## wpiea2020205-print-pdf

## Source details

**Canonical URL:** [wpiea2020205-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020205-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020205-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020205-print-pdf.pdf.json)

---

### Hiring from buyers and suppliers
- Section 3 documents that firms hire disproportionately from their buyers and suppliers.
- Observed share of movers hired by a firm that was buyer or supplier of previous employer (PC|L): 19.1 percent (766,294 movers).
- Including trades in previous or current year: 23 percent.
- Share of movers to same industry: 35 percent.
- Random-allocation counterfactual (PTrade, weighted): 1.9 percent on average; unweighted firm-level average PTrade_i = 0.2 percent.
- Odds ratio (data vs random allocation) for all movers: 13.

### Data
- Three administrative datasets combined: firm-level data, firm-to-firm transaction data, and employer-employee data from the Directorate General of Internal Taxes, the Directorate General of Customs, and the Social Security Treasury of the Dominican Republic.
- Firm-level universe: annual data for “juridical persons” between 2012 and 2017 from tax forms IR1, IR2, IT1, IR3; main industry (ISIC 3) and municipality recorded.
- Firm-to-firm purchases/sales: monthly purchases reported on tax form 606; interactions with informal-sector sellers may appear as “other expenditures.”
- Employer-employee data: monthly reported wages and worker attributes (age, gender, ethnicity); workers classified as permanent or temporary; sample restricts to firms with at least one permanent employee.
- Dataset overview (averages 2012–2017):
  - Average firms: 35,703
  - Share of firms both suppliers and buyers: 0.79
  - Share buyers only: 0.18
  - Share suppliers only: 0.03
  - Average workers: 1,638,263 (36 percent of country’s labor force)
  - Average annual transactions: 1,577,809
  - Average sales: 27,646 (millions of 2010 US dollars)
  - Sales as share of GDP: 0.40
  - Average sales per transaction: 0.018 (millions of 2010 US dollars)
- Network structure: average firm has 55 buyers and 45 suppliers; heavy right skew in connections and firm size; firms with many workers tend to be highly connected.

### Movers (measurement and sample)
- Definition: about 10 percent of workers report income from multiple firms in a year; assign each employee to the highest-paying employer within a year; a mover if highest-paying employer differs from previous year.
- Sample dynamics:
  - 15 percent of workers drop out of the sample year-to-year.
  - Observed worker moves between formal firms (2012–2017): 766,264 moves.
- Mover characteristics:
  - Mean age movers: 31; non-movers: 36.
  - Annual average (median) salary for an employee about to change job is about 64 percent (63 percent) higher than salary of an employee staying in same firm (holding age constant).
  - Female employees account for 36 percent of observations; probability of staying same firm next year: females 87 percent, males 83 percent.

### Workers move between buyers and suppliers — empirical patterns and heterogeneity
- Selected heterogeneity (probability vs random; number of movers):
  - Changing industry: probability 17.7 percent; random 1.9 percent; movers 196,620.
  - Same industry: probability 21.8 percent; random 3.3 percent; movers 169,674.
  - Changing municipality: probability 15.3 percent; random 1.5 percent; movers 356,096.
  - Same municipality: probability 22.5 percent; random 4.4 percent; movers 410,198.
  - Changing industry and municipality: probability 17.7 percent; random 0.8 percent; movers 196,620.
  - Same industry and municipality: probability 21.8 percent; random 6.2 percent; movers 169,674.
  - Within manufacturing sector: probability 17.5 percent; random 1.5 percent; movers 88,254.
  - Excluding large firms (500+ employees): probability 14.2 percent; random 0.5 percent; movers 302,423.
  - Large layoffs: probability 20.1 percent; random 0.7 percent; movers 19,323.
  - Cease to have permanent employees: probability 38.3 percent; random 1.6 percent; movers 19,323.
  - Excluding former coworkers (2016–2017 selection): probability 11.5 percent; random 1.3 percent; movers 78,252.
- Worker flows by origin firm size quintiles (probabilities and odds-ratios):
  - Prob. moving to connected firms by quintile: 1st = 9.6; 2nd = 11.0; 3rd = 11.8; 4th = 13.3; 5th = 20.0
  - Prob. randomly moving to connected firms by quintile: 1st = 0.1; 2nd = 0.1; 3rd = 0.1; 4th = 0.2; 5th = 2.3
  - Odds-ratios by quintile: 1st = 193; 2nd = 154; 3rd = 118; 4th = 79; 5th = 10
- Alternative random-allocation to job openings (100 simulations, conditional reshuffles):
  - All movers (no controls): observed 19.1 percent; random 7.1 percent; odds ratio 3.1; movers 766,294.
  - All movers (with controls): observed 19.5 percent; random 11.8 percent; odds ratio 1.8; movers 654,931.
- Firm-pair regressions (equation (3)):
  - Trade flow dummy (traded in t−1) positive and significant across specifications.
  - Coefficient magnitude range: 0.9 to 1.7 percentage points; associated odds ratio between 5.6 and 9.2.
  - Saturated specification: trade increases probability by 0.9 percentage point; unconditional weighted probability of labor connection ≈ 0.2 percentage points.
  - Distinguishing upstream vs downstream: both significant.
  - Excluding top 10 percent firms by workforce: coefficient 0.16 percentage points; odds-ratio ≈ 300 (domestic production networks more important for small and medium firms).

### Quality of firm-worker matches
- Match duration (workers who moved 2012→2013, observed up to 5 years):
  - Average duration if hiring firm was buyer/supplier: 2.8 years.
  - Average duration if hiring firm was not connected: 2.4 years.
  - Probability still at same firm in 2017: connected origin 29 percent; unconnected origin 21 percent.
  - Regression (equation (5)):
    - Unconditional: trade flow associated with 0.310 years longer duration (column 1).
    - Richest specification: trade flow associated with 0.097 years longer duration (column 3).
    - Same-firm-in-2017 dummy: unconditional +0.078; with worker controls +0.065; full controls +0.018 (columns 4–6).
  - Attrition hazard specification (equation (6)): matches between buyer-supplier firms have about 2 to 3 percentage point lower annual dissolution probability for several years.
- Wages (equation (7)):
  - Unconditional estimate: trade flow correlated with −0.018.
  - With worker controls: trade flow correlated with +0.196.
  - Full controls: trade flow correlated with +0.072 (implying additional wage growth of approximately 7 percent associated with moves to buyers/suppliers).
  - Persistence (equation (8), Figure 2): about a third of the wage increase persists for over 5 years.
- Productivity (equations (9)–(11), Table 8):
  - Destination-firm productivity (log revenue per permanent employee):
    - Unconditional: trade flow coefficient 0.286.
    - With controls: coefficient 0.096.
  - Productivity growth after hiring (conditioning on initial productivity):
    - Trade flow associated with positive productivity change: coefficients 0.106 (col 3) and 0.028 (col 4).
    - Column (4) coefficient equivalent to about 4.2 percent of one standard deviation of productivity growth and almost 50 percent of its mean.
  - Persistence (equation (11), Figure 3): productivity increase for firms hiring from buyers/suppliers persistent and significant for about four years.
  - Robustness: alternative specifications and firm-level regressions in Section A2 confirm findings.

### Human capital and knowledge transfers
- Wage quintile heterogeneity (Table 9, panel a):
  - Prob. of moving to connected firms by wage quintile: 1st quintile 11.6, 2nd quintile 14.6, 3rd quintile 19.4, 4th quintile 26.8, 5th quintile 31.8
  - Prob. of randomly moving to connected firms by quintile: 1.6, 1.6, 2.1, 3.6, 3.2
  - Odds-ratio by quintile: 8, 10, 11, 10, 14
  - Number of movers by quintile: 159,356; 212,719; 178,188; 121,055; 94,976
- Within manufacturing sector (Table 9, panel b):
  - Prob. of moving to connected firms by quintile: 8.4, 9.8, 13.2, 26.5, 42.7
  - Prob. of randomly moving to connected firms by quintile: 0.9, 1.0, 1.2, 2.4, 2.9
  - Odds-ratio by quintile: 10, 11, 13, 14, 25
  - Number of movers by quintile: 18,246; 24,294; 19,918; 12,917; 12,879
- Key implications:
  - Hiring from buyers and suppliers is much more common for high-salary workers; share of workers moving to connected firms and odds-ratio increase across wage quintiles.
  - Within manufacturing, probabilities reach 42.7 percent for top wage quintile.

### Spillovers from more productive firms (hiring and productivity)
- Define "productivity gap" = max(previous employer productivity − current employer productivity, 0).
- Table 10 selected coefficients:
  - Column (1): Trade flow 0.027*** (0.006); Productivity gap 0.097*** (0.024); Observations 620,755; R2 0.379
  - Column (2): Trade flow 0.009* (0.005); Productivity gap 0.090*** (0.025); Productivity gap×trade flow 0.036*** (0.006); Observations 620,755; R2 0.380
  - Column (3): Trade flow −0.016*** (0.005); Productivity gap 0.130*** (0.049); Productivity gap×trade flow 0.047** (0.019); Observations 261,409; R2 0.478
  - Column (4): Trade flow 0.026*** (0.007); Productivity gap 0.074*** (0.016); Productivity gap×trade flow 0.022*** (0.007); Observations 351,291; R2 0.345
- Interpretation:
  - Firms’ productivity increases when they hire workers from more productive firms.
  - Knowledge spillovers are about 40% larger when a worker is hired from a high-productivity firm that is also a buyer or supplier.
  - Spillovers present both within and across 3-digit industries.

### Coworker learning (wage effects of new hires)
- Estimation (equation (13)) in firms with at most 100 workers; selected coefficients (Table 11):
  - Ln wage persistence: 0.645*** (1 year), 0.583*** (2 years), 0.502*** (3 years)
  - New hire effect (D_nh): 0.028*** (1 year), 0.055*** (2 years), 0.047*** (3 years)
  - New hire from buyer or supplier (D_nh*): 0.057*** (1 year), 0.061*** (2 years), 0.079*** (3 years)
  - New hire from buyer: 0.028***, 0.034***, 0.055***
  - New hire from supplier: 0.054***, 0.052***, 0.055***
  - Observations: 1,378,117 (1 year); 892,476 (2 years); 533,509 (3 years)
  - R2: 0.429, 0.353, 0.276
- Key magnitudes and patterns:
  - Wage growth for employees at firms that hired is 2.7 percent higher in year 1.
  - When the new hire is from a buyer or supplier, coworker wage increase is 8.3 percent in year 1 (5.6 percentage points more than hires from outside the production network).
  - Coworker wage differential grows over time: 6 percent higher in year 2 and 7.7 percent higher in year 3 when hire from production network.
  - Hiring from suppliers yields larger immediate coworker wage gains (5.4 percentage points higher) than hiring from buyers (2.8 percentage points higher).

### Complementarities between inputs and human capital
- For firm pairs that traded in 2012, regress 2017 trade measures on a dummy for any worker moving from supplier to buyer between 2012 and 2017 (equation 14).
- Table 12 selected results:
  - Any trade (column 1): Worker flow coefficient 0.036*** (0.003)
    - Hiring from supplier increases probability of still trading in 2017 by 3.6 percentage points (average 46 percent).
  - Share of purchase (column 2): Worker flow coefficient 0.001*** (0.000)
  - Share of sales (column 4): Worker flow coefficient 0.003*** (0.000)
  - Observations range: 475,555 to 1,101,114; R2 range 0.369–0.507
- Interpretation:
  - Buyers are more likely to continue buying from suppliers they hire from, and purchase a larger share of inputs from them.
  - Productivity gains from hiring a worker from a supplier increase with the supplier’s share in the buyer’s purchases (Figure 4 summary); hires from top-quartile suppliers produce about 3 times larger initial impacts than hires from bottom-quartile suppliers.

### Information frictions and selection
- Hypothesis: information frictions and referrals could explain hiring from buyers/suppliers by reducing noise/signal problems.
- Evidence/tests:
  - Firms that hire from trading partners show stronger trade links afterwards, which information frictions alone do not explain.
  - Restricted sample excluding prior employee moves (no moves 2012–2016, examine 2016–2017 flows conditional on 2016 trading): observed share moving to buyer/supplier 11.5 percent; random allocation 1.3 percent; odds ratio 10.
- Conclusion: ex-coworker networks and referral-based information frictions cannot fully explain the main patterns.

### Conclusion (overarching findings)
- Main novel fact: workers disproportionately move to the buyer or the supplier of their previous employer when changing jobs; result robust to controls for industry, municipality, firm size, layoffs, firm closures, and excluding firms with previous/current coworkers.
- Employment and wage implications:
  - Being hired from a trading partner is associated with lower separation rates and higher wage growth for the hired worker.
  - Hiring from trading partners is associated with higher hiring-firm productivity growth, interpreted as higher match quality and knowledge transfers.
  - The propensity to move to connected firms is larger for higher-wage workers.
  - Coworkers of firms that hire from buyers/suppliers experience faster wage growth, consistent with coworker learning.
- Trade and firm-boundary implications:
  - Connected firms strengthen commercial relationships when a supplier’s employee is hired by a buyer, arguing against insourcing as the main motive.
  - Findings support complementarities between human capital and inputs along production networks as a mechanism for sizable knowledge transfers.

*Source: wpiea2020205-print-pdf - Section 3 documents that firms hire disproportionately from their buyers and suppliers.*

### Section 3 documents that firms hire disproportionately from their buyers and suppliers.

### wpiea2020205-print-pdf - Section 3 documents that firms hire disproportionately from their buyers and suppliers.

### Hiring from buyers and suppliers
- Section 3 documents that firms hire disproportionately from their buyers and suppliers.

### Quality of employer-employee matches along the domestic production network
- Section 4 analyzes the quality of the employer-employee matches formed along the domestic production network.

### Importance of human capital and knowledge transfers
- Section 5 presents the evidence on the importance of human capital and knowledge transfers to explain worker movements to buyers and suppliers.

*Source: wpiea2020205-print-pdf - Section 3 documents that firms hire disproportionately from their buyers and suppliers.*

### Section 6 proposes alternative explanations for our findings. Section 7 concludes.

### WORKER MOBILITY AND DOMESTIC PRODUCTION NETWORKS

### Data
- Three administrative datasets combined: firm-level data, firm-to-firm transaction data, and employer-employee data from the Directorate General of Internal Taxes, the Directorate General of Customs, and the Social Security Treasury of the Dominican Republic.
- Firm-level universe: annual data for “juridical persons” between 2012 and 2017 from tax forms IR1, IR2, IT1, IR3; main industry (ISIC 3) and municipality recorded.
- Firm-to-firm purchases/sales: monthly purchases reported on tax form 606; interactions with informal-sector sellers may appear as “other expenditures.”
- Employer-employee data: monthly reported wages and worker attributes (age, gender, ethnicity); workers classified as permanent or temporary; sample restricts to firms with at least one permanent employee.
- Dataset overview (averages 2012–2017):
  - Average firms: 35,703
  - Share of firms both suppliers and buyers: 0.79
  - Share buyers only: 0.18
  - Share suppliers only: 0.03
  - Average workers: 1,638,263 (36 percent of country’s labor force)
  - Average annual transactions: 1,577,809
  - Average sales: 27,646 (millions of 2010 US dollars)
  - Sales as share of GDP: 0.40
  - Average sales per transaction: 0.018 (millions of 2010 US dollars)
- Network structure: few large well-connected firms and many small sparsely connected firms; average firm has 55 buyers and 45 suppliers (marginally more buyers than suppliers); heavy right skew in connections and firm size; firms with many workers tend to be highly connected.

### Movers (worker mobility measurement and sample)
- Definition: about 10 percent of workers report income from multiple firms in a year; assign each employee to the highest-paying employer within a year; a mover if highest-paying employer differs from previous year.
- Sample dynamics:
  - 15 percent of workers drop out of the sample year-to-year.
  - Observed worker moves between formal firms (2012–2017): 766,264 moves.
- Mover characteristics:
  - Mean age movers: 31; non-movers: 36.
  - Annual average (median) salary for an employee about to change job is about 64 percent (63 percent) higher than salary of an employee staying in same firm (holding age constant).
  - Female employees account for 36 percent of observations; probability of staying same firm next year: females 87 percent, males 83 percent.

### Workers Move Between Buyers and Suppliers — Empirical Patterns
- Empirical share and counterfactuals:
  - Observed share of movers hired by firm that was buyer or supplier of previous employer (PC|L): 19.1 percent (766,294 movers).
  - Including trades in previous or current year: 23 percent.
  - Share of movers to same industry: 35 percent.
  - Random-allocation counterfactual (PTrade, weighted): 1.9 percent on average; unweighted firm-level average PTrade_i = 0.2 percent.
  - Odds ratio (data vs random allocation) for all movers: 13.
- Heterogeneity (selected results from Table 2):
  - Changing industry: probability 17.7 percent; random 1.9 percent; movers 196,620.
  - Same industry: probability 21.8 percent; random 3.3 percent; movers 169,674.
  - Changing municipality: probability 15.3 percent; random 1.5 percent; movers 356,096.
  - Same municipality: probability 22.5 percent; random 4.4 percent; movers 410,198.
  - Changing industry and municipality: probability 17.7 percent; random 0.8 percent; movers 196,620.
  - Same industry and municipality: probability 21.8 percent; random 6.2 percent; movers 169,674.
  - Within manufacturing sector: probability 17.5 percent; random 1.5 percent; movers 88,254.
  - Excluding large firms (500+ employees): probability 14.2 percent; random 0.5 percent; movers 302,423.
  - Large layoffs: probability 20.1 percent; random 0.7 percent; movers 19,323.
  - Cease to have permanent employees: probability 38.3 percent; random 1.6 percent; movers 19,323.
  - Excluding former coworkers (2016–2017 selection): probability 11.5 percent; random 1.3 percent; movers 78,252.
- Worker flows by origin firm size quintiles (Table 3):
  - Prob. moving to connected firms by quintile: 1st = 9.6; 2nd = 11.0; 3rd = 11.8; 4th = 13.3; 5th = 20.0
  - Prob. randomly moving to connected firms by quintile: 1st = 0.1; 2nd = 0.1; 3rd = 0.1; 4th = 0.2; 5th = 2.3
  - Odds-ratios by quintile: 1st = 193; 2nd = 154; 3rd = 118; 4th = 79; 5th = 10
- Alternative random-allocation to job openings (100 simulations, conditional reshuffles):
  - All movers (no controls): observed 19.1 percent; random 7.1 percent; odds ratio 3.1; movers 766,294.
  - All movers (with controls): observed 19.5 percent; random 11.8 percent; odds ratio 1.8; movers 654,931.
  - Results robust across subgroups; random-allocations to openings produce higher counterfactual shares than firm-level randomization, providing upper and lower bounds.
- Firm-pair regressions (equation (3)):
  - Dependent variable: dummy if any employee of origin firm in year t−1 works at destination firm in year t.
  - Trade flow dummy (traded in t−1) positive and significant across specifications.
  - Coefficient magnitude range: 0.9 to 1.7 percentage points; associated odds ratio between 5.6 and 9.2.
  - Saturated specification (firm-pair controls): trade increases probability by 0.9 percentage point; unconditional weighted probability of labor connection ≈ 0.2 percentage points.
  - Distinguishing upstream vs downstream: both significant.
  - Excluding top 10 percent firms by workforce: coefficient 0.16 percentage points; odds-ratio ≈ 300 (domestic production networks more important for small and medium firms).

### Quality of Firm-Worker Matches
- Match duration (workers who moved 2012→2013, observed up to 5 years):
  - Average duration if hiring firm was buyer/supplier: 2.8 years.
  - Average duration if hiring firm was not connected: 2.4 years.
  - Probability still at same firm in 2017: connected origin 29 percent; unconnected origin 21 percent.
  - Regression (equation (5)) results:
    - Unconditional: trade flow associated with 0.310 years longer duration (column 1).
    - Richest specification: trade flow associated with 0.097 years longer duration (column 3).
    - Same-firm-in-2017 dummy: unconditional +0.078; with worker controls +0.065; full controls +0.018 (columns 4–6).
  - Attrition hazard specification (equation (6)): matches between buyer-supplier firms have about 2 to 3 percentage point lower annual dissolution probability for several years.
- Wages (equation (7)):
  - Unconditional estimate: trade flow correlated with -0.018 (negative).
  - With worker controls: trade flow correlated with +0.196.
  - Full controls: trade flow correlated with +0.072 (implying additional wage growth of approximately 7 percent associated with moves to buyers/suppliers).
  - Persistence (equation (8), Figure 2): wage growth advantage persists over time; about a third of increase persists for over 5 years.
- Productivity (equations (9)–(11), Table 8):
  - Destination-firm productivity (log revenue per permanent employee):
    - Unconditional: trade flow coefficient 0.286.
    - With controls: coefficient 0.096 (workers moving to buyers/suppliers end up at firms with higher productivity).
  - Productivity growth after hiring (conditioning on initial productivity):
    - Trade flow associated with positive productivity change: coefficients 0.106 (col 3) and 0.028 (col 4).
    - Column (4) coefficient equivalent to about 4.2 percent of one standard deviation of productivity growth and almost 50 percent of its mean.
  - Persistence (equation (11), Figure 3): productivity increase for firms hiring from buyers/suppliers persistent and significant for about four years.
  - Robustness: alternative specifications and firm-level regressions in Section A2 confirm findings.

### Human Capital and Knowledge Transfers
- Wage quintile heterogeneity (Table 9, panel a summary):
  - Importance of production network increases with worker wage quintile.
  - Probability of moving to buyer or supplier by wage quintile:
    - Highest wage quintile: 32 percent (compared with 3.2 percent probability of ending in any firm of the production network under random matching for the same quintile in the random benchmark discussion).
  - Implication: higher-salary (higher human capital) workers more likely to move along buyer-supplier links, consistent with role of human capital and knowledge transfer in network-mediated mobility.

*Source: wpiea2020205-print-pdf - Section 6 proposes alternative explanations for our findings. Section 7 concludes.*

### 1.6 percent for the lowest wage quintile.  Thus, for workers in the highest wage quintile,

### wpiea2020205-print-pdf - 1.6 percent for the lowest wage quintile.  Thus, for workers in the highest wage quintile,

### Worker flows by wage
- Panel a (All moves) — Table 9:
  - Prob. of moving to connected firms by wage quintile: 1st quintile 11.6, 2nd quintile 14.6, 3rd quintile 19.4, 4th quintile 26.8, 5th quintile 31.8
  - Prob. of randomly moving to connected firms by quintile: 1.6, 1.6, 2.1, 3.6, 3.2
  - Odds-ratio by quintile: 8, 10, 11, 10, 14
  - p-value for equality test: 0.000, 0.000, 0.000, 0.000, 0.000
  - Number of movers by quintile: 159,356; 212,719; 178,188; 121,055; 94,976
- Panel b (Within manufacturing sector) — Table 9:
  - Prob. of moving to connected firms by quintile: 8.4, 9.8, 13.2, 26.5, 42.7
  - Prob. of randomly moving to connected firms by quintile: 0.9, 1.0, 1.2, 2.4, 2.9
  - Odds-ratio by quintile: 10, 11, 13, 14, 25
  - p-value for equality test: 0.000, 0.000, 0.000, 0.000, 0.000
  - Number of movers by quintile: 18,246; 24,294; 19,918; 12,917; 12,879
- Key findings:
  - For workers in the highest wage quintile, the chances of moving to a connected firm are 14 times larger than the probability of moving to a random firm in the production network (panel a).
  - Within manufacturing, probabilities reach 43 percent for top wage quintile and are five times higher than for bottom quintile; odds-ratio up to 25 for top quintile.
  - Number of movers is lower at higher quintiles, indicating high-paying jobs have lower turnover.
  - Conclusion: hiring from buyers and suppliers is much more common for high-salary workers; share of workers moving to connected firms and odds-ratio increase across wage quintiles, linking production-network role to human capital.

### Spillovers from more productive firms (hiring and productivity)
- Concept:
  - Define "productivity gap" = max(previous employer productivity − current employer productivity, 0).
  - Estimate specification (equation 12) where ∆prod is destination firm productivity growth and interest is ψ on TF × gap.
- Table 10 (selected coefficients and sample info):
  - Column (1): Trade flow 0.027*** (0.006); Productivity gap 0.097*** (0.024); Observations 620,755; R2 0.379
  - Column (2): Trade flow 0.009* (0.005); Productivity gap 0.090*** (0.025); Productivity gap×trade flow 0.036*** (0.006); Observations 620,755; R2 0.380
  - Column (3): Trade flow −0.016*** (0.005); Productivity gap 0.130*** (0.049); Productivity gap×trade flow 0.047** (0.019); Observations 261,409; R2 0.478
  - Column (4): Trade flow 0.026*** (0.007); Productivity gap 0.074*** (0.016); Productivity gap×trade flow 0.022*** (0.007); Observations 351,291; R2 0.345
- Interpretation:
  - Firms’ productivity increases when they hire workers from more productive firms.
  - Knowledge spillovers are about 40% larger when a worker is hired from a high-productivity firm that is also a buyer or supplier (column comparisons).
  - Spillovers are present both within and across 3-digit industries.

### Coworker learning (wage effects of new hires)
- Setup:
  - Estimate equation (13) where coworker wage growth depends on whether firm hired someone (D_nh) and whether hire was from buyer/supplier (D_nh*); controls include gender and age deciles; sample restricted to firms with at most 100 workers.
- Table 11 (selected coefficients):
  - Ln wage persistence: 0.645*** (1 year), 0.583*** (2 years), 0.502*** (3 years)
  - New hire effect (D_nh): 0.028*** (1 year), 0.055*** (2 years), 0.047*** (3 years)
  - New hire from buyer or supplier (D_nh*): 0.057*** (1 year), 0.061*** (2 years), 0.079*** (3 years)
  - New hire from buyer: 0.028***, 0.034***, 0.055***
  - New hire from supplier: 0.054***, 0.052***, 0.055***
  - Observations: 1,378,117 (1 year); 892,476 (2 years); 533,509 (3 years)
  - R2: 0.429, 0.353, 0.276
- Key magnitudes and patterns:
  - Wage growth for employees at firms that hired is 2.7 percent higher in year 1.
  - When the new hire is from a buyer or supplier, coworker wage increase is 8.3 percent in year 1 (i.e., 5.6 percentage points more than hires from outside the production network).
  - Coworker wage differential grows over time: 6 percent higher in year 2 and 7.7 percent higher in year 3 when hire from production network.
  - Hiring from suppliers yields larger immediate coworker wage gains (5.4 percentage points higher) than hiring from buyers (2.8 percentage points higher), though difference narrows over time.
- Robustness:
  - Results hold with additional controls (firm employment, employment growth, average coworker wage), broadly in full sample though weaker with full controls, and invariant to replacing log wages with wage growth.

### Complementarities between inputs and human capital
- Empirical test (equation 14):
  - For firm pairs that traded in 2012, regress measures of 2017 trade on a dummy for any worker moving from supplier to buyer between 2012 and 2017, controlling for firm fixed effects and firm-pair controls.
- Table 12 (selected results):
  - Any trade (column 1): Worker flow coefficient 0.036*** (0.003)
    - Interpretation: hiring from supplier increases probability of still trading in 2017 by 3.6 percentage points (about less than a tenth of average 46 percent).
  - Share of purchase (column 2): Worker flow coefficient 0.001*** (0.000)
  - Share of sales (column 4): Worker flow coefficient 0.003*** (0.000)
  - Additional columns condition on trading in 2017; results show increases in buyer’s share of purchases from that supplier.
  - Observations: 1,101,114; 1,038,014; 475,555; 1,101,114; 475,555 across columns; R2 range 0.369–0.507
- Interpretation:
  - Buyers are more likely to continue buying from suppliers they hire from, and purchase a larger share of inputs from them.
  - Evidence suggests complementarities between human capital and inputs: the knowledge to use an input increases demand for that input after worker transitions.
- Additional test (equation 15 / Figure 4 summary):
  - Productivity gains from hiring a worker from a supplier increase with the supplier’s share in the buyer’s purchases.
  - Example: initial impact of hiring from a supplier in the top quartile (β0 + ψ0·6.7%) is about 3 times larger than hiring from a supplier in bottom quartile (β0 + ψ0·0.05%).

### Information frictions and selection
- Hypothesis:
  - Information frictions and referrals could lead firms to hire workers from buyers/suppliers even if intrinsic match quality is not higher, by reducing noise/signal problems and enabling referrals.
- Evidence and tests:
  - Firms that hire from trading partners show stronger trade links afterwards, which information frictions alone do not explain.
  - To limit role of ex-coworker referrals, the authors analyze firm pairs that saw no employee moves from 2012–2016 and then examine 2016–2017 flows conditional on 2016 trading.
    - In that restricted sample: observed share moving to buyer/supplier 11.5 percent; random allocation 1.3 percent; odds ratio 10.
  - Conclusion: ex-coworker networks and referral-based information frictions cannot fully explain the main patterns.

### Conclusion (summary of overarching findings)
- Main novel fact: workers disproportionately move to the buyer or the supplier of their previous employer when changing jobs; robust to controls for industry, municipality, firm size, layoffs, firm closures, and excluding firms with previous/current coworkers.
- Employment and wage implications:
  - Being hired from a trading partner is associated with lower separation rates and higher wage growth for the hired worker.
  - Hiring from trading partners is associated with higher hiring-firm productivity growth, interpreted as higher match quality and knowledge transfers.
  - The propensity to move to connected firms is larger for higher-wage workers.
  - Coworkers of firms that hire from buyers/suppliers experience faster wage growth, consistent with coworker learning.
- Trade and firm boundary implications:
  - Connected firms strengthen commercial relationships when a supplier’s employee is hired by a buyer, arguing against insourcing as the main motive.
  - Findings support complementary human capital and inputs along production networks as a mechanism for sizable knowledge transfers.

*Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020205-print-pdf.pdf*

### References

### References

### Cited works
- Acemoglu, Daron, Vasco M. Carvalho, Asuman Ozdaglar, and Alireza Tahbaz-Salehi, “The Network Origins of Aggregate Fluctuations,”Econometrica, 2012,80(5).
- Alfaro-Urena, Alonso, Isabela Manelici, and Jos
 ́
e P V
 ́
asquez, “The Effects of Joining Multinational Supply Chains: New Evidence from Firm-to-Firm Linkages,” 2019. Available at SSRN 3376129.
- ,, and Jose P Vasquez, “The Effects of Multinationals on Workers: Evidence from Costa Rica,”

*Source: wpiea2020205-print-pdf - References*

### 2019. Working Paper, Department of Economics, UC Berkeley.

### 2019. Working Paper, Department of Economics, UC Berkeley.

### A1A Description of the Domestic Production Network — Overview
- Presents stylized facts for the domestic production network of the Dominican Republic: supplier-buyer connections, employer-employee network, and relation to firms’ turnover.
- Analysis spans years 2012–2017 with emphasis on 2017 as the last year of the sample.

### Suppliers and Buyers (A1.1)
- 2017 averages and dispersion:
  - Average supplier sold to 57 buyers, with a standard deviation of 364 buyers.
  - Average buyer bought from 48 suppliers, with a standard deviation of 68 suppliers.
  - Median supplier had 9 buyers; 99th percentile supplier had 769 buyers.
  - Median buyer had 30 suppliers; 99th percentile buyer had 311 suppliers.
- Distributional characteristics:
  - Both buyers-per-supplier and suppliers-per-buyer distributions are right skewed with concentration in a few firms.
  - About one fourth of suppliers have only 3 buyers; one fourth of buyers have only 14 suppliers.
  - Concentration increased over time; the top 1 percent of suppliers saw an increase of about 15 percent in the number of buyers over 5 years.
  - Only 1 percent of firms has more than 300 connections; only one hundredth of 1 percent have more than 1,100 connections.
- Pareto distribution parameter estimates (per-firm suppliers and per-firm customers):
  - Authors’ estimates: -0.30 and -0.43, respectively.
  - Comparative mentions: Costa Rica (Alfaro-Urena et al. (2018)) -0.58 for buyers and -0.73 for suppliers; Japan (Bernard et al. (2019a)) -1.50 for buyers and -1.32 for suppliers.
- Assortativity findings:
  - Suppliers: negative assortativity — a supplier with many buyers tends to be connected to buyers that buy from only a few suppliers.
    - Linear regression: a 10 percent increase in the supplier’s number of buyers is associated with a 1.4 reduction in the average number of suppliers of those buyers.
  - Buyers: negative assortativity — a buyer with many suppliers tends to be connected to suppliers that sell to only a few buyers.
    - Linear regression: a 10 percent increase in the buyer’s number of suppliers is associated with a 2.6 percent reduction in the average number of buyers of those suppliers.
  - Interpretation: dependence of small buyers on large suppliers and of small suppliers on large buyers; relationship for buyers is concave with stronger dependence for some ranges.
- Cross-sector heterogeneity (2017):
  - Highest average number of buyers per supplier: “finance and insurance” (mean 519), followed by “wholesale and retail” (mean 844) and “manufacturing” (mean 712).
  - Least concentrated industries: “health” (mean 123), “education” (mean 133), and “construction” (mean 166).
  - Sectoral distributions remain heavily right skewed.

- Table A1 (selected entries, Units: number of counterparties by year):
  - Number of buyers per supplier, 2017: Mean 57, St. Dev. 364, 10th 1, 25th 3, 50th 9, 75th 28, 90th 94, 99th 769.
  - Number of suppliers per buyer, 2017: Mean 48, St. Dev. 68, 10th 1, 25th 2, 50th 30, 75th 59, 90th 105, 99th 311.

### Workers / Employer-Employee Network (A1.2)
- Firm-level employment (2012–2017):
  - Number of workers per firm fluctuated between 45 and 50 (mean), with an average standard deviation of 268 workers across years.
  - 2017 specifics:
    - Mean number of workers per firm: 46.
    - Of these, 28 with permanent contracts and 18 with temporary contracts.
    - Standard deviation in 2017: 313 workers.
    - Median firm had 11 workers; 25th percentile firms had a maximum of 5 workers.
    - Firms at the top 1 percent had a minimum of 608 employees.
  - Distributional facts:
    - 3.8 percent of firms had one single employee in 2017.
    - 0.1 percent of firms had over 2,800 workers.
- Relationship between connections and employment (2017):
  - Positive relationship between number of connections and number of workers.
  - Linear regression estimates:
    - A 10 percent increase in the number of connections is associated with a 4.1 percent increase in the number of buyers.
    - A 10 percent increase in the number of connections is associated with a 7.8 percent increase in the number of suppliers.
  - Relationships are convex; marginal effect of additional connections is larger when initial connections are already large, especially for number of suppliers.
- Worker assortativity by firm size:
  - Suppliers: mildly negative — a 10 percent increase in suppliers’ workforce is associated with a 1.1 percent decline in the average number of workers of their buyers.
  - Buyers: negative — a 10 percent increase in buyers’ workforce is associated with a 1.6 percent decline in the average number of workers of their suppliers.
  - Interpretation: small suppliers depend on large buyers; small buyers depend on large suppliers.
- Cross-sector employment heterogeneity (2017):
  - Sectors with highest average workers per firm: “finance and insurance” mean 658; “manufacturing” mean 110.
  - Lowest average workers per firm: “real estate, renting, and business activities” mean 19; “education” mean 31.
  - Concentration example: manufacturing top 1 percent firm employs 1,946 workers (17.7 times the average firm in the sector); education top 1 percent employs 6.6 times the average firm in that sector.
- Table A3 (Number of Workers per Firm, Units):
  - 2017: Mean 46, St. Dev. 313, 10th 1, 25th 2, 50th 11, 75th 27, 90th 46, 99th 608.
- Table A4 (2017 workers by sector, Units) — selected entries:
  - Agriculture, hunting, and forestry: Mean 71, St. Dev. 160, 10th 3, 25th 6, 50th 17, 75th 64, 90th 177, 99th 766.
  - Finance and insurance: Mean 658, St. Dev. 1,597, 10th 4, 25th 12, 50th 72, 75th 394, 90th 2,154, 99th 7,782.
  - Manufacturing: Mean 110, St. Dev. 571, 10th 3, 25th 18, 50th 57, 75th 193, 90th 1,946.

### Sales (A1.3)
- Aggregate and firm-level sales trends (2012–2017):
  - Domestic sales increased by 7.6 percent from 2012 to 2017, reaching US$28.6 billion in 2017.
  - Average turnover fell by 16 percent to US$0.9 million; standard deviation declined by 38 percent to US$14.7 million.
  - 2017 percentiles:
    - 25th percentile firms sold less than US$0.01 million.
    - 99th percentile firms sold at least US$11.9 million.
  - Interpretation: top 1 percent turnover broadly unchanged; declines concentrated among other firms.
- Connections and sales (2017):
  - Positive correlation between number of connections and sales.
  - Linear regression estimates:
    - A 10 percent increase in the number of suppliers is associated with an 8.8 percent increase in sales.
    - A 10 percent increase in the number of buyers is associated with a 14.2 percent increase in sales.
  - Nonlinearities: slope becomes steeper for larger numbers of connections; firms with many connections have disproportionally higher sales.
- Geographic presence and sales:
  - Positive relationship between number of municipalities served and sales.
  - Linear regression estimates:
    - A 10 percent increase in the number of municipalities served by suppliers is associated with a 14.0 percent rise in sales.
    - A 1 percent increase in the number of municipalities from which firms buy inputs is associated with a 21.9 percent rise in sales.
- Intensive vs. extensive margin of sales:
  - Predicted percentiles (10th, 50th, 90th) of suppliers’ sales as function of number of buyers are broadly parallel for most values, suggesting number of buyers is generally not the main determinant of turnover; however, for firms with a very large number of buyers the number of buyers becomes relevant and nonlinearities emerge.
- Sectoral concentration of sales (Table A6, shares of domestic sales, Percent):
  - Manufacturing and wholesale and retail trade jointly accounted for 70.7 percent of domestic sales in 2017.
  - 2017 sectoral shares (selected): Manufacturing 37.0 percent; Wholesale and retail trade 33.7 percent; Finance and insurance 3.5 percent.
- Sales by sector (2017, Millions of 2010 US dollars) — selected entries (Table A7):
  - Manufacturing: Total 9,861; Mean 3.036; St. Dev. 41.506; 10th 0.003; 25th 0.018; 50th 0.091; 75th 0.475; 90th 2.258; 99th 41.595.
  - Wholesale and retail trade: Total 9,944; Mean 0.778; St. Dev. 8.295; 10th 0.002; 25th 0.009; 50th 0.051; 75th 0.230; 90th 0.872; 99th 10.669.
  - Finance and insurance: Total 1,079; Mean 8.236; St. Dev. 28.416; 10th 0.003; 25th 0.017; 50th 0.197; 75th 1.803; 90th 16.153; 99th 132.589.
- Firm-to-firm transaction values by sector (2017, Millions of 2010 US dollars) — Table A8 selected entries:
  - Agriculture, hunting, and forestry: Mean 0.049, St. Dev. 0.341, 10th 0.000, 25th 0.000, 50th 0.002, 75th 0.011, 90th 0.055, 99th 0.952.
  - Hotels and restaurants: Mean 0.004, St. Dev. 0.129, 10th 0.000, 25th 0.000, 50th 0.000, 75th 0.000, 90th 0.002, 99th 0.043.
  - Manufacturing: Mean 0.043, St. Dev. 1.481, 10th 0.000, 25th 0.000, 50th 0.001, 75th 0.006, 90th 0.026, 99th 0.439.

### A2 Alternative Productivity Specifications — Main Results
- Purpose: test robustness of baseline results linking worker movements along production network to productivity growth using alternative specifications and levels.
- Alternative specification using productivity level (equation (16)):
  - Specification: prod_{d,t+k} = φ_{o,k} + ρ_k · prod_{d,t−1} + β_k TF_{d,o,t−1} + γ X_{d,o,t−1} + δ_k X_{w,t−1} + η_{w,d,o,k}.
  - Finding: β_k estimates (Figure A11) confirm Section 4.3: firms that hire workers from their buyer and supplier are more productive to start with and become more productive after the hire, conditional on initial productivity.
- Knowledge spillovers using productivity level:
  - Unreported results confirm positive productivity spillovers are larger when firms hire workers of highly productive buyers or suppliers.
- Firm-level regressions (equation (17)):
  - Specification: ∆prod_{i,t+k} = ρ_k · prod_{i,t−1} + β_k HN_{i,t−1} + δ_k X_{i,t−1} + η_{i,t,k}.
  - HN_{i,t−1} = share of new hires that came from connected firms; X_{i,t−1} controls: calendar year, firm industry, location, size (number of permanent employees), number of new hires; observations weighted by number of new hires.
  - Results (Figure A12): firms that hire from buyers or suppliers experience significantly stronger productivity growth, and the additional growth persists over time. Results hold when focusing on firms that hired only one worker during the year.
- Firm-level knowledge spillovers with productivity gap (equation (18)):
  - Specification: ∆prod_{i,t+k} = ρ_k · prod_{i,t−1} + β_k HN_{i,t−1} + τ_k gap_{i,t−1} + ψ_k gapconnected_{i,t−1} + δ_k X_{i,t−1} + η_{i,t,k}.
  - Definitions:
    - gap_{i,t−1} = average productivity gap (difference between old and new firm) across all new hires to firm i (minimum zero).
    - gapconnected_{i,t−1} = average gap across hires who were working for one of i’s buyers or suppliers the previous year (zero if none).
  - Results (Figure A13): coefficient estimates for ψ_k indicate additional knowledge spillovers from hiring workers from more productive buyers or suppliers; these additional spillovers are persistent.
- Methodological notes:
  - Firm-level regressions do not include firm fixed effects due to short panel concerns and potential Nickell (1981) bias.
  - New hires considered in HN exclude hires who were not permanent workers of another firm in the previous year.
  - Standard errors clustered at the municipality level; figures report 90 percent confidence intervals for persistence plots.

*Source: wpiea2020205-print-pdf - 2019. Working Paper, Department of Economics, UC Berkeley.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020205-print-pdf.pdf_