## wpiea2024179-print-pdf

## Source details

**Canonical URL:** [wpiea2024179-print-pdf](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024179-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024179-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024179-print-pdf.pdf.json)

---

### Data and measures of FDI
- Data sources and definitions:
  - CDIS: Coordinated Direct Investment Survey (CDIS) administered by the IMF beginning from 2009; coverage 110 reporting countries as of 2022. CDIS bilateral values refer to total immediate FDI between source and host countries; CDIS neither distinguishes real FDI from FDI in SPEs between countries nor measures bilateral FDI by ultimate ownership.
  - OECD: Annual bilateral FDI and SPE data for reporting OECD countries used to impute the share of SPEs out of total FDI to construct real bilateral FDI for all countries (following Damgaard et al., 2019).
  - Orbis: Firm-level data for over 448 million companies; contains immediate and ultimate investor information and industry codes; used (2009–2021 vintages) to impute ultimate bilateral FDI, country-level bilateral vertical FDI, and country-by-industry bilateral FDI (following Damgaard and Elkjaer, 2017). Vertical FDI defined as shareholder and subsidiary having different four-digit industry codes; horizontal FDI where codes match.
  - BEA: Bureau of Economic Analysis reports outward FDI from the US by industry to host countries; used to complement Orbis where US firm information is missing.
  - CEPII and BI: CEPII provides bilateral gravity controls; exchange rate data from BIS used for valuation adjustments and as gravity controls.
- Definition and scope:
  - Real FDI: FDI not directed at companies with less than five employees or with assets per employee exceeding 10 million USD (Damgaard et al., 2019).
  - Analysis focuses on real FDI where possible; data period 2009 to 2021.

### Summary statistics for bilateral FDI (CDIS, 2009–2021)
- FDI measures in billions of USD; statistics for 2009–2021:
  - Real immediate FDI: Mean 1.478, Std dev 6.980, Min 0, Max 1969.13, N 199,279
  - Real ultimate FDI: Mean 1.478, Std dev 6.560, Min 0, Max 1969.13, N 199,279
  - Real immediate vertical FDI: Mean 1.095, Std dev 3.520, Min 0, Max 1969.13, N 199,279
  - Real ultimate vertical FDI: Mean 1.086, Std dev 12.220, Min 0, Max 1559.21, N 199,279
  - Total immediate FDI: Mean 2.237, Std dev 23.380, Min 0, Max 1969.13, N 234,652
  - Total ultimate FDI: Mean 2.392, Std dev 23.740, Min 0, Max 1969.13, N 219,308
- Additional summary points:
  - Average bilateral real immediate and ultimate FDI position ≈ 1.478 billion USD; mean includes zero-FDI pairs; just under 50 percent of country pairs have zero FDI.
  - Real immediate and real ultimate FDI ≈ 60 percent of total immediate and ultimate FDI.
  - Real vertical FDI ≈ just under 80 percent of real FDI (both immediate and ultimate).
  - Correlations:
    - real immediate vs real ultimate FDI: 0.76
    - real ultimate FDI vs real ultimate vertical FDI: around 0.975
  - High correlation between vertical and overall real FDI motivates focus on overall real FDI in main analysis.

### Stylized facts and regional patterns (2009–2021)
- Regional groupings: US, China, advanced Europe, emerging Europe, the Americas (excluding the US), Asia (excluding China), rest of the world (RoW).
- US outward FDI:
  - Levels to advanced Europe, the Americas and rest of Asia rose from 2009; acceleration after 2018.
  - FDI to Emerging Europe, China and RoW plateau from 2009.
  - Distribution of immediate and ultimate US FDI roughly stable; share to advanced Europe rises slightly; share to China declines slowly from 2013.
  - Share of US ultimate FDI rises for the Americas and Asia (excluding China) and advanced Europe.
- Chinese outward FDI:
  - China’s FDI to all regions rose from 2009; immediate FDI to rest of Asia and advanced Europe rises sharply after 2016; Chinese FDI to the US plateaus from 2017.
  - Chinese ultimate FDI to advanced Europe rises sharply from 2020, dipping slightly in 2022.
  - Distribution: share to Hong Kong and Macau declines 2009–2022; share to rest of Asia and advanced Europe rises.
  - From 2020, signs of round-tripping where FDI in China is ultimately controlled by Chinese firms.
- Advanced Europe outward FDI:
  - Outward immediate and ultimate FDI to all regions rose from 2009, more to the US, the Americas (excluding the US), emerging Europe and Asia (excluding China).
  - Firms invest primarily within the bloc, though intra-bloc share declines slowly from 2009; share to the US mirrors share to advanced Europe.
  - Share to China remains low and flat.
- Geopolitical groupings:
  - Countries ranked by geopolitical proximity to the US into terciles (Catalán and Tsuruga, 2023). Patterns:
    - Countries closest and farthest from the US invest most within their bloc, even as within-bloc shares decline slowly.
    - Countries less close to US invest most in the bloc farthest from the US, though that share is declining.
    - Distribution of ultimate FDI diverges after 2018: intra-bloc ultimate FDI rises for closest and farthest; for less close, ultimate FDI to countries close to the US rises steeply after 2018.
- UN voting blocs (UN GA Resolution ES-11/1, March 2022):
  - Countries that voted to condemn Russia’s invasion: immediate and ultimate FDI within their bloc remained quite stable.
  - Countries that did not condemn: immediate FDI within bloc declines after 2018; ultimate FDI within bloc rises after 2018.
- Community detection (Blondel et al., 2008):
  - 2010–2022: no increase in number of communities nor clear reshuffling toward geopolitical alignment; typically 4 large communities: North America, rest of Western Europe, China, Russia/eastern Europe (latter sometimes subsumed into Western Europe).
  - For ultimate FDI, communities hover between 4 and 5; China’s bloc becomes less interconnected over time; Russia’s network not diminishing through 2022.

### Overall assessment on fragmentation (2009–2021)
- Evidence on FDI fragmentation is small and inconclusive:
  - FDI between the US and advanced Europe rose in absolute and relative terms.
  - No evidence that China moved away from investments in the West; Chinese FDI to advanced Europe rose in absolute and relative terms.
  - Mixed results by geopolitical distance and UN voting blocs.
  - Community detection does not show a clear fragmentation pattern (number of communities stable; only modest reshuffling).
- Conclusion: no clear relative rise in intra-bloc FDI across groupings and methods for 2009–2021.

### Measurement and empirical setup (preview of determinants)
- Gravity specification: host and source fixed effects, pair fixed effects, time-varying host and source controls, year dummies, interactions between distance measures and time.
- Pairwise distance measures:
  - Distance: geographic distance between most populated cities (measured in thousands of km).
  - same_rank: dummy = 1 if both host and source in same geopolitical tercile relative to the US (time-invariant).
  - same_UN_vote: dummy = 1 if host and source cast same vote on UN Resolution ES-11/1 (March 2022) (time-invariant).
  - geopol_prox ∈ [−1, 1]: continuous bilateral geopolitical proximity index (time-varying).
- Estimation approaches:
  - OLS on log(Fhst) with fixed effects (drops zero-valued pairs).
  - Pseudo Poisson Maximum Likelihood (PPML / QPMLE) to include zero bilateral FDI positions; PPML results preferred.
- Summary statistics for main independent variables (country-level regressions):
  - Distance (thousands km): Mean 7.408, Std dev 4.541, Min 1.055, Max 19.93, N 166,947
  - same_rank: Mean .369, Std dev .233, Min 0, Max 1, N 166,947
  - same_UN_vote: Mean .634, Std dev .481, Min 0, Max 1, N 166,947
  - geopol_prox: Mean .165, Std dev .410, Min -0.923, Max 1, N 166,947
  - Trade agreement: Mean .245, Std dev .430, Min 0, Max 1, N 166,947
  - Common language: Mean .124, Std dev .330, Min 0, Max 1, N 166,947
  - Contiguous border: Mean .025, Std dev .158, Min 0, Max 1, N 166,947
  - Colonial link: Mean .015, Std dev .121, Min 0, Max 1, N 166,947

### Determinants of FDI (QPMLE results, Section 5.1)
- Estimation notes: Gravity estimates using QPMLE; base period 2009–2013; Source and Host Country FE included; additional controls include Host and Source exchange rates and volatilities, annual population and GDP.
- Geographic distance:
  - Immediate FDI: Distance coefficient = -0.112 (QPMLE, Table 3, col. 1). A one unit (1000km) increase lowers immediate FDI by 10.59 percent ((e^{-0.112}-1)*100 = 10.59).
  - Ultimate FDI: Distance coefficient = -0.135 (QPMLE, Table 4, col. 1). A one unit (1000km) increase lowers ultimate FDI by 12.62 percent ((e^{-0.135}-1)*100 = -12.62).
- Geopolitical proximity/alignment:
  - Immediate FDI magnitudes (Table 3):
    - geopol_prox coefficient = 0.418 (col. 4): a 0.1 unit (5 percentage point) increase raises immediate FDI by 51.89 percent ((e^{0.418}-1)*100 = 51.89).
    - same_rank coefficient = 0.726 (col. 2): sharing similar geopolitical rank associated with 106.67 percent more immediate FDI ((e^{0.726}-1)*100 = 106.67).
    - same_UN_vote coefficient = 0.712 (col. 3): source has 103.80 percent more immediate FDI in host that voted same way on Russian invasion ((e^{0.712}-1)*100 = 103.80).
  - Ultimate FDI (Table 4): geopolitical proximity measures generally smaller and less statistically significant (e.g., same_rank = 0.562 in col. 1; same_un_vote = -0.155 in col. 2 not significant; geopol_prox = -0.135 in col. 3 not significant).
- Other bilateral factors:
  - Trade agreement (immediate FDI): coefficients ~ 0.311 to 0.511 (Table 3).
  - Common language (immediate FDI): coefficients ~ 0.514 to 0.552 (Table 3); ultimate FDI common language ~ 1.012 to 1.046 (Table 4).
  - Contiguous border (immediate FDI): coefficients ~ 0.406 to 0.645 (Table 3); ultimate FDI smaller/less consistent.
  - Colonial link (immediate FDI): coefficients ~ 0.241 to 0.318 (Table 3); ultimate FDI colonial link ~ 0.806 to 0.854 (Table 4).
- Time effects:
  - Relative to 2009–2013, real FDI increased significantly in 2014–2017 and 2018–2021 for both immediate and ultimate FDI (immediate FDI 2014–2017 coefficient ~ 0.267–0.297; 2018–2021 coefficient ~ 0.523–0.594).
- Sample sizes:
  - Table 3 (immediate FDI): Observations = 167130 (except col. 4: 166297).
  - Table 4 (ultimate FDI): Observations = 159123 (except col. 4: 158710).
- Significance notation: * p<0.10, ** p<0.05, *** p<0.01.
- Interpretation summary:
  - Classic gravity determinants (distance, language, border, colonial link, trade agreements) have expected signs.
  - Geopolitical proximity and alignment strongly associated with larger immediate FDI flows; weaker association for ultimate FDI.
  - Some bilateral attributes (distance, language, colonial links) have larger magnitudes for ultimate than immediate FDI, while geopolitical measures larger for immediate FDI.

### Industry-level fragmentation and US outward FDI (selected findings)
- Summary findings:
  - Geopolitical fragmentation in immediate real FDI present for strategic industries: computer manufacturing, transport, professional, scientific and technical services.
  - Fragmentation begins in 2014–2017 and intensifies in 2018–2021 for these industries (immediate FDI).
  - Industry-level fragmentation generally does not carry over to ultimate real FDI except professional, scientific and technical services showing fragmentation in 2018–2021.
  - Fragmentation more pronounced in immediate FDI than ultimate FDI, suggesting ultimate parent residence attenuates apparent fragmentation.
  - Overall, fragmentation is localized in key strategic industries and largely driven by US-China patterns.
- US outward FDI (BEA immediate FDI):
  - Aggregate (all industries): no evidence of fragmentation by geographic or geopolitical distance when all industries considered.
  - Geopolitical rank finding (aggregate): significant decline of 24.5 percent ((e^{-0.281}-1)*100 = -24.49) in 2018–2021 to hosts in same geopolitical rank relative to hosts in different rank (text description).
  - Industry-specific BEA results:
    - Computer manufacturing: geopolitical proximity mattered more for attracting US FDI in 2018–2021 relative to base period.
    - Information industry: countries that voted to condemn the invasion received a 34.71 percent ((e^{0.298}-1)*100 = 34.71) larger increase in US FDI in 2018–2021 compared to countries that did not.
    - Professional, scientific and technical services: relative rise in US FDI of 53.87 percent ((e^{0.431}-1)*100 = 53.87) in 2018–2021 to condemning countries; coefficient significant at 10 percent.
  - Implication: fragmentation confined to certain industries (transport, computer manufacturing, information, professional/scientific/technical services) and most pronounced in US outward FDI; US firms invest relatively less in geopolitically distant countries and particularly less in China.

### Firm-level evidence on intensive and extensive margins (Orbis; US investors)
- Firm-level intensive margin:
  - Sample restricted to firms with positive FDI positions in China at least once in each period for intensive-margin estimation.
  - Transport and Professional, Scientific and Technical Services: evidence of greater geographic and geopolitical fragmentation — negative interaction between geographic distance and time dummy and positive interaction between at least one geopolitical measure and time.
  - Transport fragmentation concentrated in 2018–2021; Professional, Scientific and Technical Services fragmentation appears from 2014–2017.
  - Computer Manufacturing and Information sectors: no statistically significant evidence of fragmentation at intensive margin.
- Extensive margin caveats:
  - Orbis coverage varies; changes in number of US investors recorded may reflect coverage changes rather than true extensive-margin shifts.
  - Orbis coverage generally increases over time; corporate inversions could affect recorded coverage.
- Interpretation:
  - Intensive margin contributes to fragmentation in some industries but not others.
  - Decoupling of US FDI from China appears not primarily driven by intensive margin; extensive margin likely plays a larger role for US–China shift, but uncertainty remains due to coverage limitations.

### The choice of intermediaries by ultimate parent companies (Section 5.7)
- Research question: Do ultimate parents change mix of intermediary subsidiaries depending on intended FDI destination?
- Key variable: geopol_prox ∈ [−1, 1] (closer to 1 = greater geopolitical proximity).
- Descriptive evidence (Figure 19): after 2016 ultimate sources on average less geopolitically close to destinations than intermediaries; gap small; co-movement between ultimate–host and intermediary–host geopol_prox; no prima facie evidence of aggregate change in intermediary mix along geopolitical lines.
- Firm-level regression (geopol_prox_uht as dependent variable) with Host FE θ_h and Ultimate Parent FE φ_u; Year_t dummy = 1 if 2018–2021, 0 if 2015–2017; controls include exchange rates, volatilities, population, GDP; standard errors clustered at ultimate parent level.
- Main regression results (Table 22 — all ultimate parents):
  - Year_t coefficients (2018–2021 vs 2015–2017):
    - Column (1) All: 0.0196 ∗∗∗ (0.00120)
    - Column (2) Above median proximity: -0.00898 ∗∗∗ (0.00203)
    - Column (3) Below median proximity: 0.0430 ∗∗∗ (0.00185)
  - Constants:
    - Column (1) Constant: 0.854 ∗∗∗ (0.0124)
    - Column (2) Constant: 0.825 ∗∗∗ (0.0127)
    - Column (3) Constant: 0.874 ∗∗∗ (0.0397)
  - Observations: 650353; 251029; 390 (as presented) [note: table presents concatenated counts "650353251029390" in source]
  - Interpretation:
    - Pooled sample: small positive and highly significant coefficient (table) but text emphasizes no significant change once controls included.
    - Split sample:
      - Ultimate parents above median proximity: intermediary mix becomes significantly less geopolitically close in 2018–2021 (coefficient -0.00898 ∗∗∗).
      - Ultimate parents below median proximity: intermediary mix becomes significantly more geopolitically close in 2018–2021 (coefficient 0.0430 ∗∗∗).
    - Conclusion: firms geopolitically distant from their destinations shift toward intermediaries that are geopolitically closer.
- US-based ultimate parents (Table 23):
  - Year_t coefficients:
    - Column (1) All (US): 0.0291 ∗∗∗ (0.00975)
    - Column (2) Above median proximity (US): 0.01100 (0.0151)
    - Column (3) Below median proximity (US): 0.0744 ∗∗∗ (0.0169)
  - Observations: 274012; 731; 238 (as presented) [note: table presents concatenated counts "274012731238"]
  - Interpretation:
    - US pooled sample: increase in average intermediary geopolitical proximity significant at 10 percent (column 1).
    - US parents geopolitically distant from hosts: larger increase in intermediary geopol_prox (0.0744 ∗∗∗).
    - No significant change for US parents geopolitically closer to hosts.
- Robustness and limitations:
  - Supplementary exercise: no significant evidence that intermediaries being geopolitically closer reduces FDI fragmentation significantly (result omitted for brevity).
  - Puzzle: if ownership links are public, host countries could trace ultimate owners; ultimate parents may not expect host countries to perform extensive due diligence.
  - Open question: whether intermediary mix changes materially affect realized FDI outcomes remains unresolved.
- Key takeaway:
  - Evidence that firms facing larger geopolitical distance to hosts choose intermediaries geopolitically closer to destinations, notably among US ultimate parents who are geopolitically distant.
  - No clear evidence that such intermediary-mix changes reduce FDI fragmentation outcomes.

### Additional table highlights and robustness (selected table results)
- Correlation matrix (Table A1) shows high correlations among ISH real FDI, GUO real FDI, and vertical measures (e.g., ISH real FDI vs ISH vertical real FDI = 0.989; GUO real FDI vs GUO vertical real FDI = 0.976).
- OLS selected coefficients (Table A2) for Immediate Log(FDI):
  - Distance: -0.239 ∗∗∗ (0.00715) ... -0.229 ∗∗∗ (0.00722) across specifications.
  - Trade agreement: 0.933 ∗∗∗ (0.0589) ... 0.811 ∗∗∗ (0.0594).
  - Common language: 1.206 ∗∗∗ (0.0794) ... 1.171 ∗∗∗ (0.0792).
  - Period dummies: 2014–2017: 0.351 ∗∗∗ (0.0225); 2018–2021: 0.690 ∗∗∗ (0.0286) ... 0.753 ∗∗∗ (0.0292).
- OLS fragmentation interactions (Tables A3–A4) show mixed and generally small interaction coefficients for distance and geopolitical measures across periods; some significant negative interactions for same_rank×2018–2021 and same_un_vote×2018–2021 (Table A3) indicating relative declines in same-bloc immediate FDI in later period in selected specifications.
- Sectoral tables (A5–A8) display heterogeneous and sometimes large coefficients in specific sectors (e.g., Immediate Agriculture Distance: -4.911 ∗∗∗ (1.898); Immediate Transport same_rank×2018–2021: 1.945 ∗∗∗ (0.675); geopol_prox: -2.610 ∗∗∗ (0.478)).
- Fragmentation of US FDI (Table A9 All Industries):
  - Distance: -0.253 ∗∗∗ (0.0838)
  - same_rank×2018–2012: -0.281 ∗∗∗ (0.0984)
  - geopol_prox×2018–2012: -0.409 ∗∗∗ (0.131)
- Data construction (section 9 highlights):
  - Real FDI construction follows Damgaard et al., 2019 with SPE subtraction/imputation using OECD and regression-based imputation for non-reporters; lower-bounded at 0; ratio assumed constant across bilateral positions for a given CDIS country.
  - Ultimate FDI construction follows Damgaard and Elkjaer, 2017 using Orbis where possible, δ_hi ratios to scale CDIS positions, and final scaling to ensure sums of ultimate FDI to host equal sums of immediate FDI to host.
  - Vertical FDI imputed from Orbis industry codes and applied to CDIS positions separately for immediate and ultimate FDI.

*Source: wpiea2024179-print-pdf - 3.1 Data and measures of FDI; 5.1 The determinants of FDI; 5.x industry and firm-level excerpts; 5.7 The choice of intermediaries by ultimate parent companies; 8.2 Tables; Data Appendix highlights.*

### 3.1    Data and measures of FDI

### 3.1    Data and measures of FDI

### Data sources and definitions
- CDIS: Coordinated Direct Investment Survey (CDIS) administered by the IMF beginning from 2009; coverage of the CDIS is wide, with 110 reporting countries as of 2022. Bilateral FDI values in the CDIS refer to total immediate FDI between source and host countries; the CDIS neither distinguishes real FDI from FDI in SPEs between countries nor measures bilateral FDI by ultimate ownership.
- OECD: Annual bilateral FDI and SPE data for reporting OECD countries are used to impute the share of SPEs out of total FDI to construct real bilateral FDI for all countries (following Damgaard et al., 2019). A description of how real FDI is imputed for the CDIS using the available breakdown in the OECD data is provided in the Data Appendix.
- Orbis: Firm-level balance sheet and shareholder data for over 448 million companies worldwide. Contains information on immediate and ultimate investors, industry codes, employee numbers and total firm assets; used (2009–2021 vintages) to impute ultimate bilateral FDI, country-level bilateral vertical FDI, and country-by-industry bilateral FDI (following Damgaard and Elkjaer, 2017). Vertical FDI is defined as FDI from a shareholder firm that does not share the same four-digit industry code as its subsidiary; horizontal FDI is where shareholder and subsidiary share the same industry code.
- BEA: Bureau of Economic Analysis reports outward FDI from the US by industry to host countries; used to complement Orbis where US firm information is missing.
- CEPII and BI: CEPII provides bilateral country-pair data on gravity control variables (e.g., Free Trade Agreement, colonial links, common language, bilateral geographic distances). Exchange rate data obtained from Bank of International Settlements (BIS) to adjust for valuation changes in FDI stocks due to exchange rate fluctuations and as controls in gravity regressions.

### Definition of real FDI and scope
- Real FDI is defined as FDI that is not directed at companies with less than five employees or with assets per employee exceeding 10 million USD (Damgaard et al., 2019).
- Analysis focuses on real FDI where possible; data period is 2009 to 2021 (2009 chosen as the start of CDIS; data up to end-2021 available at time of writing).

### Summary statistics for bilateral FDI (CDIS, 2009–2021)
- FDI measures are in billions of USD and are constructed as described in the Data Appendix. Summary statistics are for the period 2009 to 2021.
- Summary statistics (Mean, Std dev, Min Max N):
  - Real immediate FDI: Mean 1.478, Std dev 6.980, Min 0, Max 1969.13, N 199,279
  - Real ultimate FDI: Mean 1.478, Std dev 6.560, Min 0, Max 1969.13, N 199,279
  - Real immediate vertical FDI: Mean 1.095, Std dev 3.520, Min 0, Max 1969.13, N 199,279
  - Real ultimate vertical FDI: Mean 1.086, Std dev 12.220, Min 0, Max 1559.21, N 199,279
  - Total immediate FDI: Mean 2.237, Std dev 23.380, Min 0, Max 1969.13, N 234,652
  - Total ultimate FDI: Mean 2.392, Std dev 23.740, Min 0, Max 1969.13, N 219,308
- The average bilateral real immediate and ultimate FDI position is around 1.478 billion USD. The calculation of the mean includes bilateral country pairs with zero FDI; just under 50 percent of country pairs have zero FDI between them.
- Real immediate and real ultimate FDI make up around 60 percent of total immediate and ultimate FDI.
- Real vertical FDI makes up just under 80 percent of real FDI, for both immediate and ultimate measures.
- Correlations:
  - Correlation between real immediate and real ultimate FDI: 0.76
  - Correlation between real ultimate FDI and real ultimate vertical FDI: around 0.975
- Because of the high correlation between vertical and overall real FDI measures, subsequent analysis focuses on overall real FDI, leaving vertical FDI results to the Appendix except when results differ substantially.

### Stylized facts (2009–2021)
- Regional groupings used: US, China, advanced Europe, emerging Europe, the Americas (excluding the US), Asia (excluding China), and the rest of the world (RoW).
- US outward FDI:
  - US FDI in levels to immediate recipients rose to advanced Europe, the Americas and the rest of Asia from 2009; rate of increase accelerates after 2018.
  - FDI to Emerging Europe, China and the rest of the world plateau from 2009.
  - Distribution of immediate and ultimate US FDI roughly stable over time; share to advanced Europe rises slightly and share to China declines slowly from 2013.
  - Share of US ultimate FDI rises for the Americas and Asia (excluding China) in addition to advanced Europe.
- Chinese outward FDI:
  - China’s FDI to all regions rose from 2009; immediate FDI to rest of Asia and advanced Europe rises sharply after 2016, while Chinese FDI to the US plateaus from 2017.
  - Chinese ultimate FDI to advanced Europe rises sharply from 2020, dipping slightly in 2022.
  - Distribution of Chinese immediate FDI: share to Hong Kong and Macau declines from 2009 to 2022; share to rest of Asia and advanced Europe rises.
  - From 2020, secular trends suggest some FDI in China is ultimately controlled by Chinese firms (round-tripping).
- Advanced Europe outward FDI:
  - Outward immediate and ultimate FDI from advanced Europe to all other regions rose from 2009, more so for the US, the Americas (excluding the US), emerging Europe and Asia (excluding China).
  - Firms in advanced Europe invest primarily within the bloc, though this intra-bloc share has been declining slowly from 2009; share to the US is a mirror image of share to advanced Europe.
  - Share to China remains low and flat.
- Geopolitical groupings:
  - Using a geopolitical distance index (Catalán and Tsuruga, 2023, based on Häge, 2011), countries are ranked by proximity to the US and grouped into terciles (closest, less close, farthest).
  - Countries closest and farthest from the US tend to invest most within their bloc, even as within-bloc shares of immediate investment decline slowly over time.
  - Countries less close to the US invest most in the bloc farthest from the US, though that share is declining.
  - Distribution of ultimate FDI differs after 2018: intra-bloc share of ultimate FDI rises for countries closest and farthest from the US; for countries less close to the US, their share of ultimate FDI to countries close to the US rises steeply after 2018.
- UN voting blocs:
  - Countries that voted to condemn Russia’s invasion of Ukraine (UN General Assembly Resolution ES-11/1, March 2022): immediate and ultimate FDI within their bloc remained quite stable over time.
  - Countries that did not condemn the invasion experience more volatile changes: immediate FDI within bloc declines after 2018; ultimate FDI within bloc rises after 2018.
- Community detection (network modularity):
  - Heuristic algorithm maximizing modularity (Blondel et al., 2008) groups countries into densely connected FDI communities.
  - From 2010 to 2022 neither an increase in the number of communities nor an obvious reshuffling toward geopolitically aligned communities materialized; there are 4 large communities across years: North America, the rest of Western Europe, China, and Russia/eastern Europe (the latter sometimes subsumed into Western Europe).
  - For ultimate FDI, number of communities hovers between 4 and 5 from 2010 to 2022; bloc containing China becomes less interconnected over time; network containing Russia does not appear to be diminishing even in 2022.

### Overall assessment on fragmentation
- Evidence on FDI fragmentation from 2009 to 2021 is small and inconclusive:
  - FDI between the US and advanced Europe has risen in absolute and relative terms.
  - No evidence of China moving away from investments in the West; China’s FDI to advanced Europe rose in absolute and relative terms.
  - Mixed results when grouping countries by geopolitical distance from the US and by UN voting blocs.
  - Organic community detection does not show a clear fragmentation pattern (number of communities stable and only modest reshuffling).
- Conclusion: no clear relative rise in intra-bloc FDI across the various groupings and methods examined for 2009–2021.

### Measurement and empirical setup notes (preview of subsequent analysis)
- Gravity model specification (modified gravity equation) with host and source fixed effects, pair fixed effects, time-varying host and source controls, year dummies, and interactions between distance measures and time to assess changing importance of geographic or geopolitical distance.
- Four measures of pairwise distance considered:
  - Distance: geographic distance between most populated cities of source and host (measured in thousands of km).
  - same_rank: dummy = 1 if both host and source are in the same geopolitical tercile relative to the US (Ahn et al., 2023 based on Häge, 2011); time-invariant for practical purposes.
  - same_UN_vote: dummy = 1 if host and source cast the same vote on UN Resolution ES-11/1 (March 2022); time-invariant.
  - geopol_prox ∈ [−1, 1]: continuous bilateral geopolitical proximity index (Catalán and Tsuruga, 2023 based on Häge, 2011); time-varying.
- Estimation approaches:
  - OLS on log(Fhst) with fixed effects (standard approach), but taking logs drops zero-valued pairs.
  - Pseudo Poisson Maximum Likelihood (PPML) following Head and Ries, 2008 and Silva and Tenreyro, 2006 to include zero bilateral FDI positions; PPML results are preferred and discussed in Section 5; OLS placed in the Appendix.
- Summary statistics for main independent variables (country-level regressions):
  - Distance (in thousands of km): Mean 7.408, Std dev 4.541, Min 1.055, Max 19.93, N 166,947
  - same_rank: Mean .369, Std dev .233, Min 0, Max 1, N 166,947
  - same_UN_vote: Mean .634, Std dev .481, Min 0, Max 1, N 166,947
  - geopol_prox: Mean .165, Std dev .410, Min -0.923, Max 1, N 166,947
  - Trade agreement: Mean .245, Std dev .430, Min 0, Max 1, N 166,947
  - Common language: Mean .124, Std dev .330, Min 0, Max 1, N 166,947
  - Contiguous border: Mean .025, Std dev .158, Min 0, Max 1, N 166,947
  - Colonial link: Mean .015, Std dev .121, Min 0, Max 1, N 166,947

*Source: wpiea2024179-print-pdf - 3.1    Data and measures of FDI*

### 5.1    The determinants of FDI

### 5.1    The determinants of FDI

### Key empirical findings on determinants of immediate and ultimate FDI
- Estimation method: Gravity estimates for Equation 1 using QPMLE. Base period: 2009 to 2013. OLS results are in the Appendix.
- Geographic distance:
  - Immediate FDI: Distance coefficient = -0.112 (QPMLE, Table 3, col. 1). A one unit (i.e. 1000km) increase in geographic distance lowers immediate FDI by 10.59 percent ((e^{-0.112}-1)*100 = 10.59).
  - Ultimate FDI: Distance coefficient = -0.135 (QPMLE, Table 4, col. 1). A one unit (1000km) increase lowers ultimate FDI by 12.62 percent ((e^{-0.135}-1)*100 = -12.62).
- Geopolitical proximity/alignment:
  - Closer geopolitical proximity (same_rank, same_UN_vote, geopol_prox) correlates with larger real FDI flows from Source to Host.
  - Example magnitudes (immediate FDI, Table 3):
    - geopol_prox coefficient = 0.418 (col. 4): a 0.1 unit (5 percentage point) increase raises immediate FDI by 51.89 percent ((e^{0.418}-1)*100 = 51.89).
    - same_rank coefficient = 0.726 (col. 2): sharing similar geopolitical distance from the US is associated with 106.67 percent more immediate FDI ((e^{0.726}-1)*100 = 106.67).
    - same_UN_vote coefficient = 0.712 (col. 3): a source has 103.80 percent more immediate FDI in a host that voted the same way on the Russian invasion of Ukraine ((e^{0.712}-1)*100 = 103.80).
  - For ultimate FDI (Table 4), geopolitical proximity measures are smaller in magnitude and less statistically significant than for immediate FDI (same_rank coefficient = 0.562 in col. 1; same_un_vote = -0.155 in col. 2 is not significant; geopol_prox = -0.135 in col. 3 is not significant).
- Other bilateral factors:
  - Trade agreement: immediate FDI coefficients ~ 0.311 to 0.511 across specifications (Table 3).
  - Common language: immediate FDI coefficients ~ 0.514 to 0.552 (Table 3); ultimate FDI common language coefficients ~ 1.012 to 1.046 (Table 4), larger in magnitude.
  - Contiguous border: immediate FDI coefficients ~ 0.406 to 0.645 (Table 3); ultimate FDI smaller/less consistent.
  - Colonial link: immediate FDI coefficients ~ 0.241 to 0.318 (Table 3); ultimate FDI colonial link coefficients ~ 0.806 to 0.854 (Table 4), larger in magnitude.
- Time effects:
  - Relative to 2009-2013, real FDI increased significantly in 2014-2017 and 2018-2021 for both immediate and ultimate FDI (Table 3 and Table 4; e.g., immediate FDI: 2014-2017 coefficient ~ 0.267–0.297; 2018-2021 coefficient ~ 0.523–0.594).

### Estimation details and controls
- Fixed effects: Source Country FE and Host Country FE included for Tables 3 and 4.
- Other controls: Host and Source exchange rate with respect to the USD, volatility of Host and Source exchange rate in each year, annual population and GDP of Source and Host countries.
- Sample sizes:
  - Table 3 (immediate FDI): Observations = 167130 (except col. 4: 166297).
  - Table 4 (ultimate FDI): Observations = 159123 (except col. 4: 158710).
- Significance notation: * p<0.10, ** p<0.05, *** p<0.01.

### Summary interpretation
- Geographic distance, common language, contiguous border, colonial links, and trade agreements are classic gravity determinants with expected signs and significance.
- Geopolitical proximity and alignment are strongly associated with larger immediate FDI flows; the association is generally weaker for ultimate FDI.
- The magnitudes for certain bilateral attributes (distance, language, colonial links) tend to be larger for ultimate than for immediate FDI, while geopolitical measures are larger for immediate FDI.

*Source: wpiea2024179-print-pdf - 5.1    The determinants of FDI*

### 2013.  One should note however that this result is not robust to alternative measures of

### wpiea2024179-print-pdf - 2013.  One should note however that this result is not robust to alternative measures of

### Summary findings
- Geopolitical fragmentation in immediate real FDI has occurred on a global scale for some strategic industries: computer manufacturing, transport, and professional, scientific and technical services.
- Fragmentation in these industries appears to begin in 2014-2017 and is more stark in 2018-2021.
- These industry-level fragmentation findings do not generally carry over to ultimate real FDI; only FDI in professional, scientific and technical services shows evidence of fragmentation in 2018-2021 when using ultimate-owner measures.
- Fragmentation is more pronounced in immediate FDI than in ultimate FDI, suggesting ultimate parent company residence weakens apparent fragmentation.
- Overall, FDI fragmentation is not a widespread phenomenon but is localized in key strategic industries and largely driven by US-China FDI patterns.

### Industry-level results (Orbis coverage caveat)
- The industry results are based on incomplete Orbis coverage; this motivates a narrower focus on US outward FDI where comprehensive industry information is available.
- Industries with notable fragmentation:
  - Computer manufacturing: fragmentation begins 2014-2017 and intensifies 2018-2021 (immediate FDI).
  - Transport: evidence of fragmentation in some specifications.
  - Professional, scientific and technical services: fragmentation observed, including in ultimate FDI for 2018-2021.
  - Information (and Information and Communication): selective evidence, including increased US FDI to countries that condemned the Russian invasion in 2018-2021.
- Findings indicate superficial fragmentation in some industries that attenuates when accounting for ultimate parent company residence.

### US outward FDI — aggregate and by industry (BEA data; immediate FDI)
- Aggregate (all industries): no evidence of fragmentation in outward FDI from the US by geographic or geopolitical distance when all industries are considered.
- Geopolitical rank finding (aggregate): a significant decline of 24.5 percent ((e^−0.281−1)∗100=−24.49) in 2018-2021 to host countries in the same geopolitical rank relative to host countries in a different geopolitical rank (column 2 result described in text).
- Industry-specific BEA results:
  - Computer manufacturing: some evidence that geopolitical proximity mattered more for attracting US FDI in 2018-2021 relative to the base period.
  - Information industry: countries that voted to condemn the Russian invasion of Ukraine received a significantly larger increase in FDI from the US ((e^0.298−1)∗100=34.71 percent) in 2018-2021 compared to countries that did not.
  - Professional, scientific and technical services: some evidence of a relative rise in US FDI (e^0.431−1)∗100=53.87 percent) in 2018-2021 to countries that voted to condemn the Russian invasion; coefficient significant at the ten percent level.
- Overall implication: fragmentation is confined to certain industries (transport, computer manufacturing, information, professional/scientific/technical services) and is most pronounced in outward FDI from the US; US firms invest relatively less in geopolitically distant countries and particularly less in China.

### Firm-level study of US investors (Orbis; intensive and extensive margins)
- Dataset and regressions:
  - Uses Orbis firm-level data on US shareholder firms.
  - Regression (Equation 3) is at the firm level with host fixed effects θ_h, shareholder firm fixed effects φ_i, and controls; sample restricted to firms with positive FDI positions in China at least once in each period (2009-2013, 2014-2017, 2018-2021) for intensive-margin estimation.
- Intensive margin results:
  - Transport and Professional, Scientific and Technical Services: evidence of both greater geographic and geopolitical fragmentation — significantly negative interaction between geographic distance and time dummy, and positive interaction between at least one measure of geopolitical distance and time.
  - Transport fragmentation appears concentrated in 2018-2021.
  - Professional, Scientific and Technical Services fragmentation appears from 2014-2017.
  - Computer Manufacturing and Information sectors: no statistically significant evidence of fragmentation at the intensive margin.
- Extensive margin limitations:
  - Orbis coverage varies widely across countries; changes in the number of US investors recorded may reflect coverage changes rather than true extensive-margin shifts.
  - Figure 18 (described): number of US ultimate investor firms in China fell in 2011, rose to a peak in 2015, then declined subsequently; percentage of reporting US firms investing in China over time plotted alongside counts (Figure not reproduced here).
  - Orbis coverage generally increases over time, but corporate inversions in the US or China could affect recorded coverage.
- Interpretation:
  - Intensive margin contributes to fragmentation of US outward FDI for some industries but not others.
  - Decoupling of US FDI from China appears not to be driven primarily by the intensive margin, implying the extensive margin likely plays a bigger role for the US-China shift — but this is uncertain given Orbis coverage limitations.

### Empirical robustness and overall interpretation
- Results are robust to various empirical specifications (OLS and QPMLE) and the addition of a range of controls.
- The paper finds no evidence of generalized geo-economic fragmentation in FDI; instead fragmentation is localized:
  - Largely driven by targeted FDI barriers, notably between the US and China.
  - Concentrated in strategic industries with national security implications.
- Fragmentation is primarily observed in immediate FDI measures rather than ultimate FDI measures, indicating ultimate ownership links have been less influenced by geopolitical tensions.

*Source: Excerpt from wpiea2024179-print-pdf - 2013.  One should note however that this result is not robust to alternative measures of*

### 5.7    The choice of intermediaries by ultimate parent companies

### 5.7    The choice of intermediaries by ultimate parent companies

### Research question and data
- Objective: Examine whether ultimate parent companies change their mix of intermediary subsidiaries depending on their intended FDI destination.
- Data source: Orbis data distinguishing immediate and ultimate shareholders and ownership links.
- Key variable: geopol_prox ∈ [−1, 1], where a value closer to 1 corresponds to greater geopolitical proximity between source and host countries.
- Sample note (footnote 33): Ultimate parent companies that are also the immediate investor in their FDI destination countries without going through any intermediary country are excluded from the analysis.

### Descriptive evidence (Figure 19)
- After 2016, the ultimate sources are, on average, less geopolitically close to their FDI destination countries than their intermediaries.
- The gap in geopolitical proximity between ultimate source–host and intermediary–host is small.
- The mean geopolitical proximity between ultimate source and host destination, and between intermediary and host destination, co-move closely.
- Implication: No prima facie evidence that the average ultimate parent company is changing its intermediary mix along geopolitical lines over time at the aggregate level.

### Firm-level regression specification
- Dependent variable: geopol_prox_uht — the weighted average geopolitical proximity of intermediary countries i to host country h of ultimate parent company u at time t:
  - geopol_prox_uht = (1 / N_u^I) * sum_{i=1}^{N_u^I} ω_i × geopol_prox_uiht, with sum_i ω_i = 1.
- Regression:
  - geopol_prox_uht = θ_h + φ_u + ρ_hst β_hs + X_ht β_h + Z_st β_s + Year_t + ε_uht
  - Year_t dummy = 1 if period t falls between 2018 and 2021, and 0 if it falls between 2015 and 2017.
- Fixed effects and controls:
  - θ_h: host country fixed effects
  - φ_u: ultimate parent company fixed effects
  - ρ_hst: time-varying host-source pair controls
  - X_ht, Z_st: time-varying controls for host and source country respectively (including Host exchange rate w.r.t. the USD, volatility of Host and Source exchange rate, annual population and GDP of Source and Host countries)
  - Standard errors clustered at the ultimate parent firm level
- Sub-sample analysis: regressions run for i) ultimate parents above the median in geopolitical proximity to host destinations and ii) those below the median.

### Main regression results (Table 22 — all ultimate parent companies)
- Coefficients on Year_t (2018-2021 vs 2015-2017):
  - Column (1) All: 0.0196 ∗∗∗ (0.00120)
  - Column (2) Above median proximity: -0.00898 ∗∗∗ (0.00203)
  - Column (3) Below median proximity: 0.0430 ∗∗∗ (0.00185)
- Constants:
  - Column (1) Constant: 0.854 ∗∗∗ (0.0124)
  - Column (2) Constant: 0.825 ∗∗∗ (0.0127)
  - Column (3) Constant: 0.874 ∗∗∗ (0.0397)
- Fixed effects: Host Country FE = Yes; Ultimate Parent FE = Yes
- No. of Observations650353251029390
- Significance notation: ∗ p<0.10, ∗∗ p<0.05, ∗∗∗ p<0.01
- Interpretation:
  - Pooled sample: no significant change after including controls (column 1 shows a small positive and highly significant coefficient but the text emphasizes no significant change when controls included — the table reports coefficient and SE).
  - Split sample:
    - Ultimate parents geopolitically closer to their hosts (above median): the intermediary mix is, on average, significantly less geopolitically close in 2018-2021 compared to 2015-2017 (coefficient -0.00898 ∗∗∗).
    - Ultimate parents geopolitically farther from their hosts (below median): the intermediary mix has become significantly more geopolitically close to host countries over time (coefficient 0.0430 ∗∗∗).
  - Conclusion from Table 22: firms that are geopolitically distant from their FDI destinations appear to shift toward intermediaries that are geopolitically closer to those destinations.

### Results for US-based ultimate parent companies (Table 23)
- Coefficients on Year_t (2018-2021 vs 2015-2017):
  - Column (1) All (US): 0.0291 ∗∗∗ (0.00975)
  - Column (2) Above median proximity (US): 0.01100 (0.0151)
  - Column (3) Below median proximity (US): 0.0744 ∗∗∗ (0.0169)
- Constants:
  - Column (1) Constant: 0.515 (0.491)
  - Column (2) Constant: -0.606 (0.876)
  - Column (3) Constant: 0.828 ∗ (0.435)
- Fixed effects: Host Country FE = Yes; Ultimate Parent FE = Yes
- No. of Observations274012731238
- Significance notation: ∗ p<0.10, ∗∗ p<0.05, ∗∗∗ p<0.01
- Interpretation:
  - For US ultimate parents pooled together, the increase in average intermediary geopolitical proximity is significant at the ten percent level (column 1).
  - For US ultimate parents that are geopolitically distant from their hosts (below median), there is a larger increase in average intermediary geopolitical proximity: 0.0744 ∗∗∗ (column 3).
  - No significant change for US parent firms that are geopolitically closer to their FDI host countries (column 2).

### Robustness, supplementary tests, and limitations
- Supplementary exercise: examined whether having intermediaries that are geopolitically closer to FDI destinations reduces FDI fragmentation significantly compared to geopolitically distant intermediaries (controlling as in main analysis).
  - Result: No significant impact found (supplementary result omitted for brevity and available upon request).
- Puzzle and interpretation:
  - If ownership links are public, host countries could, in principle, trace ultimate owners; changing intermediary mix should not prevent tracing.
  - Possible explanation: ultimate parents may not expect host countries to do extensive due diligence on the ownership links of their immediate investors.
  - Evidence is inconclusive regarding whether intermediary mix changes materially affect realized FDI outcomes.
- Research caveat: The author leaves the open question of whether altering intermediary mixes meaningfully reduces potential FDI barriers to future research.

### Key takeaway
- When ultimate parents invest indirectly, there is evidence that firms facing larger geopolitical distance to host destinations choose intermediaries that are geopolitically closer to those destinations (consistent with a strategy to circumvent potential FDI barriers).
- This shift is present in the pooled sample when focusing on ultimate parents that are geopolitically distant, and is evident in the US sub-sample (notably stronger for geopolitically distant US parents).
- However, there is no clear evidence that these intermediary mix changes translate into a significant reduction in FDI fragmentation outcomes; the effect on realized FDI remains to be established.

*Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024179-print-pdf.pdf*

### 8.2    Tables

### 8.2    Tables

### Correlation matrix (Table A1)
- Variables: ISH real FDI, GUO real FDI, ISH vertical real FDI, GUO vertical real FDI, ISH FDI, GUO FDI.
- Correlations (matrix entries as presented):
  - ISH real FDI with:
    - ISH real FDI: 1
    - GUO real FDI: 0.766
    - ISH vertical real FDI: 0.989
    - GUO vertical real FDI: 0.779
    - ISH FDI: 0.839
    - GUO FDI: 0.663
  - GUO real FDI with:
    - ISH real FDI: 0.766
    - GUO real FDI: 1
    - ISH vertical real FDI: 0.751
    - GUO vertical real FDI: 0.976
    - ISH FDI: 0.662
    - GUO FDI: 0.778
  - ISH vertical real FDI with:
    - ISH real FDI: 0.989
    - GUO real FDI: 0.751
    - ISH vertical real FDI: 1
    - GUO vertical real FDI: 0.775
    - ISH FDI: 0.823
    - GUO FDI: 0.650
  - GUO vertical real FDI with:
    - ISH real FDI: 0.779
    - GUO real FDI: 0.976
    - ISH vertical real FDI: 0.775
    - GUO vertical real FDI: 1
    - ISH FDI: 0.678
    - GUO FDI: 0.774
  - ISH FDI with:
    - ISH real FDI: 0.839
    - GUO real FDI: 0.662
    - ISH vertical real FDI: 0.823
    - GUO vertical real FDI: 0.678
    - ISH FDI: 1
    - GUO FDI: 0.845
  - GUO FDI with:
    - ISH real FDI: 0.663
    - GUO real FDI: 0.778
    - ISH vertical real FDI: 0.650
    - GUO vertical real FDI: 0.774
    - ISH FDI: 0.845
    - GUO FDI: 1
- Note: ISH refers to immediate shareholder while GUO refers to global ultimate owner. ISH FDI refers to bilateral country-level immediate total FDI, while GUO FDI refers to bilateral country-level ultimate total FDI.

### Determinants of real immediate FDI (Table A2: OLS — selected coefficients and inference)
- Dependent variable: Immediate Log(FDI).
- Key coefficients (columns (1)-(4) reported with standard errors and significance):
  - Distance: -0.239 ∗∗∗ (0.00715); -0.233 ∗∗∗ (0.00728); -0.239 ∗∗∗ (0.00715); -0.229 ∗∗∗ (0.00722)
  - Trade agreement: 0.933 ∗∗∗ (0.0589); 0.877 ∗∗∗ (0.0594); 0.919 ∗∗∗ (0.0591); 0.811 ∗∗∗ (0.0594)
  - Common language: 1.206 ∗∗∗ (0.0794); 1.191 ∗∗∗ (0.0791); 1.208 ∗∗∗ (0.0794); 1.171 ∗∗∗ (0.0792)
  - Common border: 1.756 ∗∗∗ (0.120); 1.733 ∗∗∗ (0.121); 1.742 ∗∗∗ (0.120); 1.711 ∗∗∗ (0.122)
  - Colonial link: 1.088 ∗∗∗ (0.138); 1.124 ∗∗∗ (0.138); 1.087 ∗∗∗ (0.138); 1.192 ∗∗∗ (0.141)
  - Period dummies:
    - 2014-2017: 0.351 ∗∗∗ (0.0225); 0.354 ∗∗∗ (0.0225); 0.351 ∗∗∗ (0.0225); 0.378 ∗∗∗ (0.0227)
    - 2018-2021: 0.690 ∗∗∗ (0.0286); 0.697 ∗∗∗ (0.0286); 0.691 ∗∗∗ (0.0286); 0.753 ∗∗∗ (0.0292)
  - Additional variables reported in some specifications:
    - same_rank: 0.245 ∗∗∗ (0.0521) [column unspecified]
    - same_un_vote: 0.0971 (0.0676)
    - geopol_prox: 0.501 ∗∗∗ (0.0580)
- Controls and fixed effects:
  - Source Country FE: Yes (all columns)
  - Host Country FE: Yes (all columns)
  - Other controls: Host and Source exchange rate with respect to the USD and the volatility of Host and Source exchange rate in each year, annual population and GDP of Source and Host countries.
  - Reference time period: 2009-2013.
- Observations: 95552; 95529; 95529; 55295 (as presented across columns).
- Significance notation: ∗ p<0.10, ∗∗ p<0.05, ∗∗∗ p<0.01.

### Real immediate FDI fragmentation along different measures of distance (Table A3: OLS — selected coefficients)
- Dependent variable: Immediate Log(FDI).
- Coefficients and interactions (columns (1)-(4)):
  - Distance: -0.0277 (0.292)
  - Period dummies:
    - 2014-2017: 0.497 ∗∗∗ (0.0277); 0.543 ∗∗∗ (0.0250); 0.590 ∗∗∗ (0.0321); 0.517 ∗∗∗ (0.0217)
    - 2018-2021: 0.983 ∗∗∗ (0.0339); 1.029 ∗∗∗ (0.0321); 1.077 ∗∗∗ (0.0414); 0.987 ∗∗∗ (0.0280)
  - Interactions:
    - Distance×2014-2017: 0.00252 (0.00375)
    - Distance×2018-2021: -0.00149 (0.00458)
    - same_rank×2014-2017: -0.0673 ∗∗ (0.0341)
    - same_rank×2018-2021: -0.126 ∗∗∗ (0.0422)
    - same_un_vote×2014-2017: -0.110 ∗∗∗ (0.0376)
    - same_un_vote×2018-2021: -0.146 ∗∗∗ (0.0472)
    - geopol_prox: -0.105 ∗∗ (0.0435)
    - geopol_prox×2014-2017: -0.0245 (0.0350)
    - geopol_prox×2018-2021: -0.103 ∗∗ (0.0436)
- Country Pair FE: Yes (all columns)
- Observations: 94556; 94569; 45569; 45569; 4295 (as presented)
- Note: Other controls and reference period as in Table A2.

### Real ultimate FDI fragmentation along different measures of distance (Table A4: OLS — selected coefficients)
- Dependent variable: Immediate Log(FDI) (columns (1)-(4) show alternative specifications).
- Coefficients:
  - Distance: -0.0763 (0.296)
  - Period dummies:
    - 2014-2017: 0.551 ∗∗∗ (0.0286); 0.548 ∗∗∗ (0.0254); 0.600 ∗∗∗ (0.0325); 0.524 ∗∗∗ (0.0221)
    - 2018-2021: 1.054 ∗∗∗ (0.0350); 1.021 ∗∗∗ (0.0325); 1.078 ∗∗∗ (0.0418); 0.995 ∗∗∗ (0.0285)
  - Interactions:
    - Distance×2014-2017: -0.00151 (0.00383)
    - Distance×2018-2021: -0.00758 (0.00466)
    - same_rank×2014-2017: -0.00933 (0.0352)
    - same_rank×2018-2021: -0.0257 (0.0432)
    - same_un_vote×2014-2017: -0.0819 ∗∗ (0.0383)
    - same_un_vote×2018-2021: -0.0964 ∗∗ (0.0478)
    - geopol_prox: -0.121 ∗∗∗ (0.0448)
    - geopol_prox×2014-2017: -0.0679 ∗ (0.0367)
    - geopol_prox×2018-2021: 0.0304 (0.0453)
- Country Pair FE: Yes (all columns)
- Observations: 94659; 94662; 94629; 46629; 4306 (as presented)
- Note: Other controls and reference period as in Table A2.

### Sectoral fragmentation results — Immediate and Ultimate FDI (Tables A5–A8: selected highlights)
- Table A5: Immediate Real FDI Fragmentation (Agriculture)
  - Dependent variable: Real FDI (immediate).
  - Distance: -4.911 ∗∗∗ (1.898)
  - Period coefficients and others (selected):
    - 2014-2017: 0.702 (0.460); 1.410 ∗ (0.808); 3.000 ∗∗∗ (1.094); 1.295 ∗ (0.738)
    - 2018-2021: -0.166 (0.688); 0.420 (1.152); 2.955 ∗∗∗ (1.016); 0.994 (1.087)
    - Distance×2014-2017: 0.0708 ∗ (0.0412)
    - Distance×2018-2021: 0.265 ∗∗∗ (0.0818)
    - same_UN_vote×2014-2017: -1.860 ∗ (1.100)
    - Observations: 36604; 27242; 72427; 24044 (as presented)
- Table A6: Immediate Real FDI Fragmentation (Transport)
  - Dependent variable: Real FDI (immediate).
  - Distance: 2.066 (1.352)
  - Period coefficients:
    - 2014-2017: 0.274 ∗∗ (0.134); 0.330 ∗∗ (0.166); 0.330 ∗∗ (0.158); -0.106 (0.194)
    - 2018-2021: -0.0753 (0.367); -0.678 (0.625); -1.054 ∗∗ (0.517); -1.115 ∗ (0.593)
    - same_rank×2014-2017: 0.641 ∗∗ (0.290)
    - same_rank×2018-2021: 1.945 ∗∗∗ (0.675)
    - geopol_prox: -2.610 ∗∗∗ (0.478)
    - Observations: 12792; 13920; 13920; 13500 (as presented)
- Table A7: Ultimate Real FDI Fragmentation (Agriculture)
  - Dependent variable: Real FDI (ultimate).
  - Distance: 3.065 (2.852)
  - Period coefficients:
    - 2014-2017: 0.902 ∗∗ (0.371); 1.565 ∗∗∗ (0.432); 1.059 ∗ (0.562); 2.319 ∗∗∗ (0.631)
    - 2018-2021: -0.0572 (0.507); 1.433 ∗∗ (0.663); 2.097 ∗∗ (0.910); 2.162 ∗∗∗ (0.767)
    - Distance×2018-2021: 0.220 ∗∗ (0.0886)
    - same_rank×2014-2017: -1.112 ∗∗∗ (0.428)
    - geopol_prox×2014-2017: -1.869 ∗∗∗ (0.631)
    - Observations: 27603; 40834; 08340; 83264 (as presented)
- Table A8: Ultimate Real FDI Fragmentation (Transport)
  - Dependent variable: Real FDI (ultimate).
  - Distance: -0.794 (0.825)
  - Period coefficients:
    - 2014-2017: -0.225 (0.333); 0.318 (0.256); 0.294 (0.251); 0.331 (0.358)
    - 2018-2021: -0.610 (0.555); 0.0432 (0.594); -0.107 (0.561); 0.329 (0.735)
    - Distance×2014-2017: 0.236 ∗∗∗ (0.0576)
    - Distance×2018-2021: 0.383 ∗∗∗ (0.0722)
    - geopol_prox: -1.778 (1.233)
    - Observations: 9732; 10872; 10872; 10488 (as presented)

### Fragmentation of US FDI (Tables A9–A10: All Industries and Mining — selected coefficients)
- Table A9: Fragmentation of US FDI (All Industries)
  - Dependent variable: US outward FDI.
  - Distance: -0.253 ∗∗∗ (0.0838)
  - Period coefficients:
    - 2014-2017: -0.219 ∗∗ (0.102); 0.0469 (0.0794); -0.0218 (0.0954); -0.0882 ∗ (0.0489)
    - 2018-2021: -0.329 ∗∗∗ (0.124); 0.0219 (0.0691); -0.0590 (0.128); -0.347 ∗∗∗ (0.103)
    - Distance×2014-2017: 0.0198 ∗ (0.0119)
    - Distance×2018-2021: 0.0179 (0.0132)
    - same_rank×2018-2021: -0.281 ∗∗∗ (0.0984)
    - geopol_prox×2014-2017: -0.275 ∗∗∗ (0.0919)
    - geopol_prox×2018-2021: -0.409 ∗∗∗ (0.131)
  - Host Country FE: Yes
  - Observations: 21612; 16121; 61216; 12148 (as presented)
- Table A10: Fragmentation of US FDI (Mining)
  - Dependent variable: US outward FDI.
  - Distance: -0.848 ∗∗∗ (0.191)
  - Period coefficients:
    - 2014-2017: -0.504 ∗ (0.292); 0.142 (0.248); -0.494 (0.365); -0.136 (0.215)
    - 2018-2022: -0.631 ∗ (0.327); 0.142 (0.233); -0.332 (0.242); -0.324 (0.266)
    - Distance×2014-2017: 0.0477 ∗∗ (0.0192)
    - Distance×2018-2022: 0.0598 ∗∗∗ (0.0221)
    - same_rank×2018-2022: -0.635 ∗∗ (0.296)
    - geopol_prox×2018-2022: -0.791 ∗∗ (0.313)
  - Host Country FE: Yes
  - Observations: 12251; 32013; 20132; 1306 (as presented)

### Data Appendix — construction of key measures (section 9 highlights)
- Data sources and general approach:
  - Starting data: CDIS data giving bilateral total FDI positions between source-host country pairs from 2009 to the present.
  - When both host and source are CDIS reporters, inward FDI reported by the host country (from the source country) is used instead of outward FDI reported by the source country.
- Construction of Real FDI (following Damgaard et al., 2019):
  1. For OECD countries that report bilateral SPEs, subtract the FDI position in SPEs from total FDI to obtain bilateral real FDI.
  2. For OECD countries reporting only aggregate SPEs, calculate share of SPE out of total FDIs and apply that share to all bilateral FDI positions reported in the CDIS.
  3. For non-OECD countries and OECD countries that don’t report SPEs:
     - i. For OECD countries (h) that report aggregate inward SPE and non-SPE breakdown, regress (with logs) the share of real FDI out of total FDI on the share of total FDI out of GDP for each period t.
     - ii. Use coefficients to impute share of real FDI for non reporters based on their FDI to GDP ratio. Lower bound real FDI to 0.
     - iii. Assume that the above ratio of real to total FDI is constant across all bilateral FDI for each CDIS country (apply same share to all bilateral positions).
- Construction of Ultimate FDI (following Damgaard and Elkjaer, 2017):
  - Definitions: ISH = immediate shareholder; GUO = global ultimate owner. FDI from an ISH firm = immediate FDI; FDI from a GUO firm = ultimate FDI.
  1. For OECD countries that report real FDI by ultimate investor economy, use that data.
  2. For source-host pairs well-represented by Orbis (Orbis bilateral FDI deviates from CDIS by ≤50%):
     - Compute δ_hi = FDI from GUO in source (i) to host (h) / FDI from ISH in source (i) to host (h).
     - Multiply bilateral CDIS FDI from source (i) to host (h) by δ_hi to obtain ultimate bilateral FDI from (i) to (h).
  3. For pairs where Orbis deviates by >50% from CDIS:
     - Let δ_hi = (∑_i FDI with ultimate investor in source i) / (∑_i FDI with immediate investor in source i).
     - Multiply bilateral CDIS FDI from source (i) to host (h) by δ_hi to obtain ultimate bilateral FDI from (i) to (h).
  4. Scale all bilateral ultimate FDI to ensure sum of all ultimate FDI to host (h) = sum of all immediate FDI to host (h), by multiplying by:
     - (sum of all FDI from immediate to host (h)) / (sum of all FDI from ultimate investor to host (h)).
- Imputation of Vertical FDI:
  1. Using industry information from Orbis, define FDI as vertical if the four digit industry code of the shareholder differs from that of the subsidiary.
  2. For each source-host pair in Orbis, calculate the share of vertical real FDI out of total real FDI by aggregating over shareholder and subsidiary firms.
  3. Apply this share to bilateral CDIS positions to obtain bilateral vertical FDI positions.
  4. Repeat separately for immediate and ultimate FDI.

*Source: wpiea2024179-print-pdf - 8.2    Tables*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024179-print-pdf.pdf_