## ch4annex - 1.2 in 2003 (the first year of data) to 1.3 in

## Source details

**Canonical URL:** [ch4annex - 1.2 in 2003 (the first year of data) to 1.3 in](https://www.imf.org/-/media/files/publications/weo/2023/april/english/ch4annex.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/weo/2023/april/english/ch4annex.pdf.md)
- [Structured JSON version](/-/media/files/publications/weo/2023/april/english/ch4annex.pdf.json)

---

### Key descriptive findings on geopolitics and FDI
- The increase in geopolitical importance for FDI decisions was markedly higher in strategic sectors (+26 percent) than for other sectors (+6 percent).
- Event-study evidence around two UN General Assembly resolutions shows FDI from countries that approved the resolution diverged from and remained higher than FDI from countries that voted against or abstained in the 16 quarters around each resolution:
  - Resolution 68/262 (adopted 27 March 2014) about the territorial integrity of Ukraine.
  - Resolution 72/191 (adopted 19 December 2017) on the situation of human rights in Syria.
- The descriptive evidence suggests geopolitical factors affect multinational corporations' investment decisions.

### Parametric results — baseline gravity estimates
- Estimation method: gravity model for foreign direct investment estimated with Pseudo-Poisson Maximum Likelihood; period 2003-2021.
- Main coefficient findings:
  - A higher ideal point distance (IPD) is associated with lower FDI, whether FDI is measured by number of projects or by value (USD).
  - Geographic distance is also associated with lower FDI.
  - Common legal origins, common language, and colonial/dependency relationship are generally associated with more FDI, but the IPD coefficient remains negative and statistically significant when these controls are included.
- Quantitative interpretation (column 3, number of projects): moving the IPD measure from the 25th to the 75th percentile (equivalent example: from the distance between South Korea and Japan to that between the UK and Russia) is associated with about a 15 percent decline in the number of FDI projects.
- Selected reported coefficients and notes:
  - Period: 2003-2021 (primary analyses); 2003-2018 for brownfield robustness.
  - Ideal point distance, lagged (various specifications): -0.3570***, -0.1448***, -0.1162*** (FDI USD panel); -0.1310***, -0.1157***, -0.1259*** (FDI number of projects panel).
  - Geographic distance examples: -0.6266***, -0.5694***, -0.6168***; -0.5102***, -0.6447***, -0.5368***.
  - Common language examples: 0.4768***, 0.5446***, 0.3911***, 0.5482***.
  - Colonial/dependency relationship examples: 0.4135***, 0.4052***, 0.2805***, 0.5056***.
  - Alternative geopolitical measures: Ideal point distance rank, lagged = -0.3281***; S measure of distance, lagged = -0.2946***; Pi measure of distance, lagged = -0.2809***.
  - Controls with strong associations: Imports coefficient example 0.2012***; Exchange rate (yearly change) examples -2.4574***, -4.3679***.
  - Observations in many specifications: 320,025 (full sample), with sub-sample counts reported (for example, 229,262; 269,436; 115,659; 291,547; 312,830).

### Heterogeneity: where and when IPD matters most
- EMDE involvement:
  - FDI responds to IPD especially when an EMDE is involved either as source or destination.
  - The negative relationship between IPD and FDI is non significantly different from zero if the source or destination country is an AE.
  - The negative relationship is larger than average (approximately twice as large) if the destination or source country is an EMDE.
- Temporal pattern:
  - The relevance of geopolitical distance for FDI declined up to 2017 and increased after 2018 through 2021, producing a U-shaped pattern in the negative coefficients.
  - The difference in semielasticities between the 2009-17 and 2018-21 periods is statistically significant.
- Sectoral heterogeneity:
  - The importance of IPD for FDI is larger for strategic sectors than for other sectors.
- Flow direction:
  - The impact of IPD on FDI flows is especially driven by South-South flows (both source and destination are EMDEs).

### Robustness and extensions
- Results hold across alternative samples:
  - Manufacturing only; services only; excluding country pairs that never registered a FDI; excluding international financial centers; excluding China.
- Results hold with alternative measures of geopolitical distance:
  - Rank of destination country in the IPD distribution; S score; Pi (π) scores (Signorino and Ritter 1999; Häge 2011).
- Additional controls included with robustness:
  - Announcement and implementation of bilateral trade barriers (Global Trade Alerts).
  - Bilateral imports (trade intensity).
  - Yearly change in the bilateral exchange rate.
- Country-pair fixed effects (saturated model) results:
  - When the model includes country-pair fixed effects, the IPD coefficient becomes smaller and loses significance for FDI measured by number of projects (top panel), but remains negative and significant for FDI measured in value (USD, bottom panel).
  - Restricting sample to EMDE destination countries:
    - Baseline specification (EMDEs-destination sample): moving IPD from the 25th to the 75th percentile is associated with a decline of about 30 percent in the number of FDI projects.
    - In the model saturated with country-pair fixed effects, this decline is reduced to 15 percent.

### Robustness to brownfield FDI (cross-border M&A)
- Replacing greenfield measures with brownfield FDI measures up to 2018 from the SDC Platinum database yields qualitatively identical results.
- Online Annex Table 4.1.4 and Figure 4.1.5 confirm robustness for cross-border M&As (number of deals and USD million).

### Definition and coverage notes (select)
- IPD measure: ideal point distance from Bailey, Strezhnev, and Voeten (2017).
- Strategic sectors (summary): includes Manufacture of basic chemicals, fertilizers and nitrogen compounds, plastics and synthetic rubber in primary forms; Manufacture of batteries and accumulators; Manufacture of coke oven products; Manufacture of consumer electronics; Manufacture of domestic appliances; Manufacture of electronic components and boards; Manufacture of general-purpose machinery; Manufacture of measuring, testing, navigating and control equipment; Manufacture of motor vehicles; Manufacture of non-metallic mineral products n.e.c.; Manufacture of pharmaceuticals, medicinal chemical and botanical products; Mining of non-ferrous metal ores; Support activities for petroleum and natural gas extraction; plus specified subsectors of ISIC code 20 (Manufacture of Chemicals and Chemical Products) such as Biological products (except diagnostic), In-Vitro diagnostic substances, Other (Biotechnology), Pesticide, fertilizers & other agricultural chemicals.
- Source and destination country lists for main regression are detailed in the annex.

### Geopolitical index: definition and construction
- Purpose: captures the idea that the vulnerability of an investment project to being relocated should increase with the geo-political distance between the host country and the source country.
- Country-level measure: v_i^geo = ∑_j share^FDI_{ij} * γ^geo(i,j)
  - v_i^geo denotes the country-level geopolitical vulnerability measure.
  - share^FDI_{ij} denotes the estimated share of FDI stock in host-country i from source-country j.
  - γ^geo(i,j) = percentile(IPD(i,j)) / 100, i.e., the percentile of the bilateral IPD amongst all bilateral IPDs across all years.
- Interpretation:
  - Index bounded between 0 and 1.
  - Higher values indicate greater vulnerability.
  - A lower percentile (γ^geo closer to 0) indicates closer geopolitical alignment between source and host.
- Estimation of share^FDI_{ij}:
  - Count number of greenfield FDI from j into i during 2010-2019 (after the GFC and before COVID).
  - Divide by the number of investments in country i over 2010-2019 from all source countries.

### Market power index: definition and construction
- Purpose: captures the idea that it may be harder to relocate projects out a sector in a host country if that host country is a major player in that sector; host countries with market power in many FDI-hosting sectors are less vulnerable.
- Country-sector measure: v_i^mkt = ∑_s share^FDI_{i,s} * γ^mkt(i;s)
  - share^FDI_{i,s} denotes the estimated share of FDI stock in country i and sector s, from all source countries over 2010-2019.
  - γ^mkt(i;s) =
    - 0.5 if country i is a top−10 exporter in sector s
    - 1 if otherwise
- Treatment of non-tradeable sectors:
  - Non-tradeable sectors (e.g., retail, finance) are treated as fully vulnerable (γ^mkt = 1) in baseline calculation.
- Robustness: an alternate calculation using tradeable sectors only yields similar results.

### Strategic index: definition and construction
- Purpose: captures the idea that source-countries may be particularly interested in re-locating investments in strategic sectors for national or economic security reasons; both geo-politically close and distant hosts may be vulnerable.
- Country-sector measure: v_i^strat = ∑_s share^FDI_{i,s} * γ^strat(s)
  - γ^strat(s) =
    - 1 if sector s is strategic
    - 0 if sector s is non−strategic

### Aggregate vulnerability index: combining dimensions
- Aggregate index: v_i^agg = 1/2 ∑_{s,j} share^FDI_{ij,s} * (γ^geo(i,j) * γ^mkt(i;s) + γ^strat(s))
  - share^FDI_{ij,s} denotes the estimated share of FDI stock in country i – sector s and from source-country j, amongst the total FDI stock in that country.
  - Rationale:
    - Market power (γ^mkt) ameliorates geopolitical vulnerability from the source country (γ^geo), so they are multiplied.
    - The strategic dimension (γ^strat) captures a separate aspect of geoeconomic fragmentation (GEF) and is added.
  - Division by 2 ensures the aggregate index is between 0 and 1.
- Proxy for distribution of FDI stocks by host, sector, and source:
  - Count of greenfield investment projects from fDi Markets since the Global Financial Crisis and prior to COVID-19 (2010-2019).
- Bilateral geopolitical distance: measured using the ideal point distance (IPD) (Bailey et al. 2017).
- Export market shares: calculated based on bilateral exports flows from Trade Monitor Data for 2019.
- Baseline assumptions:
  - Hong Kong SAR and Macao SAR are merged with mainland China.
  - Taiwan Province of China is dropped.
  - Financial centers are included.
- Robustness checks and alternate approaches considered:
  - Estimating FDI stocks starting from 2003 rather than 2010.
  - Dropping mining sectors from strategic industries.
  - Calculating indices using only sectors with non-zero exports.
  - Estimating FDI stocks using value of projects, or the number of jobs created.
  - Dropping Hong Kong and Macao SARs rather than merging with China.
  - Dropping countries that are financial centers.

### Annex Table 4.3.2 — Summary of main findings (FDI and growth)
- Strong positive correlation between FDI and growth for vertical FDI countries; this relationship does not hold for horizontal FDI countries.
- The positive correlation among vertical FDI countries is driven entirely by EMDEs.
- Using a proxy based on sector affiliation yields qualitatively similar results, except a positive relationship between FDI and growth is also found for horizontal FDI in AEs.
- Interpretation: Horizontal FDI tend to be more frequent among final goods producers that bring simple (and labor intensive) assembly technology; vertical FDI tend to be concentrated among intermediate goods producers employing more sophisticated (and skill intensive) technology.

### Country-level estimation results (selected coefficients and samples)
- All countries (Columns (1)-(3)):
  - Lagged FDI over GDP: 0.165*** (0.0595)
  - Lagged Log of real GDP: -4.679*** (0.603)
  - Observations: 5274
  - Adj-R2: 0.253
- AEs:
  - Lagged FDI over GDP: 0.0291** (0.0129)
  - Lagged Log of real GDP: -5.051*** (0.682)
  - Observations: 11034
  - Adj-R2: 0.602
- EMDEs:
  - Lagged FDI over GDP: 0.233*** (0.0820)
  - Lagged Log of real GDP: -4.589*** (0.683)
  - Observations: 4171
  - Adj-R2: 0.239
- Note: All specifications include country and year fixed effects. Standard errors clustered at the country level. ***p<0.01, **p<0.05, *p<0.1.

### Country-level estimation by FDI type (Sales Information-Based Classification)
- Horizontal FDI (All / AEs / EMDEs):
  - All: Lagged FDI over GDP: -0.0302 (0.0235); Observations: 948; Adj-R2: 0.344
  - AEs: Lagged FDI over GDP: 0.0116 (0.0132); Observations: 410; Adj-R2: 0.678
  - EMDEs: Lagged FDI over GDP: -0.0535 (0.164); Observations: 538; Adj-R2: 0.282
- Vertical FDI (All / AEs / EMDEs):
  - All: Lagged FDI over GDP: 0.0840** (0.0408); Observations: 1475; Adj-R2: 0.425
  - AEs: Lagged FDI over GDP: 0.0152 (0.0215); Observations: 491; Adj-R2: 0.589
  - EMDEs: Lagged FDI over GDP: 0.186*** (0.0666); Observations: 984; Adj-R2: 0.417

### Industry Information-Based Classification (selected)
- All / AEs / EMDEs considering All FDI:
  - Lagged FDI over GDP: 0.0946 (0.0615); 0.172*** (0.0199); 0.0774 (0.0857)
  - Observations: 1554 (All), 244 (AEs), 1310 (EMDEs)
  - Adj-R2: 0.315 (All), 0.685 (AEs), 0.305 (EMDEs)
- Vertical FDI split:
  - Lagged FDI over GDP: 0.112** (0.0455); 0.006390 (0.0118); 0.218*** (0.0797)
  - Observations: 3102 (All), 651 (AEs), 2451 (EMDEs)
  - Adj-R2: 0.288 (All), 0.589 (AEs), 0.260 (EMDEs)

### Firm-level evidence (matched fDi Markets and Enterprise Surveys)
- Data and scope:
  - More than 180,000 firms in over 150 countries between 2006 and 2022.
  - Firm-level outcomes include employment, sales, investment, and R&D; focus on labor productivity growth over three years.
  - Inter-industry linkages measured using EORA global input-output matrix.
- Baseline regression: ∆lnLP_icjt = β1 ln(FDI_within_cjt−3) + β2 ln(FDI_supplier_cjt−3) + β3 ln(FDI_user_cjt−3) + FE_cj + FE_ct + FE_jt + ε_icjt.
- Baseline results (Columns (1)-(3); coefficients):
  - Intra-industry spillover:
    - All: 0.159* (0.0899)
    - AEs: 0.890** (0.394)
    - EMDEs: 0.141 (0.0922)
  - Backward spillover:
    - All: 0.340** (0.144)
    - AEs: 0.759 (0.931)
    - EMDEs: 0.372** (0.148)
  - Forward spillover:
    - All: -0.277 (0.181)
    - AEs: -1.298 (0.999)
    - EMDEs: -0.288 (0.181)
  - Observations: 129557 (All), 16316 (AEs), 113114 (EMDEs)
  - Adj-R2: 0.178 (All), 0.159 (AEs), 0.178 (EMDEs)
- Extended results by origin of FDI (AEs vs EMDEs) report varying intra-industry, backward, and forward spillovers with statistical differences noted.
- Interpretation:
  - Intra-industry spillovers positive and statistically significant only for AEs.
  - Backward spillover effects positive and statistically significant, particularly in EMDEs.
  - Forward spillover effects negative, not statistically significant.
  - FDI from AEs tends to produce stronger intra-industry spillovers in AEs and about two times stronger backward effects compared with FDI from EMDEs (standardized coefficients).

### Modeling FDI fragmentation (Online Annex 4.4) — GIMF elements and calibration
- Model: Global Integrated Monetary and Fiscal model (GIMF), annual, multi-region DSGE with 8 regions: United States; EU+; other advanced economies; China; Southeast Asia; India and Indonesia; Latin America and the Caribbean; rest of the world.
- Additions: tradable GVC sector to represent global value chain dynamics and FDI links; Non-tariff barriers (NTBs) capture barriers to trade and investment-input flows between opposing blocs.
- Key calibration values:
  - Intertemporal elasticity of substitution: 0.2.
  - Share of liquidity constrained households: 25 percent for United States, EU+, other AEs, and China; 50 percent for remaining regions.
  - Benchmark elasticity of substitution for investment inputs sourced from different foreign regions: 1.5 (alternative value of 3 considered).
  - Demand elasticities: GVC goods demand are relatively inelastic (all well under 1); final goods elasticity combining nontradable and tradable bundle: 0.5; tradable bundle elasticity combining tradable intermediate and GVC goods: 0.95.
- Domestic sector calibration (percent of region's GDP; selected):
  - Share of Global GDP (%, US$): United States 24.4; EU+ 18.9; Other Advanced Economies 16.5; China 16.7; South-east Asia 2.3; Rest of the World 11.6; India and Indonesia 4.5; Latin America 5.1.
  - Domestic Demand — Household Consumption: United States 65.4; EU+ 54.9; Other Advanced Economies 56.3; China 51.7; South-east Asia 58.8; Rest of the World 58.7; India and Indonesia 56.4; Latin America 63.0.
  - Domestic Demand — Private Investment: United States 17.1; EU+ 32.2; Other Advanced Economies 17.5; China 22.5; South-east Asia 24.3; Rest of the World 22.2; India and Indonesia 27.7; Latin America 16.0.
  - Trade — Aggregate Exports: United States 11.5; EU+ 20.1; Other Advanced Economies 23.5; China 17.4; South-east Asia 61.7; Rest of the World 24.9; India and Indonesia 19.9; Latin America 21.0.
  - GVC (domestic use): United States 3.8; EU+ 7.1; Other Advanced Economies 10.5; China 7.5; South-east Asia 23.2; Rest of the World 14.6; India and Indonesia 8.2; Latin America 9.3.
- Construction of geo-political blocs:
  - Countries ranked by closeness to China and the US using bilateral IPD scores averaged over 2017-2021.
  - Six statistics per country computed and aggregated to regions weighted by PPP GDP to form China-leaning and US-leaning groups; least-aligned regions assigned as non-aligned in baseline fragmentation scenario.
- Calibrating productivity losses:
  - Conditional correlation estimated via panel regression: logLP_i,t = β0 + β1 logLP_i,t−1 + β2 (FDI/GDP)_i,t−1 + δ_t + γ_i + ε_i,t, estimated separately for EMDE and AE recipients using 1980-2021 data.

### China — estimated FDI–productivity mapping and fragmentation outcomes
- Conditional correlation (Online Annex Table 4.4.5):
  - Lagged FDI over GDP: -0.00399 (AEs sample), 0.147** (EMDEs sample).
  - Labor productivity (lagged): 0.960*** (AEs), 0.963*** (EMDEs).
  - Constant: 0.464*** (AEs), 0.366*** (EMDEs).
  - Observations: 1262 (AEs), 4313 (EMDEs).
  - Rsquared: 0.997 (AEs), 0.998 (EMDEs).
  - Period: 1980-2021.
- Interpretation: The coefficient of 0.147 implies that a 10 percentage point increase in FDI inflows to GDP is associated with a 1.47 percent increase in labor productivity levels in EMDEs; corresponding coefficient small and insignificant for AEs.
- Mapping to macro model:
  - Use sequences of import of investment inputs as proxy for (FDI/GDP) changes under fragmentation vs no-fragmentation.
  - Compute labor productivity changes using β2 and β1 over s = 10 (ten-year cumulation).
  - Feed estimated labor productivity losses back into the model.
- Fragmentation scenario: Barriers between the U.S. and China only:
  - Explicit barriers reduce investment input flows by 50 percent between China and the United States only.
  - Outcome summary:
    - Losses mainly arise for EMDE regions in the China bloc; non-aligned or US-bloc EMDE regions may see some increase in flows due to diversion.
    - Ten-year cumulation (s = 10) used to capture partial substitution of knowledge transfer.
    - Under benchmark elasticity 1.5: gains from diversion to non-aligned regions are dominated by negative impact of reduced external demand, so non-aligned regions experience small output losses in the US–China barriers scenario.
- Policy uncertainty regimes and modeling:
  - Certainty case: investors expect non-aligned regions will remain non-aligned permanently.
  - Uncertainty case: investors assign positive probability that non-aligned regions will join one bloc or the other.
  - Uncertainty modeled as implicit partial barriers; illustrative mapping uses implicit barriers equal to half those faced by regions in the two blocs (50 percent of explicit barriers); alternate cases (e.g., 25 percent) yield smaller losses.

### Balance sheet exposure to fragmentation risk: data and methodology
- Data sources:
  - Bilateral portfolio equity and debt investments from IMF Coordinated Portfolio Investments Statistics Survey (CPIS).
  - Bilateral cross-border loans to non-banks from BIS International Locational Banking Statistics (residency basis).
- Imputations and construction notes:
  - For India and Indonesia, BIS cross-border bank loans to non-banks were imputed after 2015 and 2016 respectively.
  - For Germany and Japan, cross-border bank loans to non-banks estimated using BIS bilateral distribution of total cross-border bank investments.
  - For China, estimates of cross-border bank loans taken from Horn and others (2021).
  - For Argentina, Russia and Saudi Arabia, BIS data not available; exposures capture only CPIS portfolio investment data (interpreted as a lower bound).
- Reallocation of CPIS positions:
  - CPIS positions reallocated to parent country using matrices based on fund holdings from Coppola and others (2021); matrices available from 2007-2021; 2007 matrix used for 2001–2007.
- Political proximity index:
  - Political proximity data from Bailey and others (2017).
  - Ideal Point Distance (IPD) averaged between 2002 and 2021 and normalized into γ in [0,1], where 0 (1) is most (least) politically distant.
  - IPDs rebased so that average IPD across destinations when destination is US equals average IPD across all destinations excluding the US.
- Definition of gross balance sheet exposure to fragmentation risk:
  - fragmentation_exposure_i,t = ∑ ( p_i,j,t − γ_i,j p_i,j,t )_j / X_i,t
    - p_i,j,t is bilateral non-FDI cross-border position (assets plus liabilities) for country i with country j at time t;
    - γ_i,j p_i,j,t is the politically-weighted version of that position;
    - X_i,t is a normalization variable (either nominal GDP in US dollars (GDP_i,t) or total cross-border positions).
  - GDP data from the WEO database.
- Sample coverage and aggregates:
  - Final sample: 38 countries (23 AE and 15 EM) accounting for 86 percent of world GDP.
  - Cross-border positions constructed represent 70.8% of total external assets and 59.3% of total external liabilities among countries in the sample (relative to Lane and Milesi-Ferretti (2018) International Investment Position data).
- Robustness and additional results:
  - Two alternative IPD transformations tested: (i) discrete measure (weights 0, 0.5, 1 by quartiles); (ii) continuous “rank” measure normalized to [0,1].
  - Baseline political proximity measure delivers more conservative results.
  - Using Häge (2011) political proximity measure increases estimated exposures relative to baseline by a factor of 1.5 for AE and roughly 2 for EM in 2021.
  - Net exposures over last 20 years:
    - AEs accumulated a positive net exposure to fragmentation risks (6 percent of GDP in 2021).
    - EM have become increasingly liable to politically distant creditors (-8 percent of GDP).
  - Concentration risks:
    - The 5 percent most politically distant countries in EM account for 20 percent of their gross mismatch (against 1 percent for AE).
  - As of 2021:
    - Assets exposures represented 9 percent of the financial system’s total assets in G20 countries (on average).
    - Liabilities exposures accounted for 8% of the total credit going to the non-financial sector.

*Source: Online Annex 4.1 and Online Annex Tables 4.1.1–4.1.5; Online Annex Tables 4.3.1–4.4.5 and related text (Ch. 4 annex), World Economic Outlook, International Monetary Fund, April 2023.*

### 1.2 in 2003 (the first year of data) to 1.3 in

### ch4annex - 1.2 in 2003 (the first year of data) to 1.3 in

### Key descriptive findings on geopolitics and FDI
- The increase in geopolitical importance for FDI decisions was markedly higher in strategic sectors (+26 percent) than for other sectors (+6 percent).
- Event-study evidence around two UN General Assembly resolutions shows FDI from countries that approved the resolution diverged from and remained higher than FDI from countries that voted against or abstained in the 16 quarters around each resolution.
  - Resolution 68/262 (adopted 27 March 2014) about the territorial integrity of Ukraine.
  - Resolution 72/191 (adopted 19 December 2017) on the situation of human rights in Syria.
- The descriptive evidence suggests geopolitical factors affect multinational corporations' investment decisions.

### Parametric results — baseline gravity estimates
- Estimation method: gravity model for foreign direct investment estimated with Pseudo-Poisson Maximum Likelihood; period 2003-2021.
- Main coefficient findings:
  - A higher ideal point distance (IPD) is associated with lower FDI, whether FDI is measured by number of projects or by value (USD).
  - Geographic distance is also associated with lower FDI.
  - Common legal origins, common language, and colonial/dependency relationship are generally associated with more FDI, but the IPD coefficient remains negative and statistically significant when these controls are included.
- Quantitative interpretation (column 3, number of projects): moving the IPD measure from the 25th to the 75th percentile (equivalent example: from the distance between South Korea and Japan to that between the UK and Russia) is associated with about a 15 percent decline in the number of FDI projects.

### Heterogeneity: where and when IPD matters most
- FDI responds to IPD especially when an EMDE (emerging market and developing economy) is involved either as source or destination.
- The negative relationship between IPD and FDI:
  - Is non significantly different from zero if the source or destination country is an AE (advanced economy).
  - Is larger than average (approximately twice as large) if the destination or source country is an EMDE.
- Temporal pattern:
  - The relevance of geopolitical distance for FDI declined up to 2017 and increased after 2018 through 2021, producing a U-shaped pattern in the negative coefficients.
  - The difference in semielasticities between the 2009-17 and 2018-21 periods is statistically significant.
- Sectoral heterogeneity:
  - The importance of IPD for FDI is larger for strategic sectors than for other sectors.
- Flow direction:
  - The impact of IPD on FDI flows is especially driven by South-South flows (both source and destination are EMDEs).

### Robustness and extensions
- Results hold across alternative samples:
  - Manufacturing only; services only; excluding country pairs that never registered a FDI; excluding international financial centers; excluding China.
- Results hold with alternative measures of geopolitical distance:
  - Rank of destination country in the IPD distribution; S score; Pi (π) scores (Signorino and Ritter 1999; Häge 2011).
- Additional controls included with robustness:
  - Announcement and implementation of bilateral trade barriers (Global Trade Alerts).
  - Bilateral imports (trade intensity).
  - Yearly change in the bilateral exchange rate.
- Country-pair fixed effects (saturated model) results:
  - When the model includes country-pair fixed effects, the IPD coefficient becomes smaller and loses significance for FDI measured by number of projects (top panel), but remains negative and significant for FDI measured in value (USD, bottom panel).
  - Restricting sample to EMDE destination countries: increasing geopolitical distance is associated with a subsequent decline in FDI regardless of measurement (number of projects or value). Quantitatively:
    - Baseline specification (EMDEs-destination sample): moving IPD from the 25th to the 75th percentile is associated with a decline of about 30 percent in the number of FDI projects.
    - In the model saturated with country-pair fixed effects, this decline is reduced to 15 percent.

### Robustness to brownfield FDI (cross-border M&A)
- Replacing greenfield measures with brownfield FDI measures up to 2018 from the SDC Platinum database yields qualitatively identical results.
- Online Annex Table 4.1.4 and Figure 4.1.5 confirm robustness for cross-border M&As (number of deals and USD million).

### Main numerical estimates reported (selected coefficients and sample notes)
- Period: 2003-2021 (primary analyses); 2003-2018 for brownfield robustness.
- Sample sizes and selected coefficient estimates from Online Annex Table 4.1.1 and robustness tables:
  - Ideal point distance, lagged (various specifications): values reported include -0.3570***, -0.1448***, -0.1162*** (FDI USD panel), and -0.1310***, -0.1157***, -0.1259*** (FDI number of projects panel) among others.
  - Geographic distance coefficients: examples include -0.6266***, -0.5694***, -0.6168***, and -0.5102***, -0.6447***, -0.5368***.
  - Common language coefficients: examples include 0.4768***, 0.5446***, 0.3911***, 0.5482***.
  - Colonial or dependency relationship coefficients: examples include 0.4135***, 0.4052***, 0.2805***, 0.5056***.
  - Robustness alternative measures: Ideal point distance rank, lagged = -0.3281***; S measure of distance, lagged = -0.2946***; Pi measure of distance, lagged = -0.2809***.
  - Controls with strong associations: Imports coefficient example 0.2012***; Exchange rate (yearly change) examples -2.4574***, -4.3679***.
- Observations in many specifications: 320,025 (full sample), with various sub-sample observation counts reported (for example, 229,262; 269,436; 115,659; 291,547; 312,830).

### Definition and coverage notes
- IPD measure: ideal point distance from Bailey, Strezhnev, and Voeten (2017).
- Strategic sectors (list summary from SEC 3 definitions): includes Manufacture of basic chemicals, fertilizers and nitrogen compounds, plastics and synthetic rubber in primary forms; Manufacture of batteries and accumulators; Manufacture of coke oven products; Manufacture of consumer electronics; Manufacture of domestic appliances; Manufacture of electronic components and boards; Manufacture of general-purpose machinery; Manufacture of measuring, testing, navigating and control equipment; Manufacture of motor vehicles; Manufacture of non-metallic mineral products n.e.c.; Manufacture of pharmaceuticals, medicinal chemical and botanical products; Mining of non-ferrous metal ores; Support activities for petroleum and natural gas extraction; plus specified subsectors of ISIC code 20 (Manufacture of Chemicals and Chemical Products) such as Biological products (except diagnostic), In-Vitro diagnostic substances, Other (Biotechnology), Pesticide, fertilizers & other agricultural chemicals.
- Source and destination country lists for main regression are detailed in the annex (extensive country lists for source and destination coverage).

*Source: Online Annex 4.1 and Online Annex Tables 4.1.1–4.1.5 (Ch. 4 annex), World Economic Outlook, International Monetary Fund, April 2023.*

### 1. Geopolitical index, which captures the idea that the vulnerability of an investment project to

### 1. Geopolitical index, which captures the idea that the vulnerability of an investment project to

### Geopolitical index: definition and construction
- Purpose: captures the idea that the vulnerability of an investment project to being relocated should increase with the geo-political distance between the host country and the source country.
- Country-level measure: v_i^geo = ∑_j share^FDI_{ij} * γ^geo(i,j)
  - v_i^geo denotes the country-level geopolitical vulnerability measure.
  - share^FDI_{ij} denotes the estimated share of FDI stock in host-country i from source-country j.
  - γ^geo(i,j) = percentile(IPD(i,j)) / 100, i.e., the percentile of the bilateral IPD amongst all bilateral IPDs across all years.
- Interpretation:
  - Index bounded between 0 and 1.
  - Higher values indicate greater vulnerability.
  - A lower percentile (γ^geo closer to 0) indicates closer geopolitical alignment between source and host.
- Estimation of share^FDI_{ij}:
  - Count number of greenfield FDI from j into i during 2010-2019 (after the GFC and before COVID).
  - Divide by the number of investments in country i over 2010-2019 from all source countries.

### Market power index: definition and construction
- Purpose: captures the idea that it may be harder to relocate projects out a sector in a host country if that host country is a major player in that sector; host countries with market power in many FDI-hosting sectors are less vulnerable.
- Country-sector measure: v_i^mkt = ∑_s share^FDI_{i,s} * γ^mkt(i;s)
  - share^FDI_{i,s} denotes the estimated share of FDI stock in country i and sector s, from all source countries over 2010-2019.
  - γ^mkt(i;s) =
    - 0.5 if country i is a top−10 exporter in sector s
    - 1 if otherwise
- Treatment of non-tradeable sectors:
  - A significant share of FDI in many host countries are in non-tradeable sectors (e.g., retail, finance).
  - These investments are treated as fully vulnerable (γ^mkt = 1) in all countries in the baseline calculation.
- Robustness: an alternate calculation using tradeable sectors only yields similar results.

### Strategic index: definition and construction
- Purpose: captures the idea that source-countries may be particularly interested in re-locating investments in strategic sectors for national or economic security reasons; both geo-politically close and distant hosts may be vulnerable.
- Country-sector measure: v_i^strat = ∑_s share^FDI_{i,s} * γ^strat(s)
  - γ^strat(s) =
    - 1 if sector s is strategic
    - 0 if sector s is non−strategic

### Aggregate vulnerability index: combining dimensions
- Aggregate index: v_i^agg = 1/2 ∑_{s,j} share^FDI_{ij,s} * (γ^geo(i,j) * γ^mkt(i;s) + γ^strat(s))
  - share^FDI_{ij,s} denotes the estimated share of FDI stock in country i – sector s and from source-country j, amongst the total FDI stock in that country.
  - Rationale:
    - Market power (γ^mkt) is considered to ameliorate geopolitical vulnerability from the source country (γ^geo), so they are multiplied.
    - The strategic dimension (γ^strat) captures a separate aspect of geoeconomic fragmentation (GEF) and is added.
  - Division by 2 ensures the aggregate index is between 0 and 1.

### Data sources and measurement choices
- Proxy for distribution of FDI stocks by host, sector, and source:
  - Count of greenfield investment projects from fDi Markets since the Global Financial Crisis and prior to COVID-19 (2010-2019).
- Bilateral geopolitical distance:
  - Measured using the ideal point distance (IPD) (Bailey et al. 2017).
- Export market shares:
  - Calculated based on bilateral exports flows from Trade Monitor Data for 2019.
- Baseline assumptions:
  - Hong Kong SAR and Macao SAR are merged with mainland China.
  - Taiwan Province of China is dropped.
  - Financial centers are included.

### Robustness checks and alternate approaches considered
- Alternate thresholds and functional forms are tested; the baseline index is broadly correlated with these alternatives and chapter findings are robust.
- Specific alternate approaches considered:
  - Estimating FDI stocks starting from 2003 (first available year in fDi Markets) rather than 2010.
  - Dropping mining sectors from strategic industries when calculating strategic vulnerability.
  - Calculating each index using only sectors with non-zero exports.
  - Estimating FDI stocks using value of projects, or the number of jobs created.
  - Dropping Hong Kong and Macao SARs rather than merging with China.
  - Dropping countries that are financial centers.

*Source: ch4annex - 1. Geopolitical index, which captures the idea that the vulnerability of an investment project to — Online Annex from the referenced PDF chapter.*

### Annex Table 4.3.2

### Annex Table 4.3.2

### Summary of main findings
- There is a strong positive correlation between FDI and growth for vertical FDI countries; this relationship does not hold for horizontal FDI countries.
- The positive correlation among vertical FDI countries is driven entirely by EMDEs.
- Using a proxy based on sector affiliation (Online Annex Table 4.3.3) yields qualitatively similar results, except a positive relationship between FDI and growth is also found for horizontal FDI in AEs.
- Interpretation: Horizontal FDI tend to be more frequent among final goods producers that bring simple (and labor intensive) assembly technology; vertical FDI tend to be concentrated among intermediate goods producers employing more sophisticated (and skill intensive) technology.

### Country-level estimation results (All countries; AEs; EMDEs)
- Columns (1)-(3) — All FDI countries (coefficients; standardized coefficients):
  - Lagged FDI over GDP: 0.165*** (0.0595)
  - Lagged Log of real GDP: -4.679*** (0.603)
  - Lagged FDI over GDP (standardized): 0.159*** (0.0595)
  - Lagged Log of real GDP (standardized): -0.985*** (0.603)
  - Observations: 5274
  - Adj-R2: 0.253
- Column (2) — AEs:
  - Lagged FDI over GDP: 0.0291** (0.0129)
  - Lagged Log of real GDP: -5.051*** (0.682)
  - Lagged FDI over GDP (standardized): 0.067** (0.0129)
  - Lagged Log of real GDP (standardized): -0.575*** (0.682)
  - Observations: 11034
  - Adj-R2: 0.602
- Column (3) — EMDEs:
  - Lagged FDI over GDP: 0.233*** (0.0820)
  - Lagged Log of real GDP: -4.589*** (0.683)
  - Lagged FDI over GDP (standardized): 0.180*** (0.0820)
  - Lagged Log of real GDP (standardized): -0.763*** (0.683)
  - Observations: 4171
  - Adj-R2: 0.239
- Note: All specifications include country and year fixed effects. Columns (1)-(3) consider all countries, advanced economies, emerging and developing market economies, respectively. Standard errors are clustered at the country level. ***p<0.01, **p<0.05, *p<0.1.

### Country-level estimation results by FDI type (Sales Information-Based Classification)
- Horizontal FDI (Columns (1)-(3); coefficients):
  - All:
    - Lagged FDI over GDP: -0.0302 (0.0235)
    - Lagged Log of real GDP: -4.230*** (1.488)
  - AEs:
    - Lagged FDI over GDP: 0.0116 (0.0132)
    - Lagged Log of real GDP: -7.570*** (1.981)
  - EMDEs:
    - Lagged FDI over GDP: -0.0535 (0.164)
    - Lagged Log of real GDP: -4.575*** (1.409)
  - Horizontal (standardized coefficients):
    - All: Lagged FDI over GDP: -0.040 (0.0235); Lagged Log of real GDP: -1.203*** (1.488)
    - AEs: Lagged FDI over GDP: 0.033 (0.0132); Lagged Log of real GDP: -0.976*** (1.981)
    - EMDEs: Lagged FDI over GDP: -0.028 (0.164); Lagged Log of real GDP: -0.997*** (1.409)
  - Observations: 948 (All), 410 (AEs), 538 (EMDEs)
  - Adj-R2: 0.344 (All), 0.678 (AEs), 0.282 (EMDEs)
- Vertical FDI (Columns (4)-(6); coefficients):
  - All:
    - Lagged FDI over GDP: 0.0840** (0.0408)
    - Lagged Log of real GDP: -3.008*** (0.794)
  - AEs:
    - Lagged FDI over GDP: 0.0152 (0.0215)
    - Lagged Log of real GDP: -3.715** (1.647)
  - EMDEs:
    - Lagged FDI over GDP: 0.186*** (0.0666)
    - Lagged Log of real GDP: -3.489*** (0.995)
  - Vertical (standardized coefficients):
    - All: Lagged FDI over GDP: 0.109** (0.0408); Lagged Log of real GDP: -0.795*** (0.794)
    - AEs: Lagged FDI over GDP: 0.037 (0.0215); Lagged Log of real GDP: -0.391** (1.647)
    - EMDEs: Lagged FDI over GDP: 0.138*** (0.0666); Lagged Log of real GDP: -0.680*** (0.995)
  - Observations: 1475 (All), 491 (AEs), 984 (EMDEs)
  - Adj-R2: 0.425 (All), 0.589 (AEs), 0.417 (EMDEs)
- Note: Columns (1)-(3) consider horizontal FDI countries only. Columns (4)-(6) consider vertical FDI countries only. Standard errors are clustered at the country level. ***p<0.01, **p<0.05, *p<0.1.

### Country-level estimation results (Industry Information-Based Classification; Online Annex Table 4.3.3)
- Columns (1)-(3) — All / AEs / EMDEs considering All FDI:
  - Lagged FDI over GDP: 0.0946 (0.0615); 0.172*** (0.0199); 0.0774 (0.0857)
  - Lagged Log of real GDP: -3.847*** (0.938); -5.683*** (0.431); -3.172*** (0.838)
  - Standardized: Lagged FDI over GDP: 0.063 (0.0615); 0.174*** (0.0199); 0.047 (0.0857)
  - Standardized: Lagged Log of real GDP: -0.951*** (0.938); -0.701*** (0.431); -0.702*** (0.838)
  - Observations: 1554 (All), 244 (AEs), 1310 (EMDEs)
  - Adj-R2: 0.315 (All), 0.685 (AEs), 0.305 (EMDEs)
- Columns (4)-(6) — Vertical FDI split:
  - Lagged FDI over GDP: 0.112** (0.0455); 0.006390 (0.0118); 0.218*** (0.0797)
  - Lagged Log of real GDP: -4.312*** (0.869); -2.387 (2.302); -4.869*** (0.980)
  - Standardized: Lagged FDI over GDP: 0.146** (0.0455); 0.0170 (0.0118); 0.197*** (0.0797)
  - Standardized: Lagged Log of real GDP: -1.019*** (0.869); -0.255 (2.302); -0.861*** (0.980)
  - Observations: 3102 (All), 651 (AEs), 2451 (EMDEs)
  - Adj-R2: 0.288 (All), 0.589 (AEs), 0.260 (EMDEs)
- Note: All specifications include country and year fixed effects. ***p<0.01, **p<0.05, *p<0.1.

### Firm-level evidence (Online Annex Table 4.3.4)
- Data and methods:
  - Matched fDi Markets Database with World Bank Enterprise Surveys covering more than 180,000 firms in over 150 countries between 2006 and 2022.
  - Firm-level outcomes include employment, sales, investment, and R&D expenditures; analysis focuses on labor productivity growth over three years.
  - Inter-industry linkages measured using EORA global input-output matrix to construct weighted sums of FDIs across input or output sectors (definitions for forward and backward linkages provided).
  - Baseline regression: ∆lnLP_icjt = β1 ln(FDI_within_cjt−3) + β2 ln(FDI_supplier_cjt−3) + β3 ln(FDI_user_cjt−3) + FE_cj + FE_ct + FE_jt + ε_icjt.
  - Fixed effects: country-sector, country-year, and sector-year. Standard errors clustered at country-sector and country-year.
- Baseline estimation results (Columns (1)-(3); coefficients):
  - Intra-industry spillover:
    - All: 0.159* (0.0899)
    - AEs: 0.890** (0.394)
    - EMDEs: 0.141 (0.0922)
  - Backward spillover:
    - All: 0.340** (0.144)
    - AEs: 0.759 (0.931)
    - EMDEs: 0.372** (0.148)
  - Forward spillover:
    - All: -0.277 (0.181)
    - AEs: -1.298 (0.999)
    - EMDEs: -0.288 (0.181)
  - Standardized coefficients (columns 1-3):
    - Intra-industry spillover: All: 0.028* (0.0899); AEs: 0.217** (0.394); EMDEs: 0.025 (0.0922)
    - Backward spillover: All: 0.107** (0.144); AEs: 0.349 (0.931); EMDEs: 0.114** (0.148)
    - Forward spillover: All: -0.085 (0.181); AEs: -0.557 (0.999); EMDEs: -0.086 (0.181)
  - Observations: 129557 (All), 16316 (AEs), 113114 (EMDEs)
  - Adj-R2: 0.178 (All), 0.159 (AEs), 0.178 (EMDEs)
- Extended results (Columns (4)-(6) FDI from AEs; Columns (7)-(9) FDI from EMDEs):
  - Intra-industry spillover (FDI from AEs):
    - All: 0.139 (0.100)
    - AEs: 0.768* (0.438)
    - EMDEs: 0.125 (0.103)
  - Backward spillover (FDI from AEs):
    - All: 0.293* (0.149)
    - AEs: 0.758 (1.124)
    - EMDEs: 0.285* (0.151)
  - Forward spillover (FDI from AEs):
    - All: -0.285* (0.168)
    - AEs: -0.734 (1.015)
    - EMDEs: -0.273 (0.171)
  - Intra-industry spillover (FDI from EMDEs):
    - All: 0.0909 (0.192)
    - AEs: 1.009 (1.007)
    - EMDEs: 0.0776 (0.196)
  - Backward spillover (FDI from EMDEs):
    - All: 0.256 (0.240)
    - AEs: 0.527 (1.529)
    - EMDEs: 0.451* (0.251)
  - Forward spillover (FDI from EMDEs):
    - All: -0.185 (0.320)
    - AEs: -1.663 (1.989)
    - EMDEs: -0.339 (0.329)
  - Standardized coefficients for extended results are reported similarly in the table.
- Interpretation of firm-level results:
  - Intra-industry spillovers are positive and statistically significant only for AEs (suggesting pro-competitive effects dominate in AEs; market-stealing may prevail in EMDEs).
  - Backward spillover effects are positive and statistically significant, particularly in EMDEs, indicating productivity spillovers via local suppliers in upstream sectors.
  - Forward spillover effects are negative, but not statistically significant.
  - FDI from AEs tends to produce stronger intra-industry spillovers in AEs and yields about two times stronger backward effects compared with FDI from EMDEs (standardized coefficients).

### Modeling FDI fragmentation (Online Annex 4.4) — key elements relevant to interpretation
- The Global Integrated Monetary and Fiscal model (GIMF) used:
  - Annual, multi-region DSGE model with 8 regions: United States; EU+; other advanced economies; China; Southeast Asia; India and Indonesia; Latin America and the Caribbean; rest of the world.
  - A tradable GVC sector was added to represent global value chain dynamics and FDI links.
  - Non-tariff barriers (NTBs) are used to capture barriers to trade and investment-input flows between opposing blocs; the chapter uses the form where source countries impose NTBs on exports to opposing bloc destinations.
- Calibration highlights:
  - Intertemporal elasticity of substitution: 0.2 (common across regions).
  - Share of liquidity constrained households: 25 percent for United States, EU+, other AEs, and China; 50 percent for remaining regions.
  - Benchmark elasticity of substitution for investment inputs sourced from different foreign regions: 1.5 (and alternative value of 3 considered).
  - Demand elasticities: GVC goods demand are relatively inelastic (all well under 1); final goods elasticity combining nontradable and tradable bundle: 0.5; tradable bundle elasticity combining tradable intermediate and GVC goods: 0.95.
- Domestic sector calibration (percent of region's GDP):
  - Share of Global GDP (%, US$): United States 24.4; EU+ 18.9; Other Advanced Economies 16.5; China 16.7; South-east Asia 2.3; Rest of the World 11.6; India and Indonesia 4.5; Latin America 5.1.
  - Domestic Demand — Household Consumption: United States 65.4; EU+ 54.9; Other Advanced Economies 56.3; China 51.7; South-east Asia 58.8; Rest of the World 58.7; India and Indonesia 56.4; Latin America 63.0.
  - Domestic Demand — Private Investment: United States 17.1; EU+ 32.2; Other Advanced Economies 17.5; China 22.5; South-east Asia 24.3; Rest of the World 22.2; India and Indonesia 27.7; Latin America 16.0.
  - Trade — Aggregate Exports: United States 11.5; EU+ 20.1; Other Advanced Economies 23.5; China 17.4; South-east Asia 61.7; Rest of the World 24.9; India and Indonesia 19.9; Latin America 21.0.
  - GVC (domestic use): United States 3.8; EU+ 7.1; Other Advanced Economies 10.5; China 7.5; South-east Asia 23.2; Rest of the World 14.6; India and Indonesia 8.2; Latin America 9.3.
  - Goods Imports and subcomponents reported similarly (see table values).
- Construction of geo-political blocs:
  - Countries ranked by closeness to China and the US using bilateral IPD scores averaged over 2017-2021.
  - Six statistics computed per country: (i) closer to China than the US; (ii) closer to the US than China; (iii) in the closest quartile to China; (iv) in the closest quartile to the US; (v) in the closest quartile to neither; (vi) in the closest quartile to both. Aggregated to regions weighted by PPP GDP.
  - Regions with relative leans are grouped into China-leaning (Southeast Asia, India and Indonesia, rest of the world) and US-leaning (EU+, other AEs, Latin America and the Caribbean); least-aligned regions in each group assigned as non-aligned in baseline fragmentation scenario.
- Calibrating productivity losses:
  - Conditional correlation estimated via panel regression: logLP_i,t = β0 + β1 logLP_i,t−1 + β2 (FDI/GDP)_i,t−1 + δ_t + γ_i + ε_i,t, estimated separately for EMDE and AE recipients using 1980-2021 data.

*Source: IMF staff calculations, Annex Table 4.3.2 and related online annex tables and text.*

### 2. China

### 2. China

### Estimated conditional correlation between FDI/GDP and labor productivity
- Dependent variable: Labor Productivity.
- Reported coefficients (Online Annex Table 4.4.5):
  - Lagged FDI over GDP: -0.00399 (AEs sample), 0.147** (EMDEs sample).
  - Labor productivity (lagged): 0.960*** (AEs), 0.963*** (EMDEs).
  - Constant: 0.464*** (AEs), 0.366*** (EMDEs).
  - Observations: 1262 (AEs), 4313 (EMDEs).
  - Rsquared: 0.997 (AEs), 0.998 (EMDEs).
  - Period: 1980-2021.
  - Sample: AEs and EMDEs.
- Interpretation: The coefficient of 0.147 implies that a 10 percentage point increase in FDI inflows to GDP is associated 1.47 percent increase in labor productivity levels in EMDEs. The corresponding coefficient is small and insignificant for AEs.

### Mapping the FDI–productivity relationship into the macro model
- Equation (1) specifications:
  - log(LP)i,t denotes the logarithm of labor productivity; log(LP)i,t−1 is the lagged value.
  - (FDI/GDP)i,t−1 is the lagged ratio of FDI inflows to GDP.
  - Time and country fixed effects are included.
- Mapping procedure:
  - Define sequence {(FDI/GDP ̃) i,t−j+1 } j=0 ∞ for the fragmentation scenario and {(FDI/GDP ^) i,t−j+1 } j=0 ∞ for the no-fragmentation scenario.
  - Use equation (2) to obtain the difference in labor productivity s periods after the shock:
    - [log(LP ̃) i,t+s − log(LP ^) i,t+s] = β2 ∑ β1^s_j=0 [ (FDI/GDP ̃) i,t+j−1 − (FDI/GDP ^) i,t+j−1 ].
  - As the model lacks a direct FDI variable, the import of investment inputs is used as a proxy on the right-hand side of equation (2).
- Implementation steps:
  - Run the model to obtain sequences of each EMDE region’s import of investment inputs from AE regions under no-fragmentation and fragmentation scenarios.
  - Compute labor productivity changes for EMDE regions using equation (2) with s = 10 (losses cumulated for ten years).
  - Feed estimated labor productivity losses back into the model to obtain overall impact of fragmentation.

### Key modeled outcomes for China and China-bloc EMDEs
- Losses mainly arise for EMDE regions in the China bloc; non-aligned or US-bloc EMDE regions may see some increase in flows due to diversion.
- The ten-year cumulation (s = 10) is used to capture partial substitution of knowledge transfer as China becomes a technological leader in many areas.
- Under the benchmark elasticity of substitution between investment inputs sourced from different regions (1.5):
  - Any gains from diversion of investment flows to non-aligned regions are dominated by the negative impact of reduced external demand, so non-aligned regions experience small output losses in the US–China barriers scenario.

### Fragmentation scenario: Barriers between the U.S. and China only
- Scenario design:
  - Explicit barriers reduce investment input flows by 50 percent between China and the United States only, while all other regions remain nonaligned.
- Outcome summary (from Online Annex Figure 4.4.2):
  - Long-term GDP losses are reported as percent deviation from no-fragmentation scenario for regions including United States, EU+Other AEs, China, SE Asia, ROW, India and Indonesia, LAC, and World.
  - Non-aligned regions are better able to mitigate losses if they remain non-aligned with certainty.
  - Smaller degrees of uncertainty (e.g., 25 percent implied barriers with either bloc) are associated with smaller losses.
  - Facing smaller barriers with respect to the US bloc leads to smaller losses in general, given the relative importance of that bloc as a source of investment flows.

### Alternative assumptions for policy uncertainty
- Regimes for regions in a fragmented world: US bloc, China bloc, or non-aligned.
- Two investor expectation cases:
  - Certainty case: investors expect non-aligned regions will remain non-aligned permanently.
  - Uncertainty case: investors assign positive probability that non-aligned regions will join one bloc or the other.
- Modeling uncertainty:
  - Policy uncertainty is modeled as implicit partial barriers that capture the implied impact on investment flows from uncertainty.
  - Illustrative case in main text: uncertainty translates into implicit barriers equal to half those faced by regions in the two blocs (50 percent of explicit barriers).
  - Online Annex Figure 4.4.3 presents alternate combinations (including 25 percent implied barriers) showing losses vary with the degree and direction of uncertainty.
- General insight: Policy uncertainty can persist indefinitely and effectively raises the bar for market entry or investment, affecting economic decisions similarly to formal barriers.

### Balance sheet exposure to fragmentation risk: data and methodology
- Data sources:
  - Bilateral portfolio equity and debt investments from IMF Coordinated Portfolio Investments Statistics Survey (CPIS).
  - Bilateral cross-border loans to non-banks from BIS International Locational Banking Statistics (residency basis).
- Construction notes and imputations:
  - For India and Indonesia, BIS cross-border bank loans to non-banks were imputed after 2015 and 2016 respectively under stated assumptions.
  - For Germany and Japan, cross-border bank loans to non-banks estimated using BIS bilateral distribution of total cross-border bank investments.
  - For China, estimates of cross-border bank loans were taken from Horn and others (2021).
  - For Argentina, Russia and Saudi Arabia, BIS data on cross-border bank loans were not available; exposures for these countries capture only CPIS portfolio investment data (interpreted as a lower bound).
- Reallocation of CPIS positions:
  - CPIS positions are reallocated to parent country using matrices based on fund holdings from Coppola and others (2021), transformed into nationality-based bilateral positions.
  - Reallocation matrices available from 2007-2021; 2007 matrix used for 2001–2007.
- Political proximity index:
  - Political proximity data from Bailey and others (2017).
  - Ideal Point Distance (IPD) averaged between 2002 and 2021 and normalized into γ in [0,1], where 0 (1) is most (least) politically distant.
  - IPDs rebased so that average IPD across destinations when destination is US equals average IPD across all destinations excluding the US.
- Definition of (gross) balance sheet exposure to fragmentation risk:
  - fragmentation_exposure_i,t = ∑ ( p_i,j,t − γ_i,j p_i,j,t )_j / X_i,t
    - where p_i,j,t is bilateral non-FDI cross-border position (assets plus liabilities) for country i with countryp j at time t;
    - γ_i,j p_i,j,t is the politically-weighted version of that position;
    - X_i,t is a country-specific normalization variable (used to express exposure as a share). In the main text, X_i,t is either nominal GDP in US dollars (GDP_i,t) or total cross-border positions (∑ p_i,j,t_j).
  - GDP data is taken from the WEO database.

### Sample coverage and aggregate results
- Final sample: 38 countries (23 AE and 15 EM) accounting for 86 percent of world GDP.
- Cross-border positions constructed represent 70.8% of total external assets and 59.3% of total external liabilities among countries in the sample (relative to Lane and Milesi-Ferretti (2018) International Investment Position data).

### Robustness and additional results
- Robustness exercises:
  - Alternative normalizations of IPD and alternative political proximity measures were tested.
  - Two alternative IPD transformations: (i) discrete measure (weights 0, 0.5, 1 by quartiles); (ii) continuous “rank” measure normalized to [0,1].
  - Baseline political proximity measure delivers more conservative results (less exposure).
  - Using Häge (2011) political proximity measure yields qualitatively similar results but increases estimated exposures relative to baseline by a factor of 1.5 for AE and roughly 2 for EM in 2021.
  - Conclusion: exposures presented should be interpreted as conservative estimates.
- Additional findings:
  - Net measure (assets minus liabilities) of exposures by country group:
    - Over last 20 years, AEs accumulated a large and positive net exposure to fragmentation risks (6 percent of GDP in 2021).
    - EM have become increasingly liable to politically distant creditors (-8 percent of GDP).
  - Concentration risks:
    - The 5 percent most politically distant countries in EM account for 20 percent of their gross mismatch (against 1 percent for AE).
  - As of 2021:
    - Assets exposures represented 9 percent of the financial system’s total assets in G20 countries (on average).
    - Liabilities exposures accounted for 8% of the total credit going to the non-financial sector.

*Source: IMF World Economic Outlook. Note: IMF staff calculations.*

---


_Source: https://www.imf.org/-/media/files/publications/weo/2023/april/english/ch4annex.pdf_