## Non-traded Gains from Trade: Selection in the Non-traded Sector

## Source details

**Canonical URL:** [Non-traded Gains from Trade: Selection in the Non-traded Sector](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023265-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023265-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023265-print-pdf.pdf.json)

---

### Empirical setting and motivation
- Focus: role of the local labor market in a country’s adjustment to trade shocks, emphasizing adjustment among producers within the non-traded sector.
- Empirical setting: Brazil’s import liberalization of the 1990s — unexpected import tariff reductions and removal of most nontariff import barriers; imports as a share of GDP roughly doubled in the early 2000s.
- Reasons to emphasize non-traded sector:
  - Non-traded sector is less nationally integrated and more influenced by local labor market conditions.
  - In Brazil, the non-traded sector accounts for nearly twice the share of employment and production as the traded sector.
  - OECD comparisons (2000s): services employment corresponded to 6 times agriculture and manufacturing combined in the United States, 4 times in France, 3 times in Germany, 3.3 times in the Eurozone, 5 times in Russia, 11 times in South Africa, and 2 times in Mexico.

### Key empirical findings (reallocation and plant-level patterns)
- Regional reallocation:
  - More exposed regions experienced more job reallocation compared to less exposed regions, in both traded and non-traded sectors.
- Plant-level regularities:
  - Small plants within the same local labor market were more likely to exit and to shrink in response to the policy compared to larger counterparts.
  - Overall reallocation observed: from small to large producers.
- Importer dynamics:
  - Pre-liberalization importers had larger survival probabilities and employment growth relative to small plants that never imported.
  - Regions more exposed to the liberalization saw non-traded producers more likely to become importers post-liberalization.
  - “New importers” tended to be medium-sized plants and experienced the largest growth in employment and wage bill once importing.
- Robustness: results hold using both Kovak (2013) regional exposure and a model-based structural measure of regional exposure.

### Quantified job flows (regional exposure effects — Table 1 & Table 2)
- Traded sector (θ_i^T coefficients from Table 1):
  - Exit: -0.053 (0.195)
  - Contract: 0.300 ∗∗∗ (0.101)
  - Entry: -0.397 ∗∗ (0.168)
  - Expand: 0.034 (0.073)
- Non-traded sector (θ_i^{NT} coefficients from Table 1):
  - Exit: -0.300 ∗ (0.168)
  - Contract: 0.112 (0.090)
  - Entry: -0.144 (0.139)
  - Expand: 0.203 ∗∗∗ (0.058)
- Quantified job reallocation (Table 2: ˆθ_i^K ∑_{r,k∈K} lib_r · L_rk,90):
  - Traded sector totals:
    - Exit: -43,039
    - Contract: 241,440
    - Entry: -233,250
    - Expand: 20,242
  - Non-traded sector totals:
    - Exit: -459,459
    - Contract: 171,586
    - Entry: -264,075
    - Expand: 373,800
- Aggregate interpretations:
  - Traded sector: over 240,000 additional jobs destroyed due to plant contraction and over 233,000 fewer jobs created due to decreased plant entry.
  - Non-traded sector: 450,000 fewer jobs disappeared due to reduced plant exit and an additional 373,000 jobs from incumbent plant expansion — positive effect exceeding 820,000 jobs.
- Fact 1: "Import liberalization caused quantitatively nontrivial job reallocation among plants within both the traded and non-traded sectors."

### Establishment-level size heterogeneity (Table 3)
- Regression setup: survival (Surv_e) probit and employment growth g_e OLS conditional on survival; size bins b(e) = 1..4 (quartiles).
- Traded sector estimates (θ_k):
  - Extensive (Survival):
    - θ_1 = -0.75 ∗∗ (0.41)
    - θ_2 = -0.48 (0.45)
    - θ_3 = -0.53 (0.43)
    - θ_4 = -0.72 (0.44)
  - Intensive (Growth):
    - θ_1 = -19.30 ∗∗∗ (4.34)
    - θ_2 = -15.10 ∗∗∗ (3.72)
    - θ_3 = -14.45 ∗∗∗ (3.64)
    - θ_4 = -17.98 ∗∗∗ (4.19)
- Non-traded sector estimates (θ_k):
  - Extensive (Survival):
    - θ_1 = -0.80 ∗∗∗ (0.28)
    - θ_2 = -0.18 (0.26)
    - θ_3 = 0.26 (0.30)
    - θ_4 = -0.71 ∗∗∗ (0.25)
  - Intensive (Growth):
    - θ_1 = -3.68 ∗∗∗ (1.11)
    - θ_2 = -1.25 ∗∗ (0.55)
    - θ_3 = 2.83 ∗∗∗ (0.53)
    - θ_4 = -5.61 ∗∗∗ (1.07)
- Interpretation:
  - Coefficients roughly follow an inverse-U shape across initial size distribution; reallocation from smallest (and largest) producers towards top-middle of size distribution.
- Fact 2: "The import liberalization induced job reallocation from small to larger producers in both traded and non-traded sectors."

### Importing behavior and selection (Tables 4–6, Figures)
- Descriptive import patterns:
  - Between 1986–1990, over 30% of importing establishments operated in the non-traded sector; this share grew to over half in 1997–2001.
  - Intermediate inputs account for more than half of aggregate imports in Brazil; inputs from traded industries account for over 20% of total input costs in the non-traded sector.
- Plant characteristics by importing status (Table 4):
  - Always importers (Importers_90) relative to never importers:
    - ln(Empl): 1.612 *** (0.129)
    - ln(WB): 2.233 *** (0.134)
    - ln(Wage): 0.621 *** (0.029)
    - SkillShare: 0.184 *** (0.015)
  - New importers (NewImp_99) relative to never importers:
    - ln(Empl): 1.187 *** (0.096)
    - ln(WB): 1.645 *** (0.097)
    - ln(Wage): 0.459 *** (0.031)
    - SkillShare: 0.120 *** (0.021)
  - Interpretation: always importers employed 161% more employees and paid 62% higher wages; new importers employed 120% more and paid 46% higher wages initially.
- Survival and growth by importing behavior (Table 5):
  - Survived (Importers_90): 0.095 *** (0.013) → importers 9.5 p.p. more likely to survive 1990–1999.
  - ln(Empl) among survivors (NewImp_99): 0.683 *** (0.027) → employment at new importers grew 68% more than never importers.
  - dln(WB) among survivors (NewImp_99): 0.843 *** (0.024)
- Causal effect of exposure on import adoption (Table 6):
  - Example coefficient: 0.033 (interpretation: a 10 p.p. larger effective tariff decline → 0.33 p.p. larger probability of becoming an importer; 0.33 p.p. = 48% of overall share of new importers).
  - Model-based exposure specifications (Table C.2):
    - θ: 0.048*** (0.001) in column (1)
    - θ: 0.045*** (0.007) in column (2)
    - θ: 0.013** (0.005) in column (3)
- Fact 3 and Fact 4:
  - Fact 3: Majority of Brazil’s imports during the liberalization were intermediate goods; many importers operated in non-traded industries.
  - Fact 4: New importers were initially higher performing than never importers but lower than always importers; they grew the most and their adoption was concentrated in more exposed regions and larger initial plants.

### Theoretical model: selection into importing and predictions (Sections 3)
- Environment:
  - Partial equilibrium model with heterogeneous non-traded producers producing differentiated varieties, sourcing intermediate inputs locally (m^d_j) or internationally (m_j), with fixed importer cost F^m_j and participation fixed cost F^n_j.
  - Productivity cutoffs: operating cutoff z^n_j and importer cutoff z^m_j with z^n_j < z^m_j in baseline.
  - Importer advantage: Δ_j ≡ (m^d_j / m_j)^{α_j}.
- Comparative statics and predicted reallocation when importing becomes cheaper:
  - Extensive margin:
    - Small establishments exit (higher z^n_j).
    - Medium-sized plants become importers (lower z^m_j).
  - Intensive margin:
    - Small surviving non-importers shrink.
    - Large establishments that do not change importing status grow.
    - Medium-sized producers that become importers grow the most.
- Role of selection:
  - Selection into importing is critical to reproduce empirical reallocation patterns in the non-traded sector.
  - Special no-selection case (F^m_j = 0): all active producers import; without selection, non-traded producers respond uniformly and within-industry reallocation channel disappears.
- Welfare implication:
  - Reallocation away from small to larger producers increases industrial productivity of the non-traded sector — a novel margin for welfare gains from trade.

### Structural model, regional exposure measure, and counterfactuals (Appendix C / Section 3)
- General-equilibrium version derived to produce a theoretically consistent regional exposure measure that:
  - Incorporates input-output linkages, producer-level selection into imports, and labor allocation across non-traded industries.
  - Yields a model-based exposure measure used in regressions; results qualitatively similar to Kovak (2013) exposure and reduced-form findings.
- Counterfactual analytic result for regional wage response (Equation 14 structure):
  - dlnw = (weighted sum over non-traded industries of αj 1−αj αj δj (1−δj) swb j normalized across non-traded industries) × (∑_{i∈T} ηij · dlnp∗i) + other terms
  - Mechanisms captured:
    - Input-output linkages (ηij).
    - Dependence on industry wage-bill share swb jr.
    - Importance of intermediate input intensity αj.
    - Nonlinear dependence on importer share δjr via δjr (1−δjr) (inverse-U shape).
- Empirical mapping choices for counterfactual exercise:
  - dlnp∗i = observed industrial tariff cuts.
  - α from National Accounts; δj and swb jr from RAIS; ηij from IBGE Input-Output table; tariffs from Kume, Piani, and Souza (2003).
- Model-based regression evidence (Table C.1 and Table C.2) supports:
  - Labor reallocation toward larger producers in non-traded sector (Table C.1).
  - Higher probability of import adoption in more exposed regions (Table C.2).

### Policy-relevant implications and welfare insights
- Import liberalization affects regional wages and local labor reallocation through:
  - Structure of input-output linkages (ηij).
  - Industry composition and local wage-bill shares (swb jr).
  - Firm-level importing decisions and their within-industry distribution (δjr).
  - Intensity of intermediate inputs (αj).
- Selection into importing is an important channel for local labor reallocation; policies altering costs of importing intermediates can shift employment from small to larger producers, raising productivity and concentration in non-traded industries.

### Conclusion (summary)
- Brazil’s unilateral import liberalization in the early 1990s generated differential local exposure and induced substantial reallocation of workers within local labor markets.
- Reallocation occurred in both traded and non-traded sectors but was quantitatively larger in non-traded sectors due to their size.
- Empirical evidence and a parsimonious model indicate that selection into importing intermediate inputs by larger/mid-sized producers drove much of the observed reallocation: small producers exited or shrank, while new importers (often medium-sized) expanded the most.
- The importing channel constitutes a novel margin for gains from trade through within-industry productivity improvements and increased concentration in non-traded industries.

*Source: wpiea2023265-print-pdf — Non-traded Gains from Trade: Selection in the Non-traded Sector*

### 1. Introduction1

### 1. Introduction

### Empirical setting and motivation
- Focus: role of the local labor market in a country’s adjustment to trade shocks, emphasizing adjustment among producers within the non-traded sector.
- Motivation:
  - The non-traded sector is less nationally integrated and thus more influenced by local labor market conditions.
  - In Brazil (empirical setting), the non-traded sector accounts for nearly twice the share of employment and production as the traded sector.
  - OECD comparisons (2000s): services employment corresponded to 6 times the employment in agriculture and manufacturing put together in the United States, 4 times in France, 3 times in Germany, 3.3 times in the Eurozone, 5 times in Russia, 11 times in South Africa, and 2 times in Mexico.
- Policy event studied: the Brazilian import liberalization of the 1990s — unexpected import tariff reductions and the removal of most nontariff import barriers; imports as a share of GDP roughly doubled in the early 2000s.

### Key empirical findings
- More exposed regions experienced more job reallocation compared to less exposed regions, in both traded and non-traded sectors.
- Plant-level regularities:
  - Small plants within the same local labor market were more likely to exit and to shrink in response to the policy compared to larger counterparts.
  - The pattern holds strongly in both traded and non-traded industries.
  - Overall reallocation observed: from small to large producers.
- Importer versus non-importer differences:
  - Pre-liberalization importers had larger survival probabilities and employment growth relative to small plants that never imported.
  - Regions more exposed to the liberalization saw non-traded producers more likely to become importers in the post-liberalization period.
  - “New importers” tended to be medium-sized plants and experienced the largest growth in employment and wage bill once importing.
- Robustness:
  - Empirical findings remain when using both the regional exposure measure proposed by Kovak (2013) and a model-based structural measure of regional exposure developed in Section 3 and Appendix C.

### Mechanism: self-selection into importing
- Suggestive evidence: producers in the non-traded sector self-select into importing, expanding as they gain access to inputs from international markets.
- Implication: selection into importing by more productive/medium-sized firms drives local job reallocation and differential plant dynamics.

### Theoretical model and predictions
- Model features:
  - Parsimonious partial equilibrium model of selection into importing.
  - Heterogeneous producers in non-traded industries produce differentiated varieties and source intermediate inputs either locally or internationally.
  - Importing involves an additional fixed cost but reduces variable costs (cheaper, higher quality, or larger variety of inputs).
  - Only the largest and most productive plants engage in importing under baseline costs.
- Predicted reallocation patterns when importing costs fall:
  - Extensive margin:
    - Small establishments exit.
    - Medium-sized plants become importers.
  - Intensive margin:
    - Small producers that survive but do not import shrink.
    - Large establishments that do not change importing status grow.
    - Medium-sized producers that become importers grow the most (notable expansion in employment and wage bill).
- Role of selection:
  - Selection into importing is critical to reproduce empirical reallocation patterns in the non-traded sector.
  - In a special case with no fixed cost of importing, selection patterns in the traded sector persist but non-traded producers respond uniformly regardless of size or productivity.
- Welfare implication:
  - Reallocation away from small to larger producers increases industrial productivity of the non-traded sector — a novel margin for welfare gains from trade.

### Contributions to the literature
- Connects to three strands:
  1. Local labor market effects of trade shocks (e.g., Autor, Dorn, Hanson; Kovak; Asquith et al.): documents establishment death and within-industry reallocation; complements by providing plant-level analysis distinguishing effects by plant characteristics.
  2. Effects of trade shocks in Brazil (e.g., Dix-Carneiro and Kovak): complements prior work by uncovering distributional effects among non-traded employers and a new margin for gains from trade.
  3. Effects of imported intermediate inputs (e.g., Amiti and Konings (2007); Pinelopi K. Goldberg, Khandelwal, et al. (2010); Bas and Paunov (2021)): relates to firm productivity gains from imported inputs and complements by emphasizing non-traded sector reallocation and selection into importing.
- Novelty claim: first to propose reallocation of workers in the non-traded sector as a margin for gains from an import liberalization.

### Paper organization
- Section 2: empirical findings.
- Section 3: partial equilibrium model to study welfare consequences of the reallocation patterns.
- Section 4: conclusion.

*Source: wpiea2023265-print-pdf - 1. Introduction1*

### 2. Empirical analysis

### 2. Empirical analysis

### Policy environment and data
- Policy event: Brazilian import liberalization comprised of sizable, heterogeneous, and unexpected one-sided tariff cuts between 1990 – 1995.
- March 1990: announcement of atarificação converting non-tariff barriers into tariffs, then steadily lowered over next five years.
- Tariff changes vary by industry; pre-liberalization protection level was main determinant of tariff cut size.
- Primary data source: RAIS (Relação Anual de Informações Sociais) — matched employer-employee panel covering the universe of formal Brazilian employees.
  - RAIS contains unique identifiers for workers and establishments, municipality and industry information, worker age, education, and earnings.
  - Exclusions/adjustments: workers with faulty IDs and zero earnings excluded; restricted to highest-paying job per worker; producers with missing ID, municipality, or industry dropped; producers in mining, utilities, and public industries not considered; all nominal variables converted to 2000 Brazilian Reais.
- Supplementary data:
  - 1991 and 2000 Demographic Censuses for initial distribution of industrial activity and informal worker shares.
  - Industry-level import tariffs compiled by Kume, Piani, and Souza (2003).
  - Imports, GDP, and input-output (IO) table from IBGE’s Sistema de Contas Nacionais (SCN).
  - Imports by product type from Fundação Centro de Estudos do Comércio Exterior (Funcex).
  - Identity of establishments that imported or exported from the Brazilian Ministry of Economy.
- Data limitation noted: RAIS omits self-employed and informal workers.

### Measuring local exposure to the trade liberalization
- Local labor markets: 404 geographic microregion units (IBGE), grouping economically related municipalities; terms “local labor market”, “region”, and “microregion” used interchangeably.
- Main explanatory variable: microregion effective exposure (effective regional tariff change), using measure proposed by Kovak (2013).
- Effective tariff change formula (as presented):
  - ras_lib_r = − ∑_k λ_rk / α_k  ∑_j λ_rj / α_j  dln(1+τ_k)
  - where λ_rk = initial share of employees in microregion r employed in industry k (calculated using 1991 Census) and α_k = share of industry k’s non-labor inputs (from 1990 input-output matrix).
- Liberalization measure weights industry-level tariff changes by microregion initial employment concentration and industry expenditure on labor input.

### Reallocation in local labor markets — definitions and empirical specification
- Outcome variables for each microregion r and industry k:
  - s_exit_rk = share of jobs in 1990 destroyed between 1990 and 1999 due to plant exit.
  - s_contract_rk = share of jobs in 1990 destroyed between 1990 and 1999 due to plant contraction.
  - s_entry_rk = share of jobs in 1999 that were created between 1990 and 1999 due to entry of new plants.
  - s_expand_rk = share of jobs in 1999 that were created between 1990 and 1999 due to plant expansion.
- Empirical specification for each margin i ∈ {exit, contract, entry, expand}:
  - s_i_rk = α_i0 + θ_i^{K(k)} lib_r + β_i X_rk + α_i^{s(r)} + α_i^k + ε_i_rk
  - K = T (traded) or NT (non-traded). θ_i^K captures effect of regional liberalization exposure on job flow patterns by sector.
- Controls X_rk include:
  - industry k’s share of microregion r’s total workers in 1990;
  - share of total payroll in microregion r attributable to industry k in 1990;
  - log mean wage in industry k in 1990;
  - pre-trends (1986–1989) for these variables;
  - 1990 informal share in each region-industry pair and 1990–1999 evolution of this variable;
  - pre-trend values of each margin s_i_rk.

### Reallocation in local labor markets — key regression estimates (Table 1)
- Table 1 reports θ_i^T and θ_i^{NT} coefficients (with robust standard errors) for each margin.

- Traded sector (θ_i^T):
  - Exit: -0.053 (0.195)
  - Contract: 0.300 ∗∗∗ (0.101)
  - Entry: -0.397 ∗∗ (0.168)
  - Expand: 0.034 (0.073)

- Non-traded sector (θ_i^{NT}):
  - Exit: -0.300 ∗ (0.168)
  - Contract: 0.112 (0.090)
  - Entry: -0.144 (0.139)
  - Expand: 0.203 ∗∗∗ (0.058)

- Sample size reported: N = 6617 (columns differ slightly across margins but table displays N6617661766176617 in source).

- Empirical interpretation:
  - Regions more exposed experienced more job destruction due to contraction and less job creation due to entry in both sectors.
  - In non-traded industries, exposed areas experienced higher job creation due to plant expansion (θ_expand^{NT} = 0.203 ∗∗∗).

### Quantification of total jobs reallocated attributable to differences in exposure (Table 2)
- Calculated as: ˆθ_i^K ∑_{r,k∈K} lib_r · L_rk,90 for each margin i and sector K, using coefficients from Table 1 and 1990 employment L_rk,90.

- Traded sector totals:
  - Exit: -43,039
  - Contract: 241,440
  - Entry: -233,250
  - Expand: 20,242

- Non-traded sector totals:
  - Exit: -459,459
  - Contract: 171,586
  - Entry: -264,075
  - Expand: 373,800

- Aggregate interpretation:
  - Traded sector: over 240,000 additional jobs destroyed due to plant contraction, and over 233,000 fewer jobs created due to decreased plant entry — net negative effect totaling over half a million jobs.
  - Non-traded sector: 450,000 fewer jobs disappeared due to reduced plant exit and an additional 373,000 jobs from incumbent plant expansion — positive effect exceeding 820,000 jobs.
- Fact 1: "Import liberalization caused quantitatively nontrivial job reallocation among plants within both the traded and non-traded sectors."
- Emphasis: although some coefficients larger in traded sector, most quantified job flows are larger in the non-traded sector because the non-traded sector accounts for almost double total employment in Brazil.

### Establishment-level employment patterns — methodology
- Two establishment-level outcomes (for establishment e):
  - Surv_e = 1{e survives between 90–99}.
  - g_e = (L_e99 − L_e90) / L_e90 (employment growth conditional on survival).
- Regression framework:
  - Surv_e = Φ( α_0 + ∑_{k=1}^4 θ_k 1_{b(e)=k} lib_{r(e)} + β X_e + α_{s(e)} + α_{j(e)} + α_{b(e)} + ε_e )
    - Probit model for survival (Φ denotes standard normal CDF).
  - g_e = α_0 + ∑_{k=1}^4 θ_k 1_{b(e)=k} lib_{r(e)} + β X_e + α_{s(e)} + α_{j(e)} + α_{b(e)} + ε_e
    - OLS for growth conditional on survival, with Heckman correction term from (2).
- b(e) denotes 1990 regional employment quartile of establishment e (size bins 1–4).
- Controls X_e include: plant’s initial share of employment and wage bill in microregion-industry, log mean wage, 1986–1989 pre-trends, initial skill intensity, pre-period employment growth, sector-specific regional informal share and its evolution (1990–1999).
- Fixed effects: state, industry, and initial regional employment quartiles.

### Establishment-level employment patterns — key estimates (Table 3)
- Table 3 reports θ_k coefficients for k = 1..4 for both extensive (Surv) and intensive (g) margins, by sector.

- Traded sector
  - Extensive (θ_1..θ_4):
    - θ_1 = -0.75 ∗∗ (0.41)
    - θ_2 = -0.48 (0.45)
    - θ_3 = -0.53 (0.43)
    - θ_4 = -0.72 (0.44)
  - Intensive (θ_1..θ_4):
    - θ_1 = -19.30 ∗∗∗ (4.34)
    - θ_2 = -15.10 ∗∗∗ (3.72)
    - θ_3 = -14.45 ∗∗∗ (3.64)
    - θ_4 = -17.98 ∗∗∗ (4.19)

- Non-traded sector
  - Extensive (θ_1..θ_4):
    - θ_1 = -0.80 ∗∗∗ (0.28)
    - θ_2 = -0.18 (0.26)
    - θ_3 = 0.26 (0.30)
    - θ_4 = -0.71 ∗∗∗ (0.25)
  - Intensive (θ_1..θ_4):
    - θ_1 = -3.68 ∗∗∗ (1.11)
    - θ_2 = -1.25 ∗∗ (0.55)
    - θ_3 = 2.83 ∗∗∗ (0.53)
    - θ_4 = -5.61 ∗∗∗ (1.07)

- Sample sizes displayed in source: N17878773082825194363200 (reported across rows/columns in Table 3).
- Empirical interpretation:
  - Coefficients roughly follow an inverse-U shape across initial size distribution, peaking at second or third employment quartiles in both sectors and for both margins.
  - Pattern implies systematic reallocation from the smallest (and largest) producers towards producers in the top-middle of the size distribution.
  - In non-traded sector extensive margin: significant effects on survival for bottom and top size bins, not for middle bins.
  - Intensive margin in non-traded sector: coefficients increase across first three quartiles, indicating distribution of labor tilts toward larger establishments in exposed regions.
- Fact 2: "The import liberalization induced job reallocation from small to larger producers in both traded and non-traded sectors, through differential patterns of plant exit and surviving plant growth through the size distribution."

### Descriptive statistics of importing in Brazil (summary of Figure 2 panels)
- Panel (a) (period averages reported in source):
  - NT share (%) among importers, plants, workers, wage bill: plotted for periods 1986–1990 and 1997–2001 showing non-traded sector participation in importing establishments’ activity.
- Panel (b) (time series reported in source):
  - Intermediaries import share (left axis) and traded share of NT total inputs (right axis) plotted over years 1985, 1990, 1995, 2000.
  - Shaded area indicates years of the liberalization policy.
- Sources for these descriptive statistics: RAIS, the Brazilian Ministry of Economy, Funcex and IBGE/SCN.

*Source: IMF Working Paper — chapter "2. Empirical analysis" (wpiea2023265-print-pdf).*

### Appendix Figures A.3 and A.4 replicates the empirical exercises industry by industry. Additionally,

### Appendix Figures A.3 and A.4 replicates the empirical exercises industry by industry. Additionally,

### Plant size comparability across regions
- Figure A.2 in Appendix A shows that plants in each quartile are of similar size regardless of region, eliminating the possibility that these results are driven by differences across regions of the size of plants in each quartile.

### Importing behavior and new importers — overview
- The analysis proceeds in three steps:
  - Show that importing was a key potential margin for adjustment among producers in the non-traded sector.
  - Establish that importers are qualitatively different from non-importers both before and after the liberalization, supporting selection into importing and the role of new importers in growth of middle-size producers.
  - Use the liberalization measure to provide direct evidence that otherwise similar producers in high exposure labor markets were more likely to become importers than those in low exposure regions.

### Sector-level patterns of importing
- Figure 2a: Between 1986-1990, over 30% of importing establishments operated in the non-traded sector; this share grew to over half of all importers in 1997-2001.
- Figure 2b:
  - Intermediate inputs account for more than half of aggregate imports in Brazil.
  - Inputs from traded industries account for over 20% of total input costs in the non-traded sector.

- Fact 3. A majority of Brazil’s imports during the liberalization period were intermediate goods. At the same time, many importers operated in non-traded industries, using as inputs goods from industries that account for a reasonably large share of imports.

### Plant characteristics versus importing status (Table 4) — key estimates
- Regressions of plant outcomes on importing-status dummies (never importers = comparison group), controlling for industry and microregion dummies; sample restricted to plants that survived throughout the sample period.
- Estimates (standard errors clustered at microregion level):
  - Always importers (Importers_90):
    - ln(Empl): 1.612 *** (0.129)
    - ln(WB): 2.233 *** (0.134)
    - ln(Wage): 0.621 *** (0.029)
    - SkillShare: 0.184 *** (0.015)
  - New importers (NewImp_99):
    - ln(Empl): 1.187 *** (0.096)
    - ln(WB): 1.645 *** (0.097)
    - ln(Wage): 0.459 *** (0.031)
    - SkillShare: 0.120 *** (0.021)

- Interpretations:
  - Always importers (relative to never importers) employed on average 161% more employees, paid 62% higher wages, and had 18 percentage points more college-educated workers.
  - New importers were initially higher performing than never importers: employed on average 120% more workers, paid 46% higher wages, and had 12 percentage points more college-educated workers.

### Plant exit, growth, and importing behavior in the non-traded sector (Table 5) — key estimates
- Regressions include state, industry, initial employment quartiles, plant initial shares, log mean wage, 1986-1989 pre-trends, employment pre-period growth, initial skill intensity, sector-specific regional informal share and its evolution; standard errors clustered at microregion level.
- Outcomes and estimates:
  - Survived (90):
    - Importers_90: 0.095 *** (0.013) → Importers were 9.5 p.p. more likely to survive between 1990 and 1999 than non-importers.
  - ln(Empl) among survivors:
    - NewImp_99: 0.683 *** (0.027) → Employment at new importers grew 68% more than never importers’ employment.
  - dln(WB) among survivors:
    - NewImp_99: 0.843 *** (0.024)

- Interpretation:
  - New importers, largely from the middle of the size distribution, grew the most during the liberalization period and may drive regional reallocation effects.

### Causal effect of trade exposure on import participation (regression (4) and Table 6)
- Specification:
  - NewImp_e = α0 + θ·lib_r(e) + βX_e + α_s(e) + α_j(e) + α_b(e) + ε_e
  - Coefficient θ captures how much more likely non-importing plants in exposed labor markets are to become importers in 1999 compared to similar plants in low-exposure regions.
- Key result from Table 6 (example specification):
  - Coefficient: 0.033 (third column example)
    - Interpretation: An establishment whose initial region faced a 10 p.p. larger effective tariff decline had, on average, a 0.33 p.p. larger probability of becoming an importer.
    - This 0.33 p.p. increase corresponds to 48% of the overall share of new importers over the sample period.
  - The effect of import exposure on import adoption was more pronounced at initially larger plants in the fourth quartile of the regional employment distribution.

### Summary facts
- Fact 4. New importers were initially higher performing than producers that never imported, but lower performing than those already importing. However, they grew the most. At the same time, plants in regions more exposed to the import shock were more likely to adopt importing than comparable plants in low-exposure regions. Entry into imports in more exposed regions was more pronounced in initially larger plants.

### Conclusion — cohesive interpretation
- Local labor markets with stronger reductions in effective import tariffs during the liberalization experienced different worker reallocation patterns compared to less exposed markets.
- In the non-traded sector, workers reallocated systematically away from small plants to larger plants.
- Larger producers selected into importing intermediate inputs and expanded as they gained access to international inputs, consistent with selection and growth of new importers driving observed reallocation and growth patterns.

*Source: wpiea2023265-print-pdf - Appendix Figures A.3 and A.4 replicates the empirical exercises industry by industry. Additionally,*

### 3. A theoretical model of importing

### 3. A theoretical model of importing

### Framework
- Economy: mass L of identical households, a traded sector, and a non-traded sector composed of industries indexed by j.
- Households: Cobb-Douglas preferences over non-traded industries and the traded sector:
  - U = [∏_j (Q_j^β_j)^{β_j}] Q_T^{1−∑_j β_j}
  - Within industry j, varieties ω ∈ Ω_j aggregated with CES elasticity σ:
    - Q_j = [∫_{ω∈Ω_j} q_j(ω)^{(σ−1)/σ} dω]^{σ/(σ−1)}
  - Demand for a particular variety ω from industry j: β_j X (p_j(ω)/P_j)^{1−σ}, where P_j is the CES price index for industry j.
- Non-traded producers:
  - Continuum of monopolistically competitive producers; each characterized by productivity z and unique variety ω.
  - Production is Cobb-Douglas with intermediates weight α_j and labor weight 1−α_j.
  - Marginal cost for a producer with efficiency z: w^{1−α_j} m_j(z)^{α_j} / z, where w is common wage and m_j(z) depends on import participation.
    - Non-importer unit price of inputs: m^d_j.
    - Importer unit price of inputs: m_j (with m^d_j > m_j assumed).
  - Optimal pricing: constant markup σ/(σ−1) over marginal cost.
  - Variable profits (conditional on importer status):
    - π_j(z) ≡ (1/σ) β_j X [ w^{1−α_j} m_j(z)^{α_j} z P_j ]^{1−σ }   (Equation (5))
  - Import fixed cost: F^m_j; importer threshold z^m_j satisfies:
    - F^m_j = π_j(z^m_j, m_j) − π_j(z^m_j, m^d_j)   (Equation (6))
  - Participation fixed cost (overhead): F^n_j; participation cutoff z^n_j satisfies:
    - F^n_j = π_j(z^n_j, m^d_j)   (Equation (7))
  - Assumption: F^n_j and F^m_j such that z^n_j < z^m_j (some operating producers do not import).
- Entry and free entry:
  - Entry fixed cost F^e_j paid before drawing z from distribution G_j(z).
  - Free entry condition:
    - F^e_j = ∫_{z_n_j}^∞ [ π_j(z) − F^n_j − 1_{z>z^m_j} F^m_j ] dG_j(z)   (Equation (8))
  - Combining conditions yields relationships (ZZ) and (FE) characterizing z^n_j and z^m_j, including the importer advantage:
    - Δ_j ≡ (m^d_j / m_j)^{α_j} (the “importer advantage”)

### Import liberalization policy: mechanism and comparative statics
- Interpretation of liberalization:
  - Can be modeled as either an increase in relative variable importing benefits (m_j / m^d_j) or a decrease in relative fixed importing costs (F^m_j / F^n_j).
  - Both types of liberalization shift (ZZ) and (FE) upwards (see Figure 3 description).
- Predicted equilibrium responses to liberalization:
  - Participation cutoff z^n_j increases.
  - Importing cutoff z^m_j decreases.
  - Intuition: importing becomes more profitable on average; free entry pins down expected profits so non-importing activity must become less profitable → higher participation cutoff and lower importing cutoff.
- Extensive margin implications (Figure 4 summary):
  - Exit of the smallest non-importers (due to higher z^n_j).
  - Largest non-importers (middle of size distribution) begin importing (z^m_j falls) — these are “new importers”.
  - Continues to match empirical patterns: new importers of middling size and more likely where import shock larger.

### Intensive margin: labor demand and heterogenous responses
- Assumption for intensive margin analysis: fixed costs paid in labor units: F^n_j ≡ w f^n_j and F^m_j ≡ w f^m_j.
- Labor demand of a participating non-importer with efficiency z:
  - ℓ = (1−α_j) [ (σ−1) f^n_j (z / z^n_j)^{σ−1} + f^n_j ]   (Equation (9))
  - This decreases with the participation cutoff z^n_j.
  - Consequence: surviving non-importers (the smallest surviving producers) unambiguously shrink after liberalization.
- Labor demand of an incumbent importer with efficiency z:
  - ℓ = (1−α_j) [ (σ−1) f^n_j ( (z / z^n_j) (m^d_j / m_j)^{α_j} )^{σ−1} + f^n_j + f^m_j ]   (Equation (10))
  - Responds positively to increases in (m^d_j / m_j)^{α_j}, f^n_j, f^m_j and negatively to increases in z^n_j.
  - Ambiguity in general, but proved that if liberalization raises importer advantage (Δ′_j > Δ_j) while leaving fixed costs unchanged, labor demand at incumbent importers unambiguously increases.
- New importers:
  - Before policy: labor demand as non-importer (Equation (9)); after policy: follows Equation (10).
  - New importers experience larger relative increases in labor demand than continuing importers because their importer advantage grows from 1 to Δ′_j (whereas incumbent importers go from Δ_j to Δ′_j).
- Summary of intensive-margin predictions (under fixed production costs unchanged):
  - Continuing non-importers shrink.
  - New importers grow.
  - New importers grow by more than continuing importers.
  - These predictions align with empirical intensive margin results (new importers from middle of size distribution grew most in workforce).

### Welfare implications
- Real wage (representative household) expression:
  - w / ∏_j P_j^{β_j} (P_T)^{1−∑_j β_j} = ∏_j [ (w / P_j)^{β_j} (w / P_T)^{1−∑_j β_j} ]   (Equation (11))
- Industry-j real wage proportional change (from operating cutoff condition (7)):
  - (w / P_j)^̂ = d(w / X)^{1/(1−σ)} [ (w m^d_j) ]^{α} ẑ^n_j   (Equation (12) — as stated)
  - Interpretation of terms in (12):
    - First term: common economy-wide income effect (depends on the relative importance of labor income in expenditure).
    - Second term: industry-wide “import competition” effect via impact of import policy on price of domestically-produced traded goods.
    - Third term: novel within-industry productivity consequence from intra-industry reallocation — change in participation cutoff ẑ^n_j is a sufficient statistic for this reallocation welfare effect.
- Quantitative relevance: non-traded sector can be very large (nearly double the size of traded sector in Brazil), so within-industry reallocation channel can be quantitatively significant.

### Structural shift-share exposure and general equilibrium
- Derived a parsimonious general equilibrium version of the model to produce a theoretically consistent regional exposure measure.
  - Measure incorporates input-output linkages, producer-level selection into imports, and labor allocation across non-traded industries.
  - Qualitatively similar effects of trade liberalization on worker reallocation and import decisions as in the partial equilibrium and empirical findings.
  - Claimed novelty: first derivation of a shift-share regional exposure measure in models of heterogeneous firms with selection into international trade (Appendix C referenced for details).

### The role of selection (special no-selection case)
- Special case: importing frictionless with F^m_j = 0 ⇒ all active producers import (z^m_j = z^n_j).
  - Free entry simplifies to:
    - F^e_j / F^n_j = ∫_{z^n_j}^∞ [ (z / z^n_j)^{σ−1} − 1 ] dG(z)
  - Importer advantage absent; marginal producer z^n_j determined by free entry and not affected by changes in input import prices m_j.
  - Implications:
    - No change in productivity distribution of existing producers following import liberalization — no extensive-margin selection effects.
    - Labor demand of a z-efficiency producer:
      - ℓ = (1−α_j) [ (σ−1) f^n_j (z / z^n_j)^{σ−1} + f^n_j ]
      - Changes in import prices m_j have no effect on producer-level labor demand; adjustments occur only via industry-level entry (aggregate scale).
    - Welfare expression (Equation (12)) loses the within-industry reallocation term since ẑ^n_j = 1. Only economy-wide income effect and import-competition effect remain.
- Key message: differential importing decisions across producers (selection) are essential for generating reallocation among producers and the associated intensive-margin and within-industry welfare effects. In the full model, looser import policy benefits importers and new importers while non-importers are indirectly harmed through the free entry condition, necessitating redistribution from non-importers to importers.

*Source: 3. A theoretical model of importing (wpiea2023265-print-pdf).*

### 4. Conclusion

### 4. Conclusion

### Main findings on local labor-market adjustment
- Brazil’s import market was unilaterally liberalized in the early 1990s; local labor markets were exposed differently depending on initial industrial composition.
- The study uncovers a new margin of adjustment within local labor markets: labor re-location between producers in non-traded industries.
- Reallocation patterns in non-traded industries are qualitatively similar to those in traded industries: large producers in import-exposed areas are more likely to survive and grow than smaller equivalents in the same industry.
- Reduced-form evidence links importing behavior of large non-traded producers to these patterns:
  - Plants that were already importing before liberalization were more likely to survive.
  - Plants located in regions more exposed to the shock were more likely to become importers.
  - Newly importing producers were the ones that grew the most over the sample period.
- The observed reorganization and the importing channel imply a novel implication of importing for industrial productivity and concentration in non-traded industries.

### Model and mechanisms
- The paper proposes a model where non-traded producers self-select into importing intermediate inputs.
- Key mechanism: only a portion of non-traded producers select into international trade, generating differential exposure to trade shocks.
- As importing becomes cheaper:
  - Larger producers switch to imported intermediate inputs, become more productive, and compete away demand from low-productivity competitors.
  - The producer size distribution tilts: the smallest producers exit and the smallest survivors shrink, while survivors that switch importing status grow relatively the most.
- The model qualitatively reproduces the differential effects across producers of non-traded varieties observed in the data.

### Implications for productivity and market structure
- Importing by a subset of non-traded producers can raise productivity and increase concentration within non-traded industries via selection and reallocation.
- The importing channel operates through changes in intermediate input sourcing that disproportionately benefit larger producers who adopt imports.

*Source: wpiea2023265-print-pdf - 4. Conclusion*

### Section 3. The last equation relates traded goods prices at home and abroad to fundamentals and

### Section 3. The last equation relates traded goods prices at home and abroad to fundamentals and the domestic wage rate.

### Counterfactual analysis
- Setup and assumptions:
  - Study how the equilibrium wage changes when we shock the prices of foreign inputs, p∗i.
  - Assume that the productivity distribution in all non-traded industries are the same: Gj ∼ Pareto(θ).
- Main analytic result (effect of an import liberalization on regional wages):
  - dlnw =
    - ∑j∈NT αj 1−αj αj δj (1−δj) swb j
      ∑k∈NT αk 1−αk αk δk (1−δk) swb k
      |{z} regional weights
      ∑i∈T ηij · dlnp∗i |{z} effective tariff reduction
    - (Equation 14 reproduced in structure from source)
- Mechanisms captured:
  - Input-output linkages: international prices directly affect each non-traded industry through input-output shares ηij.
  - Dependence on industry wage-bill share: positive dependence on non-traded industry j’s share of wage bill in region r: swb jr.
  - Role of intermediates: increases with the Cobb-Douglas share coefficient for intermediaries, αj.
  - Importer share nonlinearity: inverse-U relationship with the regional share of importers in non-traded industry j: δjr (1−δjr), highlighting the importance of differential importing decisions across producers in a given industry-region.

### Empirical analysis (mapping structural shock to data)
- Mapping choices:
  - Assume dlnp∗i equals the observed industrial tariff cuts in Brazil.
  - α (materials share relative to labor) obtained directly from National Accounts.
  - δj (share of employment in non-importers relative to importers) and the regional distribution of wage bill, swb j, calculated directly from RAIS.
  - ηij (share of materials in each traded industry) calculated from the Brazilian Input-Output table.
  - Tariff changes for each industry obtained from Kume, Piani, and Souza (2003).
- Regional exposure measure used in regressions:
  - exposure r =
    - ∑j∈NT αj 1−αj αj δjr (1−δjr) swb jr
      ∑k∈NT αk 1−αk αk δkr (1−δkr) swb kr
      ∑i∈T ηij · dln(1+τi).

### Regression evidence and key findings
- Reallocation and selection:
  - Table C.1: Labor in the non-traded sector was reallocated, in response to the liberalization, towards larger producers in the middle of the initial size distribution.
  - Table C.2: Non-traded producers in more exposed labor markets were more likely to become importers in the post-liberalization period.
  - Combined with evidence on selection into imports from Tables 4 and 5 (referenced), these results suggest selection into imports as a mechanism explaining the reallocation of labor at the local labor market level.
- Table C.1 (differential effects of the trade liberalization on plants of different size: a model-based measure of exposure) — selected coefficient estimates and statistics:
  - Extensive margin / Intensive margin columns report θ and θk for k=2,...,4 for regressions using Zr = {libr, exposurer}.
  - Coefficients (with clustered standard errors at microregion level) include:
    - θ: -0.80 ∗∗∗ (0.28) and -0.30 (0.19) and -3.68 ∗∗∗ (1.11) and -1.19 ∗ (0.68) [presented across columns in table]
    - θ2: 0.62 ∗∗∗ (0.23), 0.43 ∗ (0.22), 2.43 ∗∗ (0.97), 1.38 (0.88)
    - θ3: 1.06 ∗∗∗ (0.33), 0.70 ∗∗ (0.33), 6.51 ∗∗∗ (1.23), 3.83 ∗∗∗ (0.91)
    - θ4: 0.09 (0.21), 0.03 (0.18), -1.93 ∗ (0.99), -1.88 ∗∗ (0.88)
  - Sample sizes: N 82519 4363200 (as displayed in table).
  - Notes: Regressions include controls and fixed effects as described in the table notes. Significance levels: ***p<1%, **p<5%, *p<10%.
- Table C.2 (effects of the trade liberalization on plants importing decision: a model-based measure for exposure) — selected coefficient estimates and statistics:
  - Columns (1), (2), (3) report θ coefficients for different specifications replacing libr with exposurer:
    - θ: 0.048*** (0.001) in column (1).
    - θ: 0.045*** (0.007) in column (2).
    - θ: 0.013** (0.005) in column (3).
  - Specification details:
    - Column (1) controls for industry and microregion fixed effects.
    - Column (2) additionally controls for the initial size bin of a plant, measured by its number of employees.
    - Column (3) includes the full set of controls discussed in Table 5.
  - Sample sizes: N 363199 for each column as reported.
  - Notes: Restrict attention to plants that survived throughout the sample period. Standard errors clustered at the microregion level. Significance levels: ***p<1%, **p<5%, *p<10%.

### Interpretation and implied policy-relevant insights
- Import liberalization affects regional wages through:
  - The structure of input-output linkages (ηij).
  - Industry composition and local wage-bill shares (swb jr).
  - Firm-level importing decisions and their distribution within industries (δjr).
  - The intensity of intermediate inputs in production (αj).
- Selection into importing post-liberalization is an important channel for local labor reallocation, with larger/mid-sized producers gaining employment share in non-traded sectors where exposure is higher.

*Source: Non-traded Gains from Trade: Selection in the Non-traded Sector: Evidence from Brazil — Working Paper No. WP/2023/265 (Section 3 and accompanying tables reproduced from the provided content).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023265-print-pdf.pdf_