## wpiea2020117-print-pdf

## Source details

**Canonical URL:** [wpiea2020117-print-pdf](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020117-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020117-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020117-print-pdf.pdf.json)

---

### Literature review: origins, definitions, and channels
- GVCs emerged globally; 19th century steam engines initiated international trade and production unbundling; the 1980s information technology revolution enabled the “second unbundling.”
- Drivers of GVC rise: technological progress, trade costs, transport and communication costs, access to resources and markets, and trade policy reforms.
- Terminology and intellectual origins listed include fragmented production blocks and service links (Jones and Kierzkowski, 1990), global commodity chain (Gereffi, 1994), offshore sourcing (Arndt, 1997), disintegration of production (Feenstra, 1998), global production sharing (Yeats, 2001), vertical specialization (Hummels et al., 2001), outsourcing (Grossman and Helpman, 2002), vertical production networks (Hanson et al., 2005), trade in tasks (Grossman and Rossi-Hansberg, 2008), and the second unbundling (Baldwin, 2013).
- Producer-driven versus buyer-driven distinction; emphasis on “network” over “chain” in recent research.
- Documented benefits of GVC participation:
  - Increases efficiency in global production.
  - Helps developing countries increase participation in global trade and diversify exports.
  - Creates more employment opportunities.
  - Facilitates technology transfer and knowledge spillovers.
  - Gains depend on country/firm position in the chain.
- Empirical links between GVC participation and productivity:
  - An increase of 10 percent in GVC participation increased average productivity by close to 1.7 percent (Constantinescu et al., 2017).
  - Increased GVC participation associated with more firm patenting, higher employment growth for the average firm and faster employment growth for patenting firms.
  - Exporting firms in Latvia and Estonia are more productive, larger, pay higher wages, and are more capital intensive than non-exporters (Benkovskis et al., 2018).
  - North African firms participating in GVCs perform better ex ante and obtain productivity gains ex post (Del Prete et al., 2017).
- Firm-level empirical work expanding: links between firm characteristics and GVC integration choices and firms’ responses to exchange rate movements in the context of GVCs.

### Data sources, sample construction, and deflators
- Firm-level data: ORBIS commercial database covering sales, material costs, number of employees, and capital stock.
- Estonian coverage: more than 103,000 firms in 19 NACE sectors, from 1999 to 2016; final estimation sample: 27,451 firms with more than 216,000 observations.
- Data cleaning and sample restrictions:
  - Drop firms whose accounting period is not 12 months.
  - Remove observations with missing or negative values of sales, employees, capital, and material costs.
  - Restrict sample to years 2002 to 2016 after Orbis–Statistics Estonia comparison.
- Deflators:
  - Material costs and capital stock deflated using industrial producer price index (PPI) from Eurostat.
  - Sales deflated using domestic output price indexes from Eurostat; where missing, HICP used and matched to two-digit NACE sectors.
- Coverage of cleaned sample relative to national aggregates:
  - Covers more than 60% of total output.
  - Above 30% of total employees.
  - Over 30% of total value added.
  - About 60% of total exports in Estonia.
- GVC measures: Eora Multi-Region Input-Output (MRIO) database covering 26 industries in 189 countries from 1990 to 2015.

### Key sample summary statistics (Table 1)
- Observations: 216,020
- ln(Sales) mean: 7.35; standard deviation: 1.80
- ln(Materials) mean: 6.37; standard deviation: 2.25
- ln(Capital) mean: 5.22; standard deviation: 2.20
- ln(Employees) mean: 1.29; standard deviation: 1.21
- ln(TFP) observations: 215,979; mean: 1.27; standard deviation: 0.21
- Note: Values are in real terms, the base year for the price index: 2015.

### GVC measurement and position indices
- GVC participation index (Koopman et al. (2010) formulation):
  - GVCp_cs = (FVA_cs + DVX_cs) / Gross Exports_c
  - FVA_cs: foreign value added of sector s in country c — “backward participation”.
  - DVX_cs: indirect value added of sector s in country c used by other countries in their exports — “forward participation”.
- Position (upstreamness) index:
  - Upstream_cs = ln(1 + DVX_cs / Gross Exports_c) − ln(1 + FVA_cs / Gross Exports_c)
  - Upstream > 0: upstream sector; Upstream < 0: downstream sector.

### Productivity estimation methodology
- Production function framework:
  - General Q_it = F(L_it, M_it, K_it; β); log form y_it = h(l_it, m_it, k_it; β) + ω_it + ε_it where ω_it is firm-observed productivity and ε_it unobservables.
- Translog production function used with squares and interactions of ln labor, ln capital, ln material; controls include firm, industry, and year fixed effects.
- Endogeneity addressed using materials as proxy for productivity following Ackerberg et al. (2015).
- Material demand inverted and approximated non-parametrically by a third-order polynomial; productivity law of motion approximated by a fourth-order polynomial.
- Estimation via generalized method of moments using lagged variables as instruments; productivity computed as ω̂_it = Φ̂_it − ĥ(l_it, m_it, k_it; β).
- Estimation sample restricted to firms manufacturing a single product for at least three consecutive years.

### Stylized facts: Estonia’s GVC participation and productivity
- Aggregate and geographic patterns:
  - Estonia’s total value-added export and import share in gross export (GVC participation measure) was always higher than 70% during 1995 to 2015.
  - Intensity increased slightly from 2000 until 2008, declined after the global economic crisis, recovered to highest level in 2011 (at around 80%), then fell again.
  - General trend: deeper GVC participation in Estonia from 1996 to 2015.
  - Estonia more involved in regional than global value chains; top sources/destinations mostly European; Sweden and Finland each account for more than 10% of Estonia’s value-added exports in 2015; top 10 destinations account for more than three-quarters of Estonia’s total value-added exports.
  - Around 15 percent of Estonia’s value-added imports originate in Russia; Germany, Finland and Sweden are other important source countries; China is a significant Asian provider.
- Sectoral patterns (selected entries from Table 2, 2015):
  - Electrical and Machinery: Total GVC Participation index 15.65; Backward linkage 14.40; Forward linkage 1.25; Position index Upstream −0.122.
  - Textiles and Wearing Apparel: Total 8.49; Backward 7.42; Forward 1.07; Upstream −0.061.
  - Transport: Total 6.90; Backward 4.09; Forward 2.81; Upstream −0.012.
  - Petroleum, Chemical Products: Total 6.32; Backward 4.35; Forward 1.97; Upstream −0.023.
  - Average Manufacturing: Total 6.24; Backward 5.09; Forward 1.15; Upstream −0.038.
  - Average Services: Total 2.14; Backward 0.74; Forward 1.40; Upstream 0.007.
  - Total economy: Total 3.13; Backward 2.09; Forward 1.04; Upstream −0.010.
- Sectoral insights:
  - GVC participation mainly driven by backward participation (use of foreign intermediates).
  - Manufacturing: higher GVC participation intensity, higher backward linkages, lower upstream indexes (closer to final consumer).
  - Services: transport and financial services show high participation; transport shows higher backward participation while many services exhibit higher forward linkages and higher upstreamness.
  - Low GVC participation sectors: recycling, mining and quarrying, construction.
- Time evolution by industry:
  - Construction, transport, agriculture and trading sectors show increasing GVC participation trends.
  - Manufacturing sectors saw a sharp decline after 2008.
- Productivity patterns:
  - Average productivity in Estonia fluctuated but broadly increased over 2002–2016.
  - Largest productivity gains in manufacturing; some manufacturing industries increased average productivity by close to 20 percent during 2002-2016.
  - Services experienced little increases or declines in average productivity.
  - Average productivity increased across all firm-size categories with small variability.

### Industry-level evidence on GVC participation and productivity
- Baseline sectoral specification:
  - ln(ω)_st = α1 GVCp_st−1 + β2 ln(Output)_st + γ_t + ε_st where ln(ω)_st is median log productivity and GVCp_st−1 is lagged industry-level GVC participation.
- Endogeneity strategy:
  - Lag GVC participation by one year.
  - Instrument industry-level GVC participation using world average GVC participation for the same industry: GVCp_IV_s = (1/C) ∑_c (GVCp_cs).
  - Sectoral gross output and year fixed effects included.
- Reported empirical magnitudes:
  - OLS Lagged GVC: -0.012 (standard error 0.120)
  - IV Lagged GVC: 0.478* (standard error 0.256)
  - Log industry output: OLS 0.019*** (0.006); IV 0.012* (0.007)
  - Regressions run for 26 sectors from 2002 to 2016, Observations 301
  - Instrument diagnostics: KP LM stat: 64.90; KP Wald F stat: 109.8
- Interpretation:
  - IV evidence indicates a positive effect of lagged GVC participation on sectoral median productivity; OLS near zero.

### Firm-level direct IV evidence
- Main firm-level IV (log TFP dependent, clustered SEs at industry-year):
  - Lagged GVC (column 1): 3.405** (1.717)
  - Observations: 165,055
  - KP LM stat: 8.154
  - KP Wald F stat: 25.88
  - First stage F-stat 1: 25.88; First stage F-stat 2: 403.5
- Magnitude interpretation:
  - Moving an industry from the 10th percentile to the 90th percentile of GVC participation would increase a firm’s productivity in that industry by approximately 23 percent (example provided in source).
- Heterogeneity by firm characteristics (selected interactions):
  - GVC * Initial Frontier (column 2): 1.351*** (0.120)
  - GVC * Large firm (sales) (column 4): 0.233*** (0.043)
  - GVC * Initial export status (column 5): 0.257*** (0.040)
- Interpretation of heterogeneity:
  - Frontier firms (top quartile), large firms by sales, and exporters gain more from GVC participation.
  - Sample sizes and diagnostics across heterogeneity specifications generally: Observations 165,055 (main); column (5) subsample 107,771; KP LM stat ranges 8.153 to 11.92; KP Wald F stat ranges 12.94 to 25.88; First stage F-stat 1 examples include 25.88, 15.02, 24.87, 22.87, 12.03; First stage F-stat 2 examples include 403.5, 368.5, 412.3, 219.9.

### Spillover effects via buyer and supplier linkages
- Spillover specification (no IV for spillovers):
  - Lagged GVC main effect remains positive and significant in most specifications (example: Lagged GVC 3.605** (1.566)).
  - From BUYERS (Lag) coefficients are negative and sometimes significant (example: From BUYERS (Lag) -5.619* (3.061)).
  - From SUPPLIERS (Lag) coefficients generally negative and often insignificant (example: From SUPPLIERS (Lag) -0.585 (1.777)).
- Heterogeneous spillover interactions:
  - GVC Buyer * Initial Frontier: -0.437** (0.213)
  - GVC Supplier * Initial Frontier: 1.321*** (0.115)
- Interpretation:
  - Downstream (buyer) industry GVC participation tends to be associated with lower productivity for frontier/large/export firms relative to others.
  - Upstream (supplier) industry GVC participation benefits frontier and large firms, suggesting technology transfer and higher-quality inputs drive client-industry productivity gains.

### Robustness checks
- Alternative aggregate measure (mean log productivity):
  - OLS Lagged GVC: 0.003 (0.119)
  - IV Lagged GVC: 0.506** (0.258)
  - Observations 301; KP LM stat 64.90; KP Wald F stat 109.8
- Alternative firm-level outcome (labor productivity = value-added / employees):
  - Lagged GVC example: 14.042 (9.773)
  - Observations 151,221; First stage F-stat 1 example 26.21; KP LM stat example 8.099; KP Wald F stat example 26.21
- Continuous heterogeneity specifications:
  - Example: Lagged GVC -48.690*** (7.724) with interaction GVC * Initial Productivity 40.210*** (5.803); reported marginal effect 1.972 with S.E. 2.332.
- Excluding multinationals:
  - Signs maintained; significance reduced with smaller sample.
  - Example: Lagged GVC 5.538* (3.177); Observations 7,100; First stage F-stat 1 example 23.38; KP LM stat example 8.105; KP Wald F stat example 23.38.

### Key empirical magnitudes and sample statistics (selected)
- Industry-level IV Lagged GVC: 0.478* (0.256)
- Firm-level IV Lagged GVC (main specification): 3.405** (1.717)
- Firm-level labor productivity Lagged GVC (alternative): 14.042 (9.773)
- Spillover from buyers (example): -5.619* (3.061)
- Spillover from suppliers (example): -0.585 (1.777)
- Sample sizes:
  - Industry regressions Observations 301
  - Firm regressions Observations 165,055 (main); subsamples 107,771; robustness sample 151,221; excluding multinationals 7,100
- Instrument/test statistics examples:
  - KP LM stat: 64.90 (industry IV), 8.154 (firm IV example)
  - KP Wald F stat: 109.8 (industry IV), 25.88 (firm IV example)
  - First stage F-stat 1 examples: 25.88, 15.02, 24.87, 22.87, 12.03
  - First stage F-stat 2 examples: 403.5, 368.5, 412.3, 219.9

### Policy implications and main conclusions (Section VI)
- Main conclusion:
  - GVC participation has a positive and significant impact on the productivity of Estonian firms, with heterogeneous effects across firm types.
- Heterogeneity implications:
  - Initially frontier firms, large firms by sales, and exporting firms reap more benefits from GVC participation than non-frontier firms, small firms, and non-exporting firms.
  - Upstream (supplier) GVC participation is positively correlated with productivity of client industries; frontier and large firms benefit more from upstream participation, while non-frontier and small firms benefit more from downstream participation.
- Policy considerations in the context of COVID-19 and regional integration:
  - Estonia’s deep involvement in regional value chains suggests policies that strengthen European integration or closer Baltic cooperation (for example, infrastructure projects) may be warranted given supply disruptions and onshoring incentives.
  - Policies to boost productivity should account for firm heterogeneity: measures that help non-frontier, small, and non-exporting firms access GVCs (or the benefits of supplier improvements) could broaden productivity gains across the economy.

*Source content derived from wpiea2020117-print-pdf.*

### Section VI provides main conclusions.

### Section VI provides main conclusions.

### Literature review: origins, definitions, and channels
- GVCs emerged as a worldwide phenomenon; 19th century steam engines triggered international trade and unbundling of consumption from production, with local clustering in factories and industrial districts.
- The 1980s information technology revolution (telexes, facsimiles, Internet, high speed international communication networks) enabled the “second unbundling” and coordination of production at a distance (FGI, NTU, and WTO, 2013).
- Drivers of the rise and changes of GVCs: technological progress, trade costs, transport and communication costs, access to resources and markets, and trade policy reforms (De Backer and Miroudot, 2014).
- Terminology and intellectual origins of GVCs include: fragmented production blocks and service links (Jones and Kierzkowski, 1990), global commodity chain (Gereffi, 1994), offshore sourcing (Arndt, 1997), disintegration of production (Feenstra, 1998), global production sharing (Yeats, 2001), vertical specialization (Hummels et al., 2001), outsourcing (Grossman and Helpman, 2002), vertical production networks (Hanson et al., 2005), trade in tasks (Grossman and Rossi-Hansberg, 2008), and the second unbundling (Baldwin, 2013).
- Distinction in GVCs: “producer-driven” (lead firms upstream controlling design and assembly) versus “buyer-driven” (retailers and branded marketers control production, often outsourced). Recent research emphasizes “network” over “chain” (Coe and Hess, 2007). This paper focuses on “global value chains” by measuring position and participation of countries in global production.
- Documented benefits of GVC participation:
  - Increases efficiency in global production (Baldwin, 2012; Grossman and Rossi-Hansberg, 2008).
  - Provides opportunities for developing countries to increase participation in global trade and diversify exports (Kowalski et al., 2015).
  - Creates more employment opportunities (UNCTAD, 2013).
  - Facilitates technology transfer and knowledge spillovers (Pietrobelli and Rabellotti, 2011; Kawakami et al., 2012).
  - Gains vary depending on a country/firm’s position in the chain.
- Empirical findings linking GVC participation and productivity:
  - Constantinescu et al. (2017): an increase of 10 percent in GVC participation increased average productivity by close to 1.7 percent.
  - Increased GVC participation associated with more firm patenting, higher employment growth for the average firm and faster employment growth for patenting firms (IMF World Economic Outlook, 2018).
  - Benkovskis et al. (2018): exporting firms in Latvia and Estonia are more productive, larger, pay higher wages, and are more capital intensive than non-exporters.
  - Del Prete et al. (2017): North African firms participating in GVCs perform better ex ante and obtain productivity gains ex post.
- Firm-level empirical work is expanding: Alfaro et al. (2018) and Del Prete and Rungi (2017) link firm characteristics to GVC integration choices; Ahmed et al. (2015) and Berthou and Dhyne (2018) study firms’ responses to exchange rate movements in the context of GVCs.

### Data overview and sample construction
- Firm-level data source: ORBIS commercial database covering sales, material costs, number of employees, and capital stock.
- Estonian firm-level coverage: more than 103,000 firms in 19 NACE sectors, from 1999 to 2016.
- Data cleaning and sample restrictions:
  - Drop firms whose accounting period is not 12 months.
  - Remove observations with missing or negative values of sales, employees, capital, and material costs.
  - After comparing Orbis with Statistics Estonia, restrict sample to years from 2002 to 2016.
  - Final estimation sample: 27,451 firms with more than 216,000 observations.
- Deflators:
  - Material costs and capital stock deflated using industrial producer price index (PPI) from Eurostat.
  - Sales deflated using domestic output price indexes from Eurostat.
  - For two-digit NACE industries missing domestic output price index from Eurostat, domestic output price replaced by Harmonized Indices of Consumer Prices (HICP) and matched to two-digit NACE sectors.
- Coverage of cleaned sample relative to national aggregates:
  - Covers more than 60% of total output.
  - Above 30% of total employees.
  - Over 30% of total value added.
  - About 60% of total exports in Estonia.
- Eora Multi-Region Input-Output (MRIO) database used for GVC measures:
  - Eora covers 26 industries in 189 countries from 1990 to 2015 and provides final demand matrices, intermediate input matrices, and value-added matrices.

### Key sample summary statistics (Table 1)
- Observations: 216,020
- ln(Sales) mean: 7.35; standard deviation: 1.80
- ln(Materials) mean: 6.37; standard deviation: 2.25
- ln(Capital) mean: 5.22; standard deviation: 2.20
- ln(Employees) mean: 1.29; standard deviation: 1.21
- ln(TFP) observations: 215,979; mean: 1.27; standard deviation: 0.21
- Note: Values are in real terms, the base year for the price index: 2015.

### Measures of GVC participation and position
- GVC participation index (Koopman et al. (2010) formulation used to eliminate double counting):
  - GVCp_cs = (FVA_cs + DVX_cs) / Gross Exports_c  (equation labeled (3.2.1))
  - FVA_cs: foreign value added of sector s in country c (imported intermediate input content of exports) — “backward participation”.
  - DVX_cs: indirect value added of sector s in country c used by other countries in their exports — “forward participation”.
- Position (upstreamness) index (Koopman et al. (2010)):
  - Upstream_cs = ln(1 + DVX_cs / Gross Exports_c) − ln(1 + FVA_cs / Gross Exports_c)  (equation labeled (3.2.2))
  - Upstream > 0 indicates upstream sector (supplies intermediates for others); Upstream < 0 indicates downstream sector (uses intermediates from other countries).

### Productivity estimation methodology
- Production function framework:
  - General production function Q_it = F(L_it, M_it, K_it; β) (equation (3.3.1)).
  - Log form y_it = h(l_it, m_it, k_it; β) + ω_it + ε_it (equation (3.3.2)), where ω_it is firm productivity observed by firm, ε_it is unobservables to both firm and econometrician.
- Translog production function used (equation (3.3.3)) to allow for returns to scale; includes squares and interactions of ln labor, ln capital, ln material; C is set of controls (firm fixed effects, industry fixed effects, year fixed effects).
- Endogeneity concerns: inputs chosen after observing productivity; material used as proxy for productivity following Ackerberg et al. (2015).
- Material demand inverted: ω_it = f_t^{-1}(k_it, m_it, l_it) (equation (3.3.5)); f_t^{-1} treated non-parametrically approximated by third-order polynomial including logged inputs, squares, cubes, and interaction terms.
- Law of motion for productivity assumed: ω_it = g_t(ω_it−1) + γ_it with E[γ_it | I_it−1] = 0; g approximated by fourth-order polynomial (equation (3.3.8)).
- Estimation approach:
  - Recover productivity innovation γ̂_it(β) as residual from regressing ω̂_it(β) on ω̂_it−1(β).
  - Use generalized method of moments with moment conditions E(γ̂_it(β) B) = 0 where B contains lags of variables in the translog function and instruments list C provided.
  - Productivity computed as ω̂_it = Φ̂_it − ĥ(l_it, m_it, k_it; β) (equation (3.3.10)).
- Sample for estimation: firms that manufacture a single product for at least three consecutive years (following De Loecker et al. (2016)) to satisfy moment conditions and address multi-product timing issues.

### Stylized facts: Estonia’s GVC participation and productivity
- Estonia’s GVC participation:
  - Estonia’s total value-added export and import share in gross export (GVC participation measure) was always higher than 70% during 1995 to 2015.
  - Intensity increased slightly from 2000 until 2008, modest decline after the global economic crisis, recovered to highest level in 2011 (at around 80%), then started falling again.
  - General trend: deeper GVC participation in Estonia from 1996 to 2015.
  - Estonia is more involved in regional than global value chains; top destinations and sources are mostly European countries (exceptions: US and China).
  - Sweden and Finland each account for more than 10% of Estonia’s value-added exports in 2015; top 10 destination countries account for more than three-quarters of Estonia’s total value-added exports.
  - Around 15 percent of Estonia’s value-added imports originate in Russia; Germany, Finland and Sweden are other important source countries; China is a significant Asian provider.
- Sectoral patterns (Table 2: GVC Participation and Position by Sector in Estonia in 2015; selected entries preserved exactly):
  - Electrical and Machinery: Total GVC Participation index 15.65; Backward linkage 14.40; Forward linkage 1.25; Position index Upstream −0.122.
  - Textiles and Wearing Apparel: Total 8.49; Backward 7.42; Forward 1.07; Upstream −0.061.
  - Transport: Total 6.90; Backward 4.09; Forward 2.81; Upstream −0.012.
  - Petroleum, Chemical Products: Total 6.32; Backward 4.35; Forward 1.97; Upstream −0.023.
  - Average Manufacturing: Total 6.24; Backward 5.09; Forward 1.15; Upstream −0.038.
  - Average Services: Total 2.14; Backward 0.74; Forward 1.40; Upstream 0.007.
  - Total economy: Total 3.13; Backward 2.09; Forward 1.04; Upstream −0.010.
- Sectoral insights:
  - Estonia’s GVC participation mainly driven by backward participation (use of foreign intermediates in Estonia’s exports).
  - Manufacturing sectors: higher GVC participation intensity and higher backward linkages; lower upstream indexes (closer to final consumer).
  - Service sectors: transport and financial services show high participation; transport shows higher backward participation while many services exhibit higher forward linkages and higher upstreamness (further from final consumer).
  - Sectors with low GVC participation: recycling, mining and quarrying, construction.
- Evolution over time by industry:
  - Construction, transport, agriculture and trading sectors show increasing GVC participation trends.
  - Manufacturing sectors saw a sharp decline after 2008 due to the financial crisis.
- Productivity patterns:
  - Average productivity in Estonia fluctuated but broadly increased over 2002–2016.
  - Largest productivity gains observed in manufacturing industries; some manufacturing industries increased average productivity by close to 20 percent during 2002-2016.
  - Services experienced little increases or declines in average productivity.
  - Average productivity increased across all firm-size categories with small variability; no clear superior performance of large firms on average.

### Evidence on the relationship between GVC participation and productivity
- Visual and preliminary evidence:
  - Figure 8 shows positive correlation between GVC participation and productivity; they tend to fluctuate in the same direction.
  - Evidence suggests a time lag: it appears to take time for GVC participation to impact productivity.
- Empirical strategy at industry level:
  - Baseline specification (equation (5.1)):
    - ln(ω)_st = α1 GVCp_st−1 + β2 ln(Output)_st + γ_t + ε_st
    - ln(ω)_st is median log productivity of all firms in sector s (proxy for sectoral productivity).
    - GVCp_st−1: lagged industry-level GVC participation (Koopman measure).
    - Year fixed effects γ_t and sectoral gross output included.
  - Addressing endogeneity/reverse causality:
    - Lag GVC participation by one year to reflect delayed effects and to reduce reverse causality concerns.
    - Instrument industry-level GVC participation using world average GVC participation for the same industry:
      - GVCp_IV_s = (1/C) ∑_c (GVCp_cs)  (equation (5.2))
    - Instrument validity relies on similar drivers of GVC participation intensity between Estonia and world for the same sector and that global forces affect Estonia’s productivity only through their impact on Estonia’s GVC participation; sectoral gross output and year fixed effects included to mitigate spillover concerns.
- Reported empirical result (summary of result narrative):
  - Table 3 reports OLS and IV regressions with sectoral median log productivity as dependent variable.
  - The coefficient on lagged GVC participation in column (2) is positive and statistically significant.
  - Interpreted magnitude (text truncated in supplied content): “An increase of 1 percentage point in the intensity of GVC participation of an industry raises median productivity of that industry by around” — (the supplied content ends here and does not provide the completed numeric estimate).

*Source: ORBIS data, EORA Input-Output tables, and author’s calculations as presented in the supplied PDF content.*

### 0.48 percent. This number is consistent with what is shown in Figure 8. This impact is also similar to

### wpiea2020117-print-pdf - 0.48 percent

### Instrument validity and industry-level productivity
- IV estimates show a positive effect of GVC participation on industry productivity while OLS is close to zero:
  - OLS Lagged GVC: -0.012 (standard error 0.120)
  - IV Lagged GVC: 0.478* (standard error 0.256)
- Controls and sample:
  - Log industry output: OLS 0.019*** (0.006); IV 0.012* (0.007)
  - Regressions run for 26 sectors from 2002 to 2016, Observations 301
- Instrument diagnostics:
  - KP LM stat: 64.90
  - KP Wald F stat: 109.8
- Note on instrumentation:
  - GVC participation instrumented with average GVC participation measure at world level.
  - Tests for under identification and weak identification confirm instrument validity.

### Firm-level direct effects of GVC participation on productivity
- Main IV results (log TFP as dependent variable, clustered SEs at industry-year):
  - Lagged GVC (column 1): 3.405** (1.717)
  - Observations: 165,055
  - KP LM stat: 8.154
  - KP Wald F stat: 25.88
  - First stage F-stat 1: 25.88; First stage F-stat 2: 403.5
- Magnitude interpretation:
  - If an industry moves from the 10th percentile to the 90th percentile of GVC participation, productivity of a firm in that industry would increase by approximately 23 percent (example: a firm at the 10th percentile of productivity would move to the 50th percentile under that change).

### Heterogeneity in the direct effect (firm characteristics)
- Interaction results (selected coefficients and significance):
  - GVC * Initial Frontier (column 2): 1.351*** (0.120)
  - GVC * Large firm (sales) (column 4): 0.233*** (0.043)
  - GVC * Initial export status (column 5): 0.257*** (0.040)
- Interpretation:
  - Frontier firms (top quartile of productivity) and largest firms by sales experience larger productivity gains from GVC participation.
  - Exporting firms benefit more from industry-level GVC participation than non-exporters.
- Sample sizes and diagnostics across heterogeneity specifications:
  - Observations generally 165,055; column (5) subsample 107,771
  - KP LM stat ranges: 8.153 to 11.92
  - KP Wald F stat ranges: 12.94 to 25.88
  - First stage F-stat 1 examples: 25.88, 15.02, 24.87, 22.87, 12.03
  - First stage F-stat 2 examples: 403.5, 368.5, 412.3, 219.9

### Spillover effects (buyer and supplier linkages)
- Spillover specification (no IV used for spillovers):
  - Lagged GVC main effect remains positive and significant in most specifications:
    - Example (Table 5, column 1): Lagged GVC 3.605** (1.566)
  - From BUYERS (Lag) coefficients are negative and sometimes significant:
    - Example (Table 5, column 1): From BUYERS (Lag) -5.619* (3.061)
  - From SUPPLIERS (Lag) coefficients generally negative and often insignificant:
    - Example (Table 5, column 1): From SUPPLIERS (Lag) -0.585 (1.777)
- Heterogeneous spillover patterns:
  - Interaction signs:
    - GVC Buyer * Initial Frontier: -0.437** (0.213) — downstream GVC participation associated with lower productivity for frontier firms relative to others.
    - GVC Supplier * Initial Frontier: 1.321*** (0.115) — upstream GVC participation associated with higher productivity for frontier firms.
    - Similar patterns for firm size and sales: buyer-linkage interactions negative for large firms; supplier-linkage interactions positive for large firms.
  - Interpretation:
    - Greater GVC participation of downstream (buyer) industries tends to be associated with lower productivity for frontier/large/export firms relative to others, possibly because frontier/large/export firms exert market power over buyers and face less pressure to innovate.
    - Greater GVC participation of upstream (supplier) industries benefits frontier and large firms via technology transfer, higher-quality inputs, and knowledge spillovers.

### Robustness checks
- Alternative aggregate measure at industry level (mean log productivity):
  - OLS Lagged GVC: 0.003 (0.119)
  - IV Lagged GVC: 0.506** (0.258)
  - Observations 301; KP LM stat 64.90; KP Wald F stat 109.8
- Alternative firm-level outcome (labor productivity = value-added / employees):
  - Lagged GVC coefficients substantially larger in magnitude but same sign:
    - Example (Table 7, column 1): Lagged GVC 14.042 (9.773)
  - Observations 151,221; First stage F-stat 1 example 26.21; KP LM stat example 8.099; KP Wald F stat example 26.21
  - Interpretation: sign robustness holds; magnitudes differ due to different productivity measures and distributions.
- Continuous measures for heterogeneity (Tables 8, 9):
  - Some coefficients and standard errors increase; main qualitative conclusions largely robust.
  - Example (Table 8 column 1): Lagged GVC -48.690*** (7.724) with large interaction GVC * Initial Productivity 40.210*** (5.803); marginal effect reported 1.972 with S.E. 2.332.
- Excluding multinationals:
  - Results maintain sign; some loss of significance due to reduced sample size.
  - Example (Table 10, column 1): Lagged GVC 5.538* (3.177)
  - Observations 7,100; First stage F-stat 1 example 23.38; KP LM stat example 8.105; KP Wald F stat example 23.38

### Key empirical magnitudes and statistics (selected)
- Industry-level IV Lagged GVC: 0.478* (0.256)
- Firm-level IV Lagged GVC (main specification): 3.405** (1.717)
- Firm-level labor productivity Lagged GVC (alternative outcome): 14.042 (9.773)
- Spillover from buyers (example): -5.619* (3.061)
- Spillover from suppliers (example): -0.585 (1.777)
- Sample sizes: industry regressions Observations 301; firm regressions Observations 165,055 (main); subsamples 107,771; robustness sample 151,221; excluding multinationals 7,100
- Instrument/test statistics examples:
  - KP LM stat: 64.90 (industry IV), 8.154 (firm IV example)
  - KP Wald F stat: 109.8 (industry IV), 25.88 (firm IV example)
  - First stage F-stat 1 examples: 25.88, 15.02, 24.87, 22.87, 12.03
  - First stage F-stat 2 examples: 403.5, 368.5, 412.3, 219.9

### Policy implications and conclusion
- Main conclusion:
  - GVC participation has a positive and significant impact on the productivity of Estonian firms, with heterogeneous effects across firm types.
- Heterogeneity implications:
  - Initially frontier firms, large firms by sales, and exporting firms reap more benefits from GVC participation than non-frontier firms, small firms, and non-exporting firms.
  - Upstream (supplier) GVC participation is positively correlated with productivity of client industries; frontier and large firms benefit more from upstream GVC participation, while non-frontier and small firms benefit more from downstream GVC participation.
- Policy considerations in the context of COVID-19 and regional integration:
  - Given supply disruptions and incentives to onshore production, Estonia’s deep involvement in regional (rather than global) value chains suggests stronger integration with European trade and production or closer Baltic cooperation (e.g., via Rail Baltica).
  - Policies to boost productivity growth in Estonia should account for firm heterogeneity: measures that help non-frontier, small, and non-exporting firms access GVCs (or the benefits of supplier improvements) could broaden productivity gains across the economy.

*Italic: Source content derived from wpiea2020117-print-pdf.*

### 1. Ackerberg, A. Daniel, Caves, Kevin and Frazer, Garth. 2015. “Identification Properties of Recent

### wpiea2020117-print-pdf - 1. Ackerberg, A. Daniel, Caves, Kevin and Frazer, Garth. 2015. “Identification Properties of Recent

### Selected bibliographic entries (excerpt)
- Ackerberg, A. Daniel, Caves, Kevin and Frazer, Garth. 2015. “Identification Properties of Recent Production Function Estimators.” Econometrica 83(6): 2411-2451.
- Ahmed, Swarnali, Appendino, Maximiliano and Ruta Michele. 2015. “Global Value Chains and the Exchange Rate Elasticity of Exports.” IMF Working Paper No. 15/252.
- Amiti, Mary and Konings, Jozef. 2007. “Trade Liberalization, Intermediate Inputs, and Productivity: Evidence from Indonesia.” American Economic Review Vol. 97, No.5.
- Antras, Pol and Chor, Davin. 2013. “Organizing the Global Value Chain”. Econometrica 81(6): 2127-2204.
- Baldwin, R. 2012. “Global Supply Chains: Why They Emerged, Why They Matter, and Where They are Going.” CEPR Discussion Papers No. 9103.
- De Loecker, J., Goldberg, K. P., Khandelwal, A. K. and Pavcnik, N. (2016). “Prices, Markups, and Trade Reform.” Econometrica, Vol. 84, No. 2, pp. 445-510.
- Koopman, Robert, William Powers, Zhi Wang, and Shang-Jin Wei. 2010. “Give Credit Where Credit is Due: Tracing Value Added in Global Production Chains.” NBER Working Paper 16426.
- UNCTAD. 2013. “World Investment Report 2013: Global Value Chains: Investment and Trade for Development” United Nations Conference on Trade and Development. United Nations Publication ISBN 978-92-1-112868-0.
- Feenstra, Robert C., Robert Inklaar and Marcel P. Timmer. 2015. "The Next Generation of the Penn World Table." American Economic Review. 105(10), 3150-3182.

### Appendix A — Variables used in the productivity estimation
- Table A1: Variable definitions (ORBIS variables involved)
  - Gross output, Y: SALES (OPERATING_REVENUE_TURNOVER if SALES is missing or negative)
  - Value added, VA: VA=COSTS_OF_EMPLOYEES (wL)+ EBITDA (rK) (ADDED_VALUE if VA is missing)
  - Employment, L: NUMBER_OF_EMPLOYEES
  - Capital, K: TANGIBLE_FIXED_ASSETS, DEPRECIATION_AND_AMORTIZATION
  - Intermediate inputs, M: MATERIAL_COSTS (SALES-VA if MATERIAL_COSTS is missing)
- Value added imputation:
  - VA = wL + rK.
  - ORBIS counterparts: COSTS_OF_EMPLOYEES and EBITDA. If at least one missing, replace by ADDED_VALUE.
- Capital stock (Perpetual Inventory Method, PIM):
  - K_it = K_i,t-1 (1−δ_it) + I_it
  - I_it = (K_it_BV − K_i,t-1_BV + DEPR_it_BV) / P_t
    - where K_BV and DEPR_BV denote book value of fixed tangible assets (TANGIBLE_FIXED_ASSETS in ORBIS) and depreciation (DEPRECIATION_AND_AMORTIZATION in ORBIS), respectively.
  - P is the industrial producer price index (PPI) from Eurostat used as the investment price deflator.
  - δ_it = DEPR_it_BV / K_it_BV.
  - Trimming: trim the top 1% and the bottom 99% of the distribution of δ to mitigate measurement error concerns.
- Initial capital stock:
  - For the first observed year or for years where both real investment and depreciation rate are missing:
    - K_i0 = K_i0_BV / P_0

### Appendix B — Comparison between ORBIS and Statistics Estonia (selected tables and exact figures)
- Table B1: Output (millions euro, nominal values) — Orbis After cleaning; % over Statistics Estonia; Number of firms
  - 1999: Statistics Estonia 10870.5; After cleaning 5,670; % over Statistics Estonia 52.2; Number of firms 10,693
  - 2000: Statistics Estonia 12787.7; After cleaning 7,600; % over Statistics Estonia 59.4; Number of firms 13,405
  - 2001: Statistics Estonia 14626; After cleaning 8,920; % over Statistics Estonia 61.0; Number of firms 15,394
  - 2002: Statistics Estonia 16211.1; After cleaning 6,950; % over Statistics Estonia 42.9; Number of firms 8,113; Before cleaning 12,400; % over Statistics Estonia 76.5; Number of firms 18,246
  - 2003: Statistics Estonia 17804.5; After cleaning 9,700; % over Statistics Estonia 54.5; Number of firms 8,897; Before cleaning 16,800; % over Statistics Estonia 94.4; Number of firms 20,914
  - 2004: Statistics Estonia 19973.7; After cleaning 13,000; % over Statistics Estonia 65.1; Number of firms 9,809; Before cleaning 22,300; % over Statistics Estonia 111.6; Number of firms 23,788
  - 2005: Statistics Estonia 23055.8; After cleaning 13,400; % over Statistics Estonia 58.1; Number of firms 10,565; Before cleaning 23,500; % over Statistics Estonia 101.9; Number of firms 26,819
  - 2006: Statistics Estonia 27187.9; After cleaning 18,600; % over Statistics Estonia 68.4; Number of firms 11,576; Before cleaning 33,100; % over Statistics Estonia 121.7; Number of firms 30,913
  - 2007: Statistics Estonia 32089.7; After cleaning 24,700; % over Statistics Estonia 77.0; Number of firms 13,033; Before cleaning 44,500; % over Statistics Estonia 138.7; Number of firms 35,862
  - 2008: Statistics Estonia 33123.2; After cleaning 24,700; % over Statistics Estonia 74.6; Number of firms 14,207; Before cleaning 44,500; % over Statistics Estonia 134.3; Number of firms 40,529
  - 2009: Statistics Estonia 26499.5; After cleaning 18,300; % over Statistics Estonia 69.1; Number of firms 13,122; Before cleaning 36,200; % over Statistics Estonia 136.6; Number of firms 44,629
  - 2010: Statistics Estonia 29361.8; After cleaning 17,800; % over Statistics Estonia 60.6; Number of firms 15,973; Before cleaning 38,600; % over Statistics Estonia 131.5; Number of firms 50,411
  - 2011: Statistics Estonia 34377.3; After cleaning 20,800; % over Statistics Estonia 60.5; Number of firms 17,379; Before cleaning 47,700; % over Statistics Estonia 138.8; Number of firms 56,835
  - 2012: Statistics Estonia 36954.2; After cleaning 23,500; % over Statistics Estonia 63.6; Number of firms 19,290; Before cleaning 53,400; % over Statistics Estonia 144.5; Number of firms 65,450
  - 2013: Statistics Estonia 39024.5; After cleaning 27,000; % over Statistics Estonia 69.2; Number of firms 20,937; Before cleaning 59,200; % over Statistics Estonia 151.7; Number of firms 74,399
  - 2014: Statistics Estonia 40218.4; After cleaning 25,000; % over Statistics Estonia 62.2; Number of firms 22,684; Before cleaning 55,700; % over Statistics Estonia 138.5; Number of firms 84,020
  - 2015: Statistics Estonia 40860.1; After cleaning 20,900; % over Statistics Estonia 51.2; Number of firms 23,077; Before cleaning 48,900; % over Statistics Estonia 119.7; Number of firms 86,298
  - 2016: Statistics Estonia 42086.8; After cleaning 21,200; % over Statistics Estonia 50.4; Number of firms 19,547; Before cleaning 47,400; % over Statistics Estonia 112.6; Number of firms 81,483

- Table B2: Employees (thousands) — Orbis After cleaning; % over Statistics Estonia
  - 1999: Statistics Estonia 530.4; After cleaning 150.8; % over Statistics Estonia 28.4
  - 2000: Statistics Estonia 534.5; After cleaning 177.5; % over Statistics Estonia 33.2
  - 2001: Statistics Estonia 541.1; After cleaning 189.2; % over Statistics Estonia 35.0
  - 2002: Statistics Estonia 542.3; After cleaning 131.0; % over Statistics Estonia 24.2; Before cleaning 207.5; % over Statistics Estonia 38.3
  - 2003: Statistics Estonia 548.1; After cleaning 142.2; % over Statistics Estonia 25.9; Before cleaning 221.3; % over Statistics Estonia 40.4
  - 2004: Statistics Estonia 540.9; After cleaning 155.6; % over Statistics Estonia 28.8; Before cleaning 245.7; % over Statistics Estonia 45.4
  - 2005: Statistics Estonia 564.1; After cleaning 167.1; % over Statistics Estonia 29.6; Before cleaning 258.3; % over Statistics Estonia 45.8
  - 2006: Statistics Estonia 589.4; After cleaning 176.1; % over Statistics Estonia 29.9; Before cleaning 269.0; % over Statistics Estonia 45.6
  - 2007: Statistics Estonia 583.5; After cleaning 178.6; % over Statistics Estonia 30.6; Before cleaning 272.9; % over Statistics Estonia 46.8
  - 2008: Statistics Estonia 591; After cleaning 179.9; % over Statistics Estonia 30.4; Before cleaning 279.2; % over Statistics Estonia 47.2
  - 2009: Statistics Estonia 527.9; After cleaning 153.1; % over Statistics Estonia 29.0; Before cleaning 237.0; % over Statistics Estonia 44.9
  - 2010: Statistics Estonia 501; After cleaning 148.9; % over Statistics Estonia 29.7; Before cleaning 226.0; % over Statistics Estonia 45.1
  - 2011: Statistics Estonia 532.7; After cleaning 156.2; % over Statistics Estonia 29.3; Before cleaning 244.3; % over Statistics Estonia 45.9
  - 2012: Statistics Estonia 539.7; After cleaning 161.7; % over Statistics Estonia 30.0; Before cleaning 258.1; % over Statistics Estonia 47.8
  - 2013: Statistics Estonia 544.6; After cleaning 173.4; % over Statistics Estonia 31.8; Before cleaning 279.2; % over Statistics Estonia 51.3
  - 2014: Statistics Estonia 549.4; After cleaning 180.8; % over Statistics Estonia 32.9; Before cleaning 305.9; % over Statistics Estonia 55.7
  - 2015: Statistics Estonia 564.6; After cleaning 175.4; % over Statistics Estonia 31.1; Before cleaning 301.6; % over Statistics Estonia 53.4
  - 2016: Statistics Estonia 565.2; After cleaning 171.2; % over Statistics Estonia 30.3; Before cleaning 302.6; % over Statistics Estonia 53.5

- Table B3: Value added (millions euro, nominal values) — Orbis After cleaning; % over Statistics Estonia
  - 1999: Statistics Estonia 4836.2; After cleaning 934; % over Statistics Estonia 19.3
  - 2000: Statistics Estonia 5512.9; After cleaning 1,260; % over Statistics Estonia 22.9
  - 2001: Statistics Estonia 6227.2; After cleaning 1,410; % over Statistics Estonia 22.6
  - 2002: Statistics Estonia 6914.8; After cleaning 1,520; % over Statistics Estonia 22.0; Before cleaning 1,970; % over Statistics Estonia 28.5
  - 2003: Statistics Estonia 7773; After cleaning 2,130; % over Statistics Estonia 27.4; Before cleaning 2,610; % over Statistics Estonia 33.6
  - 2004: Statistics Estonia 8595.8; After cleaning 2,800; % over Statistics Estonia 32.6; Before cleaning 3,680; % over Statistics Estonia 42.8
  - 2005: Statistics Estonia 9972.7; After cleaning 2,980; % over Statistics Estonia 29.9; Before cleaning 3,530; % over Statistics Estonia 35.4
  - 2006: Statistics Estonia 11889; After cleaning 4,210; % over Statistics Estonia 35.4; Before cleaning 5,600; % over Statistics Estonia 47.1
  - 2007: Statistics Estonia 14258.5; After cleaning 5,540; % over Statistics Estonia 38.9; Before cleaning 6,840; % over Statistics Estonia 48.0
  - 2008: Statistics Estonia 14719.2; After cleaning 4,960; % over Statistics Estonia 33.7; Before cleaning 6,310; % over Statistics Estonia 42.9
  - 2009: Statistics Estonia 12281; After cleaning 3,830; % over Statistics Estonia 31.2; Before cleaning 5,180; % over Statistics Estonia 42.2
  - 2010: Statistics Estonia 12874.3; After cleaning 4,080; % over Statistics Estonia 31.7; Before cleaning 5,610; % over Statistics Estonia 43.6
  - 2011: Statistics Estonia 14616.4; After cleaning 4,660; % over Statistics Estonia 31.9; Before cleaning 6,330; % over Statistics Estonia 43.3
  - 2012: Statistics Estonia 15676; After cleaning 5,220; % over Statistics Estonia 33.3; Before cleaning 7,090; % over Statistics Estonia 45.2
  - 2013: Statistics Estonia 16590.9; After cleaning 6,350; % over Statistics Estonia 38.3; Before cleaning 8,050; % over Statistics Estonia 48.5
  - 2014: Statistics Estonia 17201.9; After cleaning 5,670; % over Statistics Estonia 33.0; Before cleaning 7,640; % over Statistics Estonia 44.4
  - 2015: Statistics Estonia 17610.4; After cleaning 5,050; % over Statistics Estonia 28.7; Before cleaning 7,220; % over Statistics Estonia 41.0
  - 2016: Statistics Estonia 18118.1; After cleaning 5,140; % over Statistics Estonia 28.4; Before cleaning 7,090; % over Statistics Estonia 39.1

- Table B4: Exports (millions euro, nominal values) — Orbis After cleaning; % over Statistics Estonia; Before cleaning % over Statistics Estonia
  - 2004: Statistics Estonia 4768.7; After cleaning 0; % over Statistics Estonia 0.0; Before cleaning 0; % over Statistics Estonia 0.0
  - 2005: Statistics Estonia 6201.9; After cleaning 0; % over Statistics Estonia 0.0; Before cleaning 0; % over Statistics Estonia 0.0
  - 2006: Statistics Estonia 7719.0; After cleaning 0; % over Statistics Estonia 0.0; Before cleaning 0; % over Statistics Estonia 0.0
  - 2007: Statistics Estonia 8033.5; After cleaning 0; % over Statistics Estonia 0.0; Before cleaning 0; % over Statistics Estonia 0.0
  - 2008: Statistics Estonia 8470.1; After cleaning 0; % over Statistics Estonia 0.0; Before cleaning 0; % over Statistics Estonia 0.0
  - 2009: Statistics Estonia 6486.9; After cleaning 4,060; % over Statistics Estonia 62.6; Before cleaning 6,390; % over Statistics Estonia 98.5
  - 2010: Statistics Estonia 8743.0; After cleaning 4,920; % over Statistics Estonia 56.3; Before cleaning 9,280; % over Statistics Estonia 106.1
  - 2011: Statistics Estonia 12003.4; After cleaning 6,090; % over Statistics Estonia 50.7; Before cleaning 11,700; % over Statistics Estonia 97.5
  - 2012: Statistics Estonia 12521.1; After cleaning 7,000; % over Statistics Estonia 55.9; Before cleaning 13,900; % over Statistics Estonia 111.0
  - 2013: Statistics Estonia 12288.2; After cleaning 7,780; % over Statistics Estonia 63.3; Before cleaning 15,600; % over Statistics Estonia 127.0
  - 2014: Statistics Estonia 12006.0; After cleaning 7,140; % over Statistics Estonia 59.5; Before cleaning 14,900; % over Statistics Estonia 124.1
  - 2015: Statistics Estonia 11575.3; After cleaning 6,130; % over Statistics Estonia 53.0; Before cleaning 12,800; % over Statistics Estonia 110.6
  - 2016: Statistics Estonia 11904.8; After cleaning 5,660; % over Statistics Estonia 47.5; Before cleaning 12,300; % over Statistics Estonia 103.3

*Source: wpiea2020117-print-pdf.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020117-print-pdf.pdf_