,726) (257) 823 1,057 25 2,523 (96) 8,387 Gampaha (3,369) (322) 1,820 13,099 (2,001) 259 623 (1,461) (551) 1,345 971 16 1,950 (235) 12,143 Kalutara (3,078) (228) 1,052 5,349 (805) 76 430 (809) (70) 553 454 8 1,058 (167) 3,822 Kandy (3,679) (186) 527 3,549 (395) 13 109 (616) (347) 757 475 6 1,065 (281) 998 Matale (1,923) (100) 246 1,282 (205) 11 62 (102) (104) 200 141 1 342 (589) (737) Nuwara-eliya (3,244) (25) 221 703 (77) - - (87) (98) 150 109 1 227 (2,853) (4,973) Galle (5,094) - 753 3,134 (339) 14 191 (560) (57) 552 305 4 780 (102) (420) Matara (4,297) - 382 1,611 (217) 29 286 (304) (248) 380 219 4 578 (72) (1,652) Hambantota (2,762) (169) 329 1,303 (567) - 70 (435) (125) 226 151 2 410 (636) (2,202) Jaffna (1,024) (27) 675 25 (56) - 19 (323) (84) 351 119 3 447 (547) (423) Mannar (330) (4) 76 7 (10) - - (25) (11) 46 25 0 51 (17) (193) Vavuniya (469) (8) 438 - (25) - - (35) (46) 81 43 1 102 (199) (117) Mullaitivu (448) (10) 186 25 (6) - - (30) - 43 15 0 53 (55) (228) Kilinochchi (211) (21) 171 86 (4) - - (32) (14) 61 19 1 66 (47) 73 Batticaloa (1,193) (81) 1,156 95 (60) - - (164) - 240 103 1 308 (73) 332 Ampara (1,864) (101) 1,117 266 (114) 9 23 (209) (94) 263 140 1 388 (57) (232) Tricomalee (876) (53) 235 225 (61) - - (108) (35) 214 89 1 245 (58) (183) Kurunegala (7,011) (218) 1,956 5,309 (1,272) 51 264 (994) (163) 815 556 9 1,537 (483) 356 Puttlam (2,503) (185) 1,039 1,277 (476) - 60 (398) (151) 321 230 3 419 (958) (1,321) Anuradhapura (4,395) (20) 138 1,316 (344) - 35 (184) (107) 240 222 4 741 (257) (2,611) Polonnaruwa (1,980) (148) 70 1,226 (254) - - (113) (134) 160 89 2 264 (64) (882) Badulla (3,921) - 80 543 (173) - 65 (120) (87) 234 170 2 579 (2,568) (5,197) Moneragala (2,873) (15) 29 648 (73) 10 - (64) (66) 97 98 2 242 (243) (2,207) Ratnapura (5,879) (1,823) 301 3,104 (334) - 224 (560) (214) 410 273 3 743 (440) (4,191) Kegalle (3,610) (115) 610 4,198 (492) 7 254 (379) (113) 498 245 4 686 (137) 1,656 Total (67,319) (3,998) 14,735 55,437 (9,814) 536 3,381 (9,836) (3,175) 9,060 6,319 104 15,805 (11,236) 0 Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016. Note: Green areas experience the highest labor demand, and pink areas experience negative labor demand. Table A4: Expected change in number of workers in Sri Lanka disaggregated by districts and sectors – full-reform scenario w.r.t baseline Districts Agri Energy Textiles WearingApp Chemicals TransportEq Machinery OtherManuf Utilities Construction Trade&Trans Finance SocialServ VegFruit Total Colombo (1,161) (66) 985 6,575 (535) 111 603 591 (111) (118) 1,525 66 (1,020) (50) 7,396 Gampaha (3,052) (153) 1,587 12,199 (736) 506 564 500 (238) (193) 1,400 41 (789) (122) 11,516 Kalutara (2,788) (108) 917 4,981 (296) 149 389 277 (30) (79) 655 20 (428) (86) 3,572 Kandy (3,333) (88) 460 3,305 (145) 26 98 211 (150) (109) 686 16 (431) (145) 402 Matale (1,741) (47) 214 1,194 (75) 21 56 35 (45) (29) 203 3 (138) (304) (653) Nuwara-eliya (2,938) (12) 193 654 (28) - - 30 (42) (21) 158 2 (92) (1,474) (3,572) Galle (4,614) - 656 2,918 (125) 27 173 192 (25) (79) 440 11 (315) (52) (793) Matara (3,892) - 333 1,500 (80) 56 259 104 (107) (54) 315 9 (234) (37) (1,828) Hambantota (2,502) (80) 287 1,213 (209) - 64 149 (54) (32) 218 6 (166) (329) (1,434) Jaffna (928) (13) 589 23 (21) - 17 111 (36) (50) 171 7 (181) (283) (593) Mannar (299) (2) 66 6 (4) - - 9 (5) (7) 35 1 (21) (9) (228) Vavuniya (425) (4) 382 - (9) - - 12 (20) (12) 62 3 (41) (103) (154) Mullaitivu (406) (5) 162 23 (2) - - 10 - (6) 21 1 (21) (28) (251) Kilinochchi (191) (10) 149 80 (2) - - 11 (6) (9) 27 1 (27) (24) (0) Batticaloa (1,081) (38) 1,008 88 (22) - - 56 - (34) 149 4 (124) (38) (33) Ampara (1,689) (48) 975 248 (42) 17 21 72 (41) (38) 202 3 (157) (29) (506) Tricomalee (794) (25) 205 209 (22) - - 37 (15) (31) 129 1 (99) (30) (435) Kurunegala (6,350) (103) 1,706 4,944 (468) 100 240 340 (70) (117) 803 23 (621) (250) 176 Puttlam (2,267) (88) 906 1,189 (175) - 55 136 (65) (46) 332 8 (169) (495) (678) Anuradhapura (3,981) (10) 120 1,226 (126) - 32 63 (46) (34) 321 10 (300) (133) (2,859) Polonnaruwa (1,793) (70) 61 1,141 (93) - - 39 (58) (23) 128 6 (107) (33) (802) Badulla (3,551) - 70 505 (64) - 59 41 (37) (34) 245 6 (234) (1,327) (4,321) Moneragala (2,602) (7) 25 603 (27) 19 - 22 (28) (14) 142 5 (98) (126) (2,086) Ratnapura (5,325) (864) 262 2,891 (123) - 203 192 (92) (59) 394 9 (300) (228) (3,041) Kegalle (3,270) (54) 532 3,909 (181) 14 231 130 (49) (71) 353 10 (278) (71) 1,205 Total (60,971) (1,896) 12,852 51,626 (3,609) 1,046 3,063 3,368 (1,370) (1,299) 9,115 272 (6,391) (5,806) 0 Source: Staff estimates based on LINKAGE CGE Simulations and HIES 2016. Note: Green areas experience the highest labor demand, and pink areas experience negative labor demand. Annex 2: Linkage between CGE and GIDD The ultimate focus of analysis is the evolution of the distribution of welfare in different scenarios. Starting from the base year “t”, the income or expenditure (Yi,t) of each individual living in a household can be modeled as a function of: (i) household members’ characteristics or assets (endowments) (X), (ii) the market reward for those characteristics (), and (iii) the intensity in how those endowments are used as captured by a set of parameters defining labor force participation and occupation status (|); and, finally, (iv) unobservable components ( ): , = �, , , �, � �, , � (1) The income distribution D for a population of N individuals (or households) in the base year t can be represented by a vector �1, … , … , �, where each Yi,t can be defined as in (1) in terms of endowments, prices, labor status and unobservables28: = �1, … , � = ��1, , , �1, � �, 1, � … �, , , �, � �, , �� (2) How does this distribution change dynamically, for example from year t to year t+k? This framework allows distinguishing two sources that affect the dynamic change of distribution D, both of which are relevant for the assessment of the distributive impact of trade policy reform. The first source consists of the changes in either the parameters or , namely the market rewards to the characteristics (or assets) X and parameters affecting occupational decisions. This means, for example, that inequality for distribution D can go down if the skill premia / is reduced; or if a change in labor demand in sectors with higher wages (a change in ) affects the decision to move to these sectors for some individuals working in sectors with lower wages. The second source of dynamic shift is represented by changes in the distribution of individual and household characteristics (X). Alterations of the structure of the population in terms of age and education, and changes in the size and composition of households, will all affect the distribution of income of that population. 29 Both sources of distributional change matter to the impact of trade reform. In fact, the GIDD allows generating a scenario that includes it, which is then compared to a counterfactual where trade environment remain stable at _ �+ _ �+ the levels observed in t. Comparing the distributions and derived from the two scenarios in effect isolates the distributional impact of shock. Defining the contrasting values of endowments, prices, and labor status to build the two � s can be quite challenging, especially with complex trade negotiations. To do so, the functional form of equation (1) is defined in a simple fashion using only variables available for all countries in the sample. For didactic purposes, the right-hand side of equation (1) only includes age, two levels of education endowments and two sectors of employment workers (country subscripts excluded for simplicity), so equation (1) can be re-written as: , = + 1, ( , , ) + 2, ( , , ) + 3, (, , ) + � , , + , (3) =1 where and are dummy variables identifying whether workers are skilled or unskilled, respectively; and are dummy variables taking the value of 1 if the worker is employed in the agricultural sector or in the non- 28 Note that here the representation of the income distribution is discrete. For a representation with continuous distributions, see Bourguignon, Ferreira, and Lustig (2005). 29 These two sources of dynamic change are not independent one from the other and, in the real world, they are simultaneously determined. The problem of estimating and running a fully simultaneous microsimulation framework is discussed in more detail in Bourguignon and Bussolo (2013). agricultural sector, respectively; captures the proportion of household members in each of the k age cohorts. The s are rewards (prices) to education endowments conditional on the sector of employment,30 and the s are prices associated with household composition. Finally, includes all other income determinants not included in equation (3). A counterfactual expression to (3) for year t +k is: �,+ = ̂1,+ � �+ + � ̂ � ̂ � ̂ � ,+ ,+ �,+ �� + 2,+ � ,+ ,+ �,+ �� (4) ̂3,+ � + � ̂ � � ,+ ,+ �,+ �� + � ,+ ,+ =1 Compared with equation (3), where the demographic characteristics or assets, and the rewards to these assets are obtained directly from the household surveys data, in equation (4) these variables are micro-simulated. The microsimulation proceeds in two steps. ̂ and � , In the first step, for each individual (by household), the variables � are projected using a reweighting procedure that ensures that the changes of composition of the population in terms of age and education at the micro level are consistent with the aggregate changes derived from the age-education projections (including the pipeline effect of the young moving through and up the school system). These aggregate changes, in turn, represent the shifts of the exogenous skilled and unskilled labor supplies in the LINKAGE/ENVISAGE model. With these inputs, the LINKAGE/ENVISAGE generates general equilibrium consistent returns to skilled and unskilled labor in the agriculture and non-agriculture segments of the market. In the second step, the changes in these returns with respect to the base year are used to micro-simulate the returns at the individual level. For example, if the percentage change in the return to skilled workers in the non-agriculture , segment is, according to the CGE, equal to ∆→+ , then in equation (4) the micro-simulated return is obtained as: ̂2,+ = 2, ∆, . →+ When both the simulated characteristics and rewards are determined in these two steps, 31 the new income for individual i can be calculated from equation (4)32 and, when this is replicated for each individual, a new simulated distribution is generated. Note that the variables representing characteristics ( ̂ , � , ̂ , � ) and returns ( �) are also called Linkage Aggregate Variables (LAVs) as they are the variables that link the GIDD and LINKAGE models. These steps are shown below. 30Note that unskilled workers employed in agriculture, i.e. when these dummies , , are equal to one, are the reference category, so they are excluded from the equation. 31 Characteristics and rewards should be jointly determined. For example, the decision to pursue further education depends on the return and cost of education, which in turn depend on how many people pursue education and enter the market for educated workers. However, this simultaneity is not implemented in the GIDD, nor for most microsimulation models. For a detailed discussion of these issues, see Bourguignon and Bussolo (2013). 32 Note that the intercept needs to be micro-simulated as well. In fact, this is modified so that the after all the other changes are applied, the per capita economy-wide growth can be targeted. In other words, the simulated characteristics and returns change the shape of the income distribution and its position (or its mean). However, the complete effect of income growth is not captured by just these changes. For example, accumulation of capital/land and other factors and changes in their returns are not considered in this version of the GIDD for lack of good data in the surveys. Thus, the intercepts are shifted equally for all individuals (i.e. a distribution-neutral shift) to ensure consistency between the per capita income growth simulated in the LINKAGE and the change of the mean in the GIDD microdata. This is likely to cause underestimation of changes of inequality within countries, but it does not affect inequality across countries. A couple of variants of equation (4) can be used to decompose the full change in the distribution. A first variant consists of using equation (4) with only new micro-simulated characteristics but with unchanged returns. Formally, this variant can be written as: � ′,+ = � � + 1, � ̂ � � � � ,+ ′,+ �,+ �� + 2, � ,+ ′,+ �,+ �� (4)’ � +3, � � � � ,+ ′,+ �,+ �� + � , ,+ + ,+ =1 This new equation (4)’ represents a ‘partial’ change in the incomes, and thus in the distribution, that is due to a pure demographic (or quantity) education effect. Other characteristics and rewards have not been changed yet. 33 Likewise, a second variant of equation (4), where only the rewards are modified but not the characteristics, can be set up to represent the shift in distribution exclusively due to changes in rewards. Microsimulation mechanics: reweighting household surveys The first step in the microsimulation exercise is to implement a set of changes in the household surveys’ demographic structure. The population growth adjustment is particularly important in countries with high fertility rates, such as those in Sub-Saharan Africa. In practical terms, the adjustment for population growth allows the analysis to explicitly take into account the changes in the size of the working-age population. 34 We perform population and education projections during the first stage of the microsimulation model and in creating the baseline scenario for the CGE model. For each country, we construct the demographic profile in two steps. First, the age and gender composition are exogenously determined following medium variant estimates from the World Population Prospects (UN DESA 2019). In the second step, following François Bourguignon and Bussolo (2013), country-specific educational profiles are constructed using initial educational achievement levels observed in the household surveys with some conservative yet simple assumptions about educational progress. Starting with a global collection of household surveys, country specific demographic profiles are constructed by partitioning each country’s total population into: (1) 16 age-groups (0-4, 5-9, 10-14, …, 65-69, 70-74, 75+; (2) two gender groups; and (3) three different levels of educational attainment: (i) No-education or primary; (ii) secondary; and (iii) tertiary education. As noted earlier, age and gender totals are based on data from the UN DESA (2019) medium variant population projections. In terms of education, the GIDD standard assumption is that as the population ages, the average educational attainment in a country increases through a pure pipeline effect – as younger and more educated cohorts replace older cohorts. For example, if at time t, half of the population in the cohort formed by individuals between 25 and 30 years of age have post-secondary education, then after ten years (at t+10), half of the population between 35 and 40 will have post-secondary education. Furthermore, the assumption for younger cohorts is no improvement in enrollment and graduation rates from those observed at time t. In other words, the average educational attainment of the young cohorts in the future is equal to the average 33 Note that the constraints of the reweighting procedure are in terms of (age cohorts and) education levels of the population in year t+k, and not in terms of sectoral employment. However, because skilled workers may initially be more concentrated in the non- agricultural sectors, increasing the weight of skilled workers in the total population also generates an increase in the share of non- ̂ ′ agricultural employment. This is why, in equation (4’), sectoral dummies are labeled as ,+ ′ ,+ with a prime (’) sign. These do ̂ not necessarily represent the final sectoral employment which is determined by the CGE model. 34 For the case of Sri Lanka, as will be shown later, working age population is projected to diminish by approximately 100,000 individuals by 2030. educational levels of the 20 to 24 cohort at time t. This is a conservative assumption given that the 20 to 24 cohort observed at time t may not have the maximum educational level attainable. 35 Microsimulation mechanics: Changes in factor returns and prices The second step is to adjust factor returns by skill and sector in accordance with the results of the CGE model. The GIDD imposes an entirely new vector of earnings on each worker, conditional on that worker being in sector s and having an educational attainment e. The third step adjusts the average income/consumption per capita to guarantee that it changes exactly in line with the CGE results. Lastly, GIDD constructs a household-specific deflator to adjust for changes in relative prices. The price deflator is constructed using initial and final price indexes of food versus non-food expenditure from the macro model and household-specific budget consumption shares for food and non-food expenditure observed in the micro data. Beginning with a distribution of earnings from labor by sector and skill level [, ] in the macro data, let us define a set of wage gaps as follows: , , = −1 (5) 1,1 and a similar set of wage gaps for the macroeconomic counterfactual scenario: �, �, = −1 (6) �1,1 �1,1 and where 1,1 is the average earnings from the labor of unskilled workers in agriculture and �, are their predicted values from the CGE model in the counterfactual scenario. All right-hand side values in equation (5) are observable in the CGE model benchmark data set, and all right-hand side values in equation (6) are known values in the CGE model simulations. The micro data will also have a set of wage premia which, in general, will differ from the CGE data. Analogous to equations (5) and (6), let us define: ′ ′, , = −1 (7) ′1,1 ′ �′, � , = −1 (8) �′1,1 ′ where , are the wage premia based on averages by skill group and sector in the household data; ′, are the average earnings of labor in sector s and skill group e based on the household data; ′1,1 are the average earnings of unskilled labor in agriculture based on the household data; and the ′ � are the predicted values at the household level as a result of the policy change. All right-hand side values of equation (7) are known from the initial household ′ data. In order to calculate � , , we define: 35 In practical terms, the microsimulation model recalibrates each household sample weight to match the age, gender, and education projected totals. A new probability distribution can be obtained by solving an optimization problem based on a minimum cross-entropy criterion as in (Olivieri et al. 2014). The minimum cross-entropy method assures that the new sets of age, gender and education deviate as little as possible from the initial distributions. See (Wittenberg 2010) for a technical description and implementation of this method. ′ ′ �, � , = , (9) , We may calculate the left hand side of equation (9), since the three values on the right hand side are known from equations (5), (6) and (7). Equation (9) implies that even if initial wages differ between the CGE and micro models, the percentage change in the wage gaps will be consistent across the two models. By passing on percentage changes in wage premia by type of worker, instead of percentage changes in wages, the possibility of wage gaps moving in opposite directions in the macro data and in the household data is eliminated. Within each group of workers, distributional changes occur; but, on average, for any group of workers, the relative wages for each type of worker is constrained to be consistent with the corresponding growth rates from the CGE model. Given the known values in equations (5) to (9), and defining average wages for unskilled labor in agriculture as �′1,1, it is possible to calculate the percentage changes in average wage numeraire in the GIDD, so that ′1,1 = income of households in sector s and skill level e that are consistent with wage gaps expressed in Equation (10): �′, / ′ , (10) Note that Equation (10) only operates on labor income. In order to adjust the micro data such that the weighted average percentage change in the per capita income/consumption across all households matches the change in real consumption per capita in the CGE model, a subsequent adjustment is carried out. Define as real per capita income calculated from the CGE model in the benchmark and � as its predicted value in the CGE model simulation. Define ′ ℎ = ∑∈ℎ ′,ℎ /ℎ as the per capita income of household h in the benchmark equilibrium, where ′,ℎ is the income of the ith member of household h, and n is equal to the size of household h. Similarly, define ′ �ℎ �′,ℎ /ℎ = ∑∈ℎ ′ ′ where �,ℎ and �,ℎ are the unadjusted and adjusted values, respectively, of the income of the i member of th household h in the counterfactual of the micro-model; the role of λ is explained by equation (14) below. Then define ′ as the weighted average value of real per capita income across all households, i.e., ′ � ℎ ℎ = ′ (11) ℎ where ℎ is the weight of household h in aggregate income in the benchmark data. Correspondingly, � ℎ �ℎ ′ �′ = (12) ℎ is the weighted average per capita income value in the policy simulation. Note that ∑ℎ ℎ = 1, ∑ℎ ℎ = 1 and λ is a scalar. Equations (11) and (12) allow for different household weights since the weights of the households typically change over time. So, to make sure that the percentage change in the aggregate value of household income � ′ by equation (13): is consistent with the CGE model, we constrain � � ′ = ′ (13) We implement this constraint in a distribution neutral way. That is, we adjust all household income in the ′ counterfactual by a scalar λ such that per capita household income equals �ℎ : as a result, can be defined by: � ′ � ℎ �ℎ = ′ (14) ℎ Even though the GIDD ignores other forms of income, such as capital income, this transformation guarantees consistency between the weighted average household income assessment and the CGE model assessment. For poor households, which is the focus of our work, the assumption should be reasonably accurate, since poor households have little capital income. There is a margin of error for wealthier households. But for these wealthier households, it is skilled labor rather than unskilled labor that tends to be more important and Bussolo, De Hoyos, and Medvedev (2010) have noted a tendency for skilled wage and returns to capital to be correlated. Finally, macroeconomic estimates of changes in agricultural and non-agricultural prices are distributed across heterogeneous households using the following method. Let us define the initial per capita monetary income of ′ household h, ℎ , and the purchasing power of household h, ℎ , as the ratio of its monetary income divided by a household-specific price index capturing the household’s consumption patterns in terms of food and non-food expenditure: ′ ′ ℎ ℎ ℎ = = (15) ℎ ℎ + (1 − ℎ ) where Pf and Pnf are food and non-food price indexes and αℎ is the proportion of household h budget spent on food. The αℎ parameter in the denominator of the right-hand side of Equation (15) can be estimated with household data using the following specification: ′) ℎ = 0 + 1 ln(ℎ + ℎ (16) where eh is a vector of household-specific errors that are assumed to be distributed with (ℎ ) = 0 and ̂1 remain constant, the new budget share spent on food ̂0 and (ℎ ) = 2 . Assuming that estimated parameters ′ ′ for household h, ℎ , at the counterfactual per capita income, �ℎ , can be obtained from: �ℎ ′ ̂1 ln( ̂0 + = �ℎ ′) + ̂ℎ (17) The changes in real per capita incomes brought about by a change in relative prices of food versus non-food can be approximated by the following linear expression: ′ �ℎ �ℎ = ′ ′ ′ ′ (18) �ℎ + (1 − �ℎ ) where �ℎ in Equation (18) is the real per capita income adjusted for changes in relative prices of food versus non- food expenditure. �ℎ is the counterfactual measure of real per capita income of household h for the analysis of poverty and shared prosperity. Estimating regional worker allocation As detailed under the section on disaggregated (district-level) labor market effects, to maintain the consistency of the results of the CGE-GIDD framework approach, the allocation of workers at a disaggregated level is undertaken in two additional steps. First, we derive the distribution of workers ′, = (1,1 , … … , 1, , 2,1 , … … , 2, , … … , ), which are sample weighted to obtain nationally representative, population weighted, distribution of workers in Sri Lanka from the HIES 2016, represented by = + to represent a es skilled worker, and an eu unskilled worker. The number of workers is calculated by destring and sector. These weights are simple proportions of workers in each respective sector , and district , as estimated from HIES 2016.36 ′, = � = ∀ = 1, … … , ; . . � = 1 =1 (19) = ∀ = 1, … … , ; . . � = 1 =1 The total ‘new’ workers across time, formally defined as +1 , , is informed by the United Nations population projections and allocated to conform to the initial sector and district weighted distribution of workers ′, . This provides the allocation of workers in the baseline scenario over time by sector and district. , , , +1 , = , 1 + ∆,+1 ,1 (20) The deviations from the baseline scenario, denoted by 1 , of all specific reforms scenarios, denoted by ∗ , retains the distribution of total number of workers +1 , , and reallocates workers to expanding and shrinking sectors based on changes in labor demand. +1,1 , ∗ , , , − +1 , , = ∆,+1 ,1 ′ ′ − ∆ (21) 36 Some of the sectors were collapsed to the nearest resembling sector if they were too small and hence very few observations of individuals are found in the HIES 2016.