## wpiea2020056-print-pdf - Section IX.A presents the table of summary statistics of the cleaning process.

## Source details

**Canonical URL:** [wpiea2020056-print-pdf - Section IX.A presents the table of summary statistics of the cleaning process.](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020056-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020056-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020056-print-pdf.pdf.json)

---

### Data sources and benchmark sample
- Orbis database compiled by Bureau van Dijk (BvD); coverage extends to more than 300 million firms across the world.
- Main analysis uses 2016 data for each firm on:
  - (1) number of employees,
  - (2) total asset,
  - (3) nationality of shareholders,
  - (4) direct and total percentages of shares held by the shareholders.
- 2016 chosen as the benchmark since 2016 data is the most complete recent data due to the reporting lags of Orbis as of 2019.
- Variables are taken from Orbis financial historical database and historical ownership database and matched using the unique BvD firm identifier.
- Industry-level analysis (Section V) and switching-firm analysis (Section VI) further restrict observations to those with industry information and those that can match past data respectively; the number of observations in each analysis is reported in each section.
- Known quality concerns in Orbis documented in Kalemli-Ozcan, et al. (2015) and Tørsløv et al. (2018).

### Cleaning steps and exclusions
- Standard cleaning to keep non-missing observations plus three additional filtering steps:
  - Drop observations of consolidated data in Orbis (consolidated data aggregate subsidiaries and can misrepresent employment in the economy where the parent is incorporated).
  - Classify firms into FDI and domestic firms (see section below for definition and complications).
  - Drop Special Purpose Entities (SPEs) as defined below.

### Classifying FDI and domestic firms — issues and chosen approach
- BPM6 definition: a direct investment enterprise is an entity subject to control or significant influence by a nonresident; (1) immediate relationship: nonresident directly owns more than 10 percent equity with voting power; (2) indirect relationship: nonresident exercises more than 10 percent voting power through chains of direct investment relationships.
- Orbis reports ownership as "direct percentage" and "total percentage" with subtle definitional differences; using Orbis total percentage to estimate BPM6 direct investment can under- or over-estimate direct investment enterprises.
- Additional complications: direct percentages may not sum to 100; sometimes only shareholder names are reported without percentage information.
- Adopted practical definition (simpler than BPM6) due to data limitations:
  - FDI firms are those with at least one foreign shareholder owning more than 10 percent share in either direct or total percentage.
  - Robustness: Section IX shows main results remain similar after replacing 10 by 50 so that firms are restricted to those under control of foreign parent companies.

### Treatment of Special Purpose Entities (SPEs)
- SPEs: legal entities used for special purposes (isolating financial risks, tax avoidance); often pass-through entities that do not reflect real activities FDI statistics intend to capture.
- Exclusion rule used (following IMF (2018)): FDI firms with less than five employees are defined as SPEs and dropped from main analysis.
- To prevent domestic shell companies driving results, main analysis also excludes domestic firms with less than five employees.
- Robustness checks (Sections IX.D and IX.E):
  - Alternative 1: keep all domestic firms but drop FDI firms with less than five employees.
  - Alternative 2: use the lower five percentile of the employment distribution instead of five employees.
- Summary definition (verbatim):
  - Definition. Fix a firm in Orbis identified by the unique BvD firm ID. The firm is an FDI firm if (1) it reports an unconsolidated financial statement, (2) it has at least one foreign shareholder owning more than 10 percent share in either direct or total percentage, and (3) it has at least five employees. The firm is a domestic firm if it satisfies (1), (3), and (4) all the foreign shareholders hold less than 10 percent share in both direct and total percentage.

### Main analysis: employment per asset and hypothesis testing
- Comparison metric: number of employees divided by total asset; due to fat-tailed distribution, take the log as suggested by Willett (2015).
- Each firm is represented by 푒푒 = log10( number of employees / total asset ).
- Hypotheses tested across economies:
  - H0: 피피[푒푒퐷퐷퐷퐷퐷퐷] = 피피[푒푒퐹퐹퐷퐷퐹퐹]
  - H1: 피피[푒푒퐷퐷퐷퐷퐷퐷] ≠ 피피[푒푒퐹퐹퐷퐷퐹퐹]
- t-statistic used: T = (푒푒̅퐷퐷퐷퐷퐷퐷 − 푒푒̅퐹퐹퐷퐷퐹퐹) / √(estimated variance) with observations assumed i.i.d.; economies with either the number of FDI or domestic firms less than 50 are dropped.
- Aggregate result (Figure 3):
  - In most economies, domestic firms tend to create more jobs than FDI firms for each dollar of assets.
  - Some economies exhibit large positive t-statistics; no economies exhibit large negative t-statistics.
  - Figure 4 shows large t-statistics are not driven by outliers: densities for the three largest t-statistic economies show FDI firm densities shifted left of domestic firm densities.

### Additional analyses and interpretations
- Alternative interpretation cautions: employment per asset 푒푒 may not map to firm productivity (value-added per employee) because total asset is not necessarily a measure of value-added.
- Policy implications:
  - Results can inform economies that actively attract FDI — may motivate policymakers to demand additional commitments from foreign investors to generate more benefits in the recipient economy.
  - In economies with legal restrictions that limit FDI, immediate policy responses may be limited.
- Robustness to firm-size exclusions:
  - Figure 5: Qualitative result survives if all firms with less than 100 employees are dropped; dropping SMEs does not change insight.
- Robustness checks summarized (annexes):
  - (1) total assets replaced by shareholders’ funds (Section IX.B),
  - (2) 10 percent threshold replaced by 50 percent (Section IX.C),
  - (3) SPEs restricted to FDI firms and all domestic firms included (Section IX.D),
  - (4) five-employee threshold replaced by five percentile (Section IX.E).
  - These robustness checks suggest main result does not hinge on FDI-related definition details.

### Wages, employee cost, and heterogeneity
- Alternate outcome: cost of employees per asset defined as 푒푒𝑐 = log10( cost of employees / total asset ).
- Figure 6: In the majority of economies, domestic firms spend more on the cost of employees than FDI firms for each dollar of assets, although heterogeneity increases relative to Figure 3.
  - This does not necessarily imply higher wage per employee in domestic firms; it implies domestic firms generate more total income for workers per dollar of assets in many economies.

### Industry-level findings (focus: United Kingdom)
- United Kingdom chosen for industry analysis: highest t-statistic in Figure 3 among economies with industry information for more than 90 percent of firms.
- Figure 7: Domestic firms employ more workers per asset than FDI firms in all industries of the United Kingdom (all t-statistics positive; some statistically insignificant).
- Figure 8: Services and manufacturing show the two highest t-statistics; density plots show domestic firm distributions are right-shifted relative to FDI firms in both industries.
  - Services sector exhibits the sharpest contrast between FDI and domestic firms.
- Cross-sectional pattern (Figure 9): A larger share of domestic firms in construction, wholesale trade, and services is associated with a larger aggregate difference (higher likelihood that domestic firms employ more workers than FDI firms per asset).
  - Financial industry size is not correlated with the difference in employment per asset between FDI and domestic firms (see Section IX.F).

### Analysis of switchers (time-series dimension)
- Focus: firms that switch ownership status between FDI and domestic from 2013 to 2014.
- Switcher analysis controls some firm fixed effects but is limited to brownfield FDI (excludes greenfield FDI).
- Switchers are compared with non-switchers with 2 years lag and forward; switcher-specific characteristics may confound interpretation.
- Four groups studied:
  - 1. Firms that are domestic firms from 2011 to 2016.
  - 2. Firms that are domestic firms from 2011 to 2013 and FDI firms from 2014 to 2016.
  - 3. Firms that are FDI firms from 2011 to 2016.
  - 4. Firms that are FDI firms from 2011 to 2013 and domestic firms from 2014 to 2016.

### Switcher analysis and employment per asset (Figure 10; Section IX.G)
- Time-series result: The ownership change from 2013 to 2014 does not produce an obvious change in the employment per asset ratio for the four groups of firms.
- Cross-country robustness: Similar stable patterns observed for Russia, Germany, and all firms in the world (Section IX.G) as for the United Kingdom.
- Group patterns:
  - Always-domestic firms have higher employment per asset ratio uniformly over time.
  - Always-FDI firms do not necessarily exhibit the lowest employment per asset ratio.
  - The FDI-to-domestic (FDI→domestic) group is based on a small sample size of 44.
- Interpretation: Employment per asset behavior appears driven more by the technological characteristics of firms that attract cross-border investment than by ownership structure per se.
- Causal caution: Regression results cannot be interpreted as causality since both axes are endogenous; interpretations are predictive based on correlation.

### Difference-in-difference (Switcher’s) regression results (Section IX.H; Table 4)
- Regression specification: log employment per asset for firm i at year T with firm and year fixed effects; treatment dummy for firms that switch ownership; interaction of year and treatment included; 2013 baseline; standard errors clustered at firm level.
- Interpretation rule: Ownership switch would show effects if β2011 and β2012 ≈ 0 and β2014 ≠ 0; β2015 and β2016 indicate persistence or fade-out.
- Estimation results (coefficients with p-values in parentheses):
  - Comparison: Group 1 (dom-dom v.s. dom-fdi)
    - β2011 = -0.0052 (0.8259)
    - β2012 = -0.0109 (0.6365)
    - β2014 = -0.0087 (0.7280)
    - β2015 = -0.0053 (0.8220)
    - β2016 = -0.0137 (0.5818)
  - Comparison: Group 2 (fdi-fdi v.s. fdi-dom)
    - β2011 = -0.0135 (0.6750)
    - β2012 = -0.0070 (0.8183)
    - β2014 = -0.0084 (0.7870)
    - β2015 = -0.0093 (0.7736)
    - β2016 = -0.0057 (0.8549)
- Additional regression statistics:
  - Firm fixed effect: Yes (both regressions)
  - Time fixed effect: Yes (both regressions)
  - Observations:
    - Balanced Panel: n = 206252, T = 6, N = 1237512 (first comparison)
    - Balanced Panel: n = 29354, T = 6, N = 176124 (second comparison)
  - R^2:
    - 4.5682e-07 (first comparison)
    - 1.474e-06 (second comparison)
- Conclusion from regressions: All coefficients are insignificant; the difference between FDI and domestic firms is not the consequence of ownership change but rather reflects endogenous firm characteristics.

### Aggregate conclusion and policy implications
- Empirical summary:
  - In most economies, domestic firms employ more workers than FDI firms for each dollar of assets.
  - Industry-level (United Kingdom) analysis: the aggregate result is not driven by specific industries; domestic firms tend to employ more workers per asset than FDI firms across all industries.
  - Switchers’ analysis: change of ownership itself does not have immediate impacts on employment per asset.
  - Overall interpretation: differences between FDI and domestic firms likely reflect technological differences of firms owned by foreign investors rather than ownership per se.
- Policy implications:
  - Where policies actively attract FDI, policymakers may demand additional commitments from foreign investors to generate more benefits in the recipient economy.
  - Interventions forcing ownership changes from FDI to domestic are not likely to produce immediate job creation unless the underlying production technology of firms also changes.
  - Ownership is endogenous to firm characteristics that affect employment per asset; thus ownership intervention alone is unlikely to change employment per asset behavior.

### Data cleaning and sample composition (Annex A; Table 2) — sample figures
- Examples of reported reductions and counts by economy:
  - GB: # of obs Before Cleaning = 105246; # of obs After Cleaning = 35116; Survival Rate = 0.333656; # of FDI Firms = 8207; # of Domestic Firms = 26909
  - RU: # of obs Before Cleaning = 75713; # of obs After Cleaning = 66875; Survival Rate = 0.88327; # of FDI Firms = 7260; # of Domestic Firms = 59615
  - CN: # of obs Before Cleaning = 56589; # of obs After Cleaning = 22378; Survival Rate = 0.395448; # of FDI Firms = 5071; # of Domestic Firms = 17307
  - DE: # of obs Before Cleaning = 116706; # of obs After Cleaning = 69166; Survival Rate = 0.592652; # of FDI Firms = 8673; # of Domestic Firms = 60493
  - FR: # of obs Before Cleaning = 74164; # of obs After Cleaning = 38764; Survival Rate = 0.522679; # of FDI Firms = 5391; # of Domestic Firms = 33373

### Robustness checks and alternative definitions (Annexes B–F)
- Alternative financial denominator (Annex B):
  - Replace total assets with shareholders’ funds.
  - Issue: shareholders’ funds are sometimes negative; observations with negative shareholders’ funds are dropped.
  - Result: qualitative features remain similar—domestic firms on average employ more workers than FDI firms per unit of shareholders’ funds.
- Alternative FDI threshold (Annex C):
  - Main analysis uses 10 percent foreign shareholder threshold (BPM6 standard); alternative threshold is 50 percent.
  - Result: qualitative result survives with 50 percent threshold; behavior of FDI firms mostly driven by those controlled by foreign shareholders with more than 50 percent share.
  - Supporting evidence: Table 3 reports the number of firms by share held by foreigners in categories [0,10), [10,50), [50,100]; the majority of FDI firms are controlled by foreign investors.
- Alternative SPE definitions (Annex D–E):
  - Main analysis drops SPEs defined as firms with less than five employees (per IMF 2018 recommendation).
  - Annex D: redefine SPEs as FDI firms with less than 5 employees and assume none of domestic firms are SPEs—result: inclusion of domestic firms with <5 employees changes results for several economies, but majority remain the same.
  - Annex E: define SPEs as firms in lower five percentile of employment distributions of FDI and domestic firms—result: outcome almost identical to main analysis.
- Industry structure check (Annex F):
  - Finance sector share does not explain much about difference in employment per asset (Figure 15): Y-axis = t-statistics; X-axis = share of finance industry within domestic firms.

*Source: wpiea2020056-print-pdf — Section IX.A presents the table of summary statistics of the cleaning process. Canonical URL: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020056-print-pdf.pdf*

### Section IX.A presents the table of summary statistics of the cleaning process.

### wpiea2020056-print-pdf - Section IX.A presents the table of summary statistics of the cleaning process.

### Data sources and benchmark sample
- Orbis database compiled by Bureau van Dijk (BvD); coverage extends to more than 300 million firms across the world.
- Main analysis uses 2016 data for each firm on:
  - (1) number of employees,
  - (2) total asset,
  - (3) nationality of shareholders,
  - (4) direct and total percentages of shares held by the shareholders.
- 2016 chosen as the benchmark since 2016 data is the most complete recent data due to the reporting lags of Orbis as of 2019.
- Variables are taken from Orbis financial historical database and historical ownership database and matched using the unique BvD firm identifier.
- Industry-level analysis (Section V) and switching-firm analysis (Section VI) further restrict observations to those with industry information and those that can match past data respectively; the number of observations in each analysis is reported in each section.
- Known quality concerns in Orbis documented in Kalemli-Ozcan, et al. (2015) and Tørsløv et al. (2018).

### Cleaning steps and exclusions
- Standard cleaning to keep non-missing observations plus three additional filtering steps:
  1. Drop observations of consolidated data in Orbis (consolidated data aggregate subsidiaries and can misrepresent employment in the economy where the parent is incorporated).
  2. Classify firms into FDI and domestic firms (see section below for definition and complications).
  3. Drop Special Purpose Entities (SPEs) as defined below.

### Classifying FDI and domestic firms — issues and chosen approach
- BPM6 definition: a direct investment enterprise is an entity subject to control or significant influence by a nonresident; (1) immediate relationship: nonresident directly owns more than 10 percent equity with voting power; (2) indirect relationship: nonresident exercises more than 10 percent voting power through chains of direct investment relationships.
- Orbis reports ownership as "direct percentage" and "total percentage" with subtle definitional differences:
  - Example table shows Orbis entries where total percentage of A on D is 5 = (0.5 ×0.05+0.5 ×0.05) ×100 instead of 10 because Orbis calculates total percentage by summing the multiplication of shares throughout chains, whereas BPM6 counts voting power using only the last legs (i.e., 5 +5 =10).
- Using Orbis total percentage to estimate BPM6 direct investment can under- or over-estimate direct investment enterprises (examples provided: under-estimation in Figure 1; over-estimation in Figure 2 when many middle firms dilute control).
- Additional complications: patterns of existing and missing data — direct percentages may not sum to 100 (missing shareholders or mis-recorded percentages); sometimes only shareholder names are reported without percentage information.
- Adopted practical definition (simpler than BPM6) due to data limitations:
  - FDI firms are those with at least one foreign shareholder owning more than 10 percent share in either direct or total percentage.
  - Robustness: Section IX shows main results remain similar after replacing 10 by 50 so that firms are restricted to those under control of foreign parent companies.

### Treatment of Special Purpose Entities (SPEs)
- SPEs: legal entities used for special purposes (isolating financial risks, tax avoidance); often pass-through entities that do not reflect real activities FDI statistics intend to capture.
- Exclusion rule used (following IMF (2018)): FDI firms with less than five employees are defined as SPEs and dropped from main analysis.
- To prevent domestic shell companies driving results, main analysis also excludes domestic firms with less than five employees.
- Robustness checks (Sections IX.D and IX.E):
  - Alternative 1: keep all domestic firms but drop FDI firms with less than five employees.
  - Alternative 2: use the lower five percentile of the employment distribution instead of five employees.
- Summary definition (as stated verbatim):
  - Definition. Fix a firm in Orbis identified by the unique BvD firm ID. The firm is an FDI firm if (1) it reports an unconsolidated financial statement, (2) it has at least one foreign shareholder owning more than 10 percent share in either direct or total percentage, and (3) it has at least five employees. The firm is a domestic firm if it satisfies (1), (3), and (4) all the foreign shareholders hold less than 10 percent share in both direct and total percentage.

### Main analysis: employment per asset and hypothesis testing
- Comparison metric: number of employees divided by total asset; due to fat-tailed distribution, take the log as suggested by Willett (2015).
- Each firm is represented by 푒푒=log10( number of employees / total asset ) — (formula presented in source).
- Hypotheses tested across economies:
  - H0: 피피[푒푒퐷퐷퐷퐷퐷퐷] = 피피[푒푒퐹퐹퐷퐷퐹퐹]
  - H1: 피피[푒푒퐷퐷퐷퐷퐷퐷] ≠ 피피[푒푒퐹퐹퐷퐷퐹퐹]
- t-statistic used: T = (푒푒̅퐷퐷퐷퐷퐷퐷 − 푒푒̅퐹퐹퐷퐷퐹퐹) / √(estimated variance) with observations assumed i.i.d.; economies with either the number of FDI or domestic firms less than 50 are dropped.
- Main aggregate result (Figure 3):
  - In most economies, domestic firms tend to create more jobs than FDI firms for each dollar of assets.
  - Some economies exhibit large positive t-statistics; no economies exhibit large negative t-statistics.
  - Figure 4 shows large t-statistics are not driven by outliers: densities for the three largest t-statistic economies show FDI firm densities shifted left of domestic firm densities.

### Additional analyses and interpretations
- Alternative interpretation cautions: employment per asset 푒푒 may not map to firm productivity (value-added per employee) because total asset is not necessarily a measure of value-added.
- Policy implications:
  - Results can inform economies that actively attract FDI — may motivate policymakers to demand additional commitments from foreign investors to generate more benefits in the recipient economy.
  - In economies with legal restrictions that limit FDI, immediate policy responses may be limited.
- Robustness to firm-size exclusions:
  - Figure 5: Qualitative result survives if all firms with less than 100 employees are dropped; dropping SMEs does not change insight.
- Robustness checks summarized (annexes):
  - (1) total assets replaced by shareholders’ funds (Section IX.B),
  - (2) 10 percent threshold replaced by 50 percent (Section IX.C),
  - (3) SPEs restricted to FDI firms and all domestic firms included (Section IX.D),
  - (4) five-employee threshold replaced by five percentile (Section IX.E).
  - These robustness checks suggest main result does not hinge on FDI-related definition details.

### Wages, employee cost, and heterogeneity
- Alternate outcome: cost of employees per asset defined as 푒푒𝑐 = log10( cost of employees / total asset ) — (formula presented in source).
- Figure 6: In the majority of economies, domestic firms spend more on the cost of employees than FDI firms for each dollar of assets, although heterogeneity increases relative to Figure 3.
  - This does not necessarily imply higher wage per employee in domestic firms; it implies domestic firms generate more total income for workers per dollar of assets in many economies.

### Industry-level findings (focus: United Kingdom)
- United Kingdom chosen for industry analysis: highest t-statistic in Figure 3 among economies with industry information for more than 90 percent of firms.
- Figure 7: Domestic firms employ more workers per asset than FDI firms in all industries of the United Kingdom (all t-statistics positive; some statistically insignificant).
- Figure 8: Services and manufacturing show the two highest t-statistics; density plots show domestic firm distributions are right-shifted relative to FDI firms in both industries.
  - Services sector exhibits the sharpest contrast between FDI and domestic firms.
- Cross-sectional pattern (Figure 9): A larger share of domestic firms in construction, wholesale trade, and services is associated with a larger aggregate difference (higher likelihood that domestic firms employ more workers than FDI firms per asset).
  - Financial industry size is not correlated with the difference in employment per asset between FDI and domestic firms (as noted in Section IX.F).

### Analysis of switchers (time-series dimension)
- Focus: firms that switch ownership status between FDI and domestic from 2013 to 2014.
- Switcher analysis controls some firm fixed effects but is limited to brownfield FDI (excludes greenfield FDI).
- Switchers are compared with non-switchers with 2 years lag and forward; switcher-specific characteristics may confound interpretation.
- The analysis studies four groups of firms (details follow in source beyond provided excerpt).

*Source: wpiea2020056-print-pdf — Section IX.A presents the table of summary statistics of the cleaning process. Canonical URL: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020056-print-pdf.pdf*

### 1. Firms that are domestic firms from 2011 to 2016.

### 1. Firms that are domestic firms from 2011 to 2016.

### Switcher analysis and employment per asset (Figure 10; Section IX.G)
- Time-series result: The ownership change from 2013 to 2014 does not produce an obvious change in the employment per asset ratio for the four groups of firms:
  - Group definitions:
    - 1. Firms that are domestic firms from 2011 to 2016.
    - 2. Firms that are domestic firms from 2011 to 2013 and FDI firms from 2014 to 2016.
    - 3. Firms that are FDI firms from 2011 to 2016.
    - 4. Firms that are FDI firms from 2011 to 2013 and domestic firms from 2014 to 2016.
- Cross-country robustness: Similar stable patterns observed for Russia, Germany, and all firms in the world (Section IX.G) as for the United Kingdom.
- Group patterns:
  - Always-domestic firms have higher employment per asset ratio uniformly over time.
  - Always-FDI firms do not necessarily exhibit the lowest employment per asset ratio.
  - The FDI-to-domestic (FDI→domestic) group is based on a small sample size of 44.
- Interpretation: Employment per asset behavior appears driven more by the technological characteristics of firms that attract cross-border investment than by ownership structure per se.
- Causal caution: Regression results cannot be interpreted as causality since both axes are endogenous; interpretations are predictive based on correlation.

### Difference-in-difference (Switcher’s) regression results (Section IX.H; Table 4)
- Regression specification: log employment per asset for firm i at year T with firm and year fixed effects; treatment dummy for firms that switch ownership; interaction of year and treatment included; 2013 baseline; standard errors clustered at firm level.
- Interpretation rule: Ownership switch would show effects if β2011 and β2012 ≈ 0 and β2014 ≠ 0; β2015 and β2016 indicate persistence or fade-out.
- Estimation results (coefficients with p-values in parentheses):
  - Comparison: Group 1 (dom-dom v.s. dom-fdi)
    - β2011 = -0.0052 (0.8259)
    - β2012 = -0.0109 (0.6365)
    - β2014 = -0.0087 (0.7280)
    - β2015 = -0.0053 (0.8220)
    - β2016 = -0.0137 (0.5818)
  - Comparison: Group 2 (fdi-fdi v.s. fdi-dom)
    - β2011 = -0.0135 (0.6750)
    - β2012 = -0.0070 (0.8183)
    - β2014 = -0.0084 (0.7870)
    - β2015 = -0.0093 (0.7736)
    - β2016 = -0.0057 (0.8549)
- Additional regression statistics:
  - Firm fixed effect: Yes (both regressions)
  - Time fixed effect: Yes (both regressions)
  - Observations:
    - Balanced Panel: n = 206252, T = 6, N = 1237512 (first comparison)
    - Balanced Panel: n = 29354, T = 6, N = 176124 (second comparison)
  - R^2:
    - 4.5682e-07 (first comparison)
    - 1.474e-06 (second comparison)
- Conclusion from regressions: All coefficients are insignificant; the difference between FDI and domestic firms is not the consequence of ownership change but rather reflects endogenous firm characteristics.

### Aggregate conclusion and policy implications (VII. CONCLUSION)
- Empirical summary:
  - In most economies, domestic firms employ more workers than FDI firms for each dollar of assets.
  - Industry-level (United Kingdom) analysis: the aggregate result is not driven by specific industries; domestic firms tend to employ more workers per asset than FDI firms across all industries.
  - Switchers’ analysis: change of ownership itself does not have immediate impacts on employment per asset.
  - Overall interpretation: differences between FDI and domestic firms likely reflect technological differences of firms owned by foreign investors rather than ownership per se.
- Policy implications:
  - Where policies actively attract FDI, policymakers may demand additional commitments from foreign investors to generate more benefits in the recipient economy.
  - Interventions forcing ownership changes from FDI to domestic are not likely to produce immediate job creation unless the underlying production technology of firms also changes.
  - Ownership is endogenous to firm characteristics that affect employment per asset; thus ownership intervention alone is unlikely to change employment per asset behavior.

### Data cleaning and sample composition (Annex A; Table 2)
- The paper reports reduction of observations due to lack of asset or number of employees across economies; examples of reported figures include:
  - GB: # of obs Before Cleaning = 105246; # of obs After Cleaning = 35116; Survival Rate = 0.333656; # of FDI Firms = 8207; # of Domestic Firms = 26909
  - RU: # of obs Before Cleaning = 75713; # of obs After Cleaning = 66875; Survival Rate = 0.88327; # of FDI Firms = 7260; # of Domestic Firms = 59615
  - CN: # of obs Before Cleaning = 56589; # of obs After Cleaning = 22378; Survival Rate = 0.395448; # of FDI Firms = 5071; # of Domestic Firms = 17307
  - DE: # of obs Before Cleaning = 116706; # of obs After Cleaning = 69166; Survival Rate = 0.592652; # of FDI Firms = 8673; # of Domestic Firms = 60493
  - FR: # of obs Before Cleaning = 74164; # of obs After Cleaning = 38764; Survival Rate = 0.522679; # of FDI Firms = 5391; # of Domestic Firms = 33373
  - (Table 2 contains comparable columns for other economies; survival rates and FDI/domestic counts vary by economy.)

### Robustness checks and alternative definitions (Annexes B–F)
- Alternative financial denominator (Annex B):
  - Replace total assets with shareholders’ funds (equation shown in source).
  - Issue: shareholders’ funds are sometimes negative; observations with negative shareholders’ funds are dropped.
  - Result: qualitative features remain similar—domestic firms on average employ more workers than FDI firms per unit of shareholders’ funds—indicating robustness.
- Alternative FDI threshold (Annex C):
  - Main analysis uses 10 percent foreign shareholder threshold (BPM6 standard); alternative threshold is 50 percent.
  - Result: qualitative result survives with 50 percent threshold; behavior of FDI firms mostly driven by those controlled by foreign shareholders with more than 50 percent share.
  - Supporting evidence: Table 3 reports the number of firms by share held by foreigners in categories [0,10), [10,50), [50,100]; the majority of FDI firms are controlled by foreign investors.
- Alternative SPE definitions (Annex D–E):
  - Main analysis drops SPEs defined as firms with less than five employees (per IMF 2018 recommendation).
  - Annex D: redefine SPEs as FDI firms with less than 5 employees and assume none of domestic firms are SPEs—result: inclusion of domestic firms with <5 employees changes results for several economies, but majority remain the same; caveat on Orbis coverage for SMEs/self-employment.
  - Annex E: define SPEs as firms in lower five percentile of employment distributions of FDI and domestic firms—result: outcome almost identical to main analysis.
- Industry structure check (Annex F):
  - Finance sector share does not explain much about difference in employment per asset (Figure 15): Y-axis = t-statistics; X-axis = share of finance industry within domestic firms.

*Source: wpiea2020056-print-pdf - 1. Firms that are domestic firms from 2011 to 2016.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020056-print-pdf.pdf_