## wpiea2024196-print-pdf

## Source details

**Canonical URL:** [wpiea2024196-print-pdf](https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024196-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2024/english/wpiea2024196-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2024/english/wpiea2024196-print-pdf.pdf.json)

---

### Introduction — Background and research question
- Geopolitical tensions have increased globally amid deteriorating relations between the United States and China, Russia’s invasion of Ukraine, and the most recent war in the Middle East, raising the specter of global economic and financial fragmentation.
- Main question: Do geopolitical tensions between countries influence the cross-border asset allocation of investment funds?
- Contribution: First paper to identify the role of bilateral geopolitical distance (dissimilarity in UN General Assembly voting behavior) on cross-border portfolio investments, extending gravity models of international finance to include bilateral geopolitical distance.

### Data and measures
- Geopolitical distance:
  - Baseline: (negative of the) S score of Signorino and Ritter (1999).
  - Robustness measures: π measure (Häge, 2011) and ideal point distance (IPD) (Bailey and others, 2017).
  - Correlations among measures: range from 0.66 (π vs. IPD) to 0.84 (S vs. IPD).
  - Distance measure ranges from -1 (agreement) to 1 (disagreement).
- Portfolio allocations: bilateral (country-level) “equity” and “bond” portfolio allocations from the Emerging Portfolio Fund Research (EPFR) database.
- Gravity controls: CEPII Gravity Database (Conte, Cotterlaz, and Mayer 2022).
- Bilateral trade: IMF Direction of Trade Statistics; nominal GDPs from IMF World Economic Outlook.
- Institutional quality: ICRG average score for bureaucracy quality, corruption, democratic accountability, government stability, and law and order.
- Geopolitical distance measures updated up to 2022 using the United Nations voting database and Häge (2011) methodology.

### Empirical model and identification
- Baseline non-linear gravity specification:
  - X_{c,c′,t} = exp(β ⋅ Geopolitical Distance_{c,c′,t−1} + δ ⋅ Gravity Controls_{c,c′} + ν_{c′,t} + ν_{c,t}) ε_{c,c′,t}
  - X_{c,c′,t} is the portfolio share of recipient country c in total cross-border allocation from source country c′ at time t.
  - Geopolitical distance is lagged: Geopolitical Distance_{c,c′,t−1}.
  - Gravity controls include:
    - Distance_{c,c′} (log of geographical distance in kilometers between most populated cities)
    - Common language_{c,c′} (dummy = 1 if a common language is spoken by at least nine percent of the population)
    - Common colonial history_{c,c′} (dummy = 1 if countries share a common colonizer post 1945)
    - Common religion_{c,c′} (index bounded between 0 and 1)
    - Contiguity_{c,c′} (dummy = 1 if countries share a common border)
  - ν_{c′,t} and ν_{c,t} denote source and recipient country time fixed effects.
- Additional time-varying bilateral controls (in some regressions):
  - Source country’s geopolitical distance to others_{c,c′,t−1}: exposure-weighted average geopolitical distance between source country c′ and its financial partner countries excluding recipient c.
  - Bilateral trade_{c,c′,t−1}: total bilateral volume of trade divided by the geometric average of nominal GDPs.
- Identification strategy:
  - Instruments: democracy scores of countries in a dyad as instruments for geopolitical distance.
  - Estimation methods: two-stage instrumental variable method where Poisson Pseudo Maximum Likelihood (PPML) is applied in the second stage; control function method (Wooldridge, 2014).

### Key empirical findings (summary)
- Main result: Investment funds allocate smaller shares of their equity and bond portfolios to recipient countries with greater geopolitical distances to their country of origin.
- Magnitude (baseline and IV-corrected):
  - Baseline: A one standard deviation increase in geopolitical distance is associated with a reduction in both equity and bond portfolio allocations by about 25 percent.
  - Instrumental variable (IV) estimates (endogeneity-corrected):
    - A one standard deviation increase in (instrumented) geopolitical distance reduces cross-border equity allocation by 42 percent (vs. 25 percent in baseline).
    - A one standard deviation increase in (instrumented) geopolitical distance reduces cross-border bond allocation by 60 percent (vs. 25 percent in baseline).
  - Example: the one standard deviation increase is equivalent, for example, to the observed increase in distance between the US and China from 2016 to 2019.
- Interpretation: Funds reallocate away from geopolitically distant countries to mitigate risks such as expropriation, asset freezing, financial restrictions, and reduced access to local information; geopolitical distance serves as a proxy capturing these risks.

### Baseline regression results and gravity controls (section 3)
- Dependent variables: cross-border equity allocation (columns 1–3) and cross-border bond allocation (columns 4–6).
- Estimation: Poisson Pseudo Maximum Likelihood; standard errors clustered at source-recipient country level.
- Specification differences:
  - Columns 1 and 4: bilateral geopolitical distance only.
  - Columns 2 and 5: add source and recipient country time effects.
  - Columns 3 and 6: add gravity controls.
- Key patterns:
  - Inclusion of gravity-type controls reduces estimated effects of geopolitical distance on equity and bond allocations by almost ½.
  - Gravity controls largely have expected signs: lower geographic distance, common language, and common colonial origin imply higher cross-border portfolio allocation.
  - Common religion affects bond (but not equity) allocations; contiguity is not robustly positive once geographic distance is included.
- Baseline economic magnitude example:
  - If source-country funds allocated 5 percent of foreign investments to a recipient, a one standard deviation increase in geopolitical distance reduces allocation to 3.8 percent.

### Robustness, extended specifications, and heterogeneity (section 3)
- Source country’s geopolitical distance to others:
  - Positive and significant coefficient indicates cross-border investment diversion: recipient countries attract additional investments when their source countries become more geopolitically distant from other financial partners.
  - A one standard deviation increase in Source country’s geopolitical distance to others doubles the investment allocation to the recipient country (economic effect reported).
  - Inclusion can render Geopolitical distance insignificant in some specifications.
- Interaction with institutional quality:
  - Funds reduce allocations more strongly in recipient countries with lower institutional quality when geopolitical distance rises.
- Bilateral trade:
  - Inclusion increases (absolute) magnitude of geopolitical distance coefficient while preserving statistical significance.
  - Bilateral trade coefficient is positive.
- Alternative geopolitical distance measures:
  - Effect sizes of a one standard deviation increase in geopolitical distance: equity investments range between 25–29 percent; bond investments range between 26–36 percent.
- Exclusions:
  - Main result holds when excluding the United States or international financial centers (IFCs).

### Endogeneity, instruments, and IV results (section 3)
- Potential endogeneity concern: investments could influence geopolitical distance; recipient and source country incentives may bias estimates.
- Instruments considered:
  - Composite Indicator of National Capabilities (CINC) measures: Log capabilities of the larger economy in the dyad, Larger economy’s share of total capabilities.
  - Democracy measures: Democracy score (lower within the dyad) and Democracy score (higher within the dyad) from Boix, Miller, and Rosato (2012).
  - Alternative instrument (robustness): bilateral arms trade.
- First-stage and second-stage highlights:
  - First-stage: lower democracy score within dyad remains significant with expected sign in preferred specifications.
  - Second-stage IV (Poisson ML with manual bootstrap, clustered by source-recipient, 500 replications):
    - IV increases estimated sensitivities: equity from ≈25 percent to 42 percent per one standard deviation increase; bond from ≈25 percent to 60 percent per one standard deviation increase.
    - Under IV, bond investments are substantially more sensitive than equity investments.
  - Including Source country’s geopolitical distance to others reduces instrumented Geopolitical distance coefficients by 55 percent for equity allocation and 18 percent for bond allocation; effects remain negative and significant for bond allocation.
- Robustness: IV results remain robust to interaction with institutional quality, bilateral trade, alternative distance measures, and exclusion of the United States or IFCs.

### Distributional sensitivity (illustrative percentiles)
- Geopolitical distance distribution: bimodal and highly skewed.
- Sensitivity examples (Table 4, columns 2):
  - Increase from 25th percentile to median (0.4 standard deviation increase):
    - Equity allocation declines by 7 percent.
    - Bond allocation declines by 18 percent.
  - Increase from median to 75th percentile (1.7 standard deviation increase):
    - Equity allocation declines by 33 percent.
    - Bond allocation declines by 83 percent.

### Control function estimation and method comparison
- Control function estimation (Table 5) includes first-stage residuals in second stage.
- Wooldridge (2010) note: IV yields more consistent but less efficient estimates than control function in non-linear models; both approaches used.
- Main message: primary results are robust to estimation method (IV and control function).

### Conclusion — policy relevance and suggested applications (section 4)
- Main empirical conclusions:
  - Bilateral geopolitical distance significantly influences cross-border portfolio equity and bond allocation beyond multilateral push/pull factors and gravity variables.
  - Countries with lower institutional quality are more vulnerable to geopolitical shifts.
  - Geopolitical tensions can generate cross-border investment diversion and sudden reversals of capital flows, exacerbating financial stability risks.
- Policy relevance and potential uses:
  - Framework could be extended to design stress test scenarios that capture effects of geopolitical shocks.
  - Framework could be used for assessment of adequacy of policy and regulatory buffers (e.g., international reserves and bank capital requirements).
  - Framework could improve design of safety nets.

### Appendix — selected data, samples, and coefficients
- Country samples:
  - Equity source countries AE (32) and EMDE (19); Equity recipient countries AE (31) and EMDE (75).
  - Bond source countries AE (29) and EMDE (14); Bond recipient countries AE (31) and EMDE (86).
- International financial centers included: Ireland, Luxembourg, Netherlands, and Singapore.
- Selected coefficient estimates from Table A1 (second stage reported):
  - Geopolitical distance_{c,c',t}: -2.204**; -2.415**; -2.907**; -2.895** (standard errors reported in parentheses).
  - Distance_{c,c'}: -0.231***; -0.217**; -0.227**; -0.240**.
  - Common language_{c,c'}: 0.209***; 0.228***; 0.256**; 0.263**.
  - Common colonial history_{c,c'}: 1.072**; 1.198**; 0.340; 0.381.
  - Common religion_{c,c'}: 0.098; 0.112; 0.333*; 0.338**.
  - Contiguity_{c,c'}: -0.229*; -0.211; -0.174; -0.178.
  - Observations: 28,295; 28,295; 25,339; 25,339 (columns (1) to (4) respectively).
- Instruments and additional controls listed: Log capabilities of the larger economy; Larger economy's share of total capabilities; Capital inflow controls; Expected real GDP growth; Democracy score (lower and higher within the dyad); Bilateral arms trade; CEPII Gravity database; Correlates of War; EPFR Global; UN COMTRADE; IMF WEO; The International Country Risk Guide Database.

*Source: wpiea2024196-print-pdf*

### Introduction

### Introduction

### Background
- Geopolitical tensions have increased globally amid deteriorating relations between the United States and China, Russia’s invasion of Ukraine, and the most recent war in the Middle East, raising the specter of global economic and financial fragmentation.
- Recent theoretical studies focus on financial effects of sanctions, freezing of reserves, and expropriation of foreign assets (Bianchi and Sosa Padilla, 2023 and 2024; Lorenzoni and Werning, 2023) and on geoeconomics (Clayton, Maggiori, and Schreger, 2024).
- Empirical evidence on the impact of geopolitical conflict on cross-border portfolio allocation is scant.

### Research question and contribution
- Main question: Do geopolitical tensions between countries influence the cross-border asset allocation of investment funds?
- Contribution: First paper to identify the role of bilateral geopolitical distance (dissimilarity in UN General Assembly voting behavior) on cross-border portfolio investments, extending gravity models of international finance to include bilateral geopolitical distance.

### Data and measures
- Geopolitical distance:
  - Baseline: (negative of the) S score of Signorino and Ritter (1999).
  - Robustness measures: π measure (Häge, 2011) and ideal point distance (IPD) (Bailey and others, 2017).
  - Correlations among measures: range from 0.66 (π vs. IPD) to 0.84 (S vs. IPD).
  - Distance measure ranges from -1 (agreement) to 1 (disagreement).
- Portfolio allocations: bilateral (country-level) “equity” and “bond” portfolio allocations from the Emerging Portfolio Fund Research (EPFR) database.
- Gravity controls: CEPII Gravity Database (Conte, Cotterlaz, and Mayer 2022).
- Bilateral trade: IMF Direction of Trade Statistics; nominal GDPs from IMF World Economic Outlook.
- Institutional quality: ICRG average score for bureaucracy quality, corruption, democratic accountability, government stability, and law and order.
- Geopolitical distance measures updated up to 2022 using the United Nations voting database and Häge (2011) methodology.
- Table 1 contains descriptive statistics; Appendix 1 provides further data source and transformation details.

### Empirical model and identification
- Baseline empirical specification (non-linear gravity form):
  - X_{c,c′,t} = exp(β ⋅ Geopolitical Distance_{c,c′,t−1} + δ ⋅ Gravity Controls_{c,c′} + ν_{c′,t} + ν_{c,t}) ε_{c,c′,t}
  - X_{c,c′,t} is the portfolio share of recipient country c in total cross-border allocation from source country c′ at time t.
  - Geopolitical distance is lagged: Geopolitical Distance_{c,c′,t−1}.
  - Gravity controls include:
    - Distance_{c,c′} (log of geographical distance in kilometers between most populated cities)
    - Common language_{c,c′} (dummy = 1 if a common language is spoken by at least nine percent of the population)
    - Common colonial history_{c,c′} (dummy = 1 if countries share a common colonizer post 1945)
    - Common religion_{c,c′} (index bounded between 0 and 1)
    - Contiguity_{c,c′} (dummy = 1 if countries share a common border)
  - ν_{c′,t} and ν_{c,t} denote source and recipient country time fixed effects.
  - ε_{c,c′,t} is log normal with mean 1 and variance σ_{c,c′,t}^2; ln ε assumed independent across country-pairs at a given t.
- Additional time-varying bilateral controls (in some regressions):
  - Source country’s geopolitical distance to others_{c,c′,t−1}: exposure-weighted average geopolitical distance between source country c′ and its financial partner countries excluding recipient c.
  - Bilateral trade_{c,c′,t−1}: total bilateral volume of trade divided by the geometric average of nominal GDPs.
- Heterogeneity analysis: interaction of Geopolitical Distance_{c,c′,t−1} with Institutional quality_{c,t−1}.
- Identification strategy to address endogeneity:
  - Instruments: democracy scores of countries in a dyad as instruments for geopolitical distance.
  - Estimation methods: two-stage instrumental variable method where Poisson Pseudo Maximum Likelihood (PPML) (Santos Silva and Tenreyro, 2006) is applied in the second stage; control function method (Wooldridge, 2014).

### Key findings
- Main result: Investment funds allocate smaller shares of their equity and bond portfolios to recipient countries with greater geopolitical distances to their country of origin.
- Magnitude:
  - An increase of one standard deviation in geopolitical distance between a source and a recipient country is associated with a reduction in bilateral cross-border portfolio allocation of investment funds by about 40 percent for equity investments, and 60 percent for bond investments.
  - Example: the one standard deviation increase is equivalent, for example, to the observed increase in distance between the US and China from 2016 to 2019.
- Interpretation: Funds reallocate away from geopolitically distant countries to mitigate risks such as expropriation, asset freezing, financial restrictions, and reduced access to local information; geopolitical distance serves as a proxy capturing these risks.

### Robustness and heterogeneity
- Robustness:
  - Results hold when using alternative geopolitical distance measures (Häge, 2011; Bailey and others, 2017).
  - Results hold when excluding off-shore financial centers and when excluding the United States (the main investor country).
- Recipient-country heterogeneity:
  - Effects of geopolitical distance on portfolio allocation are weaker for recipient countries with stronger institutions.
  - Poor institutional quality can amplify the effects of geopolitical shifts.
- Investment diversion effect:
  - Evidence that a recipient country attracts additional investments when the geopolitical distance between its source countries and third-party countries (their financial partner countries, excluding the recipient country) increases.
  - Implication: Some countries could potentially benefit from rising global geopolitical tensions by attracting new portfolio investments; macro-financial implications depend on absorptive capacity and policy frameworks.

### Relation to literature
- Builds on gravity in international finance literature (Portes and Rey, 2005; Okawa and van Wincoop, 2012) and recent contributions (Mercado, 2020).
- Extends literature that documents role of push and pull factors on aggregate capital flows by explaining bilateral capital flows among country pairs.
- Connects to studies on institutional quality and capital flows (Alfaro, Kalemni-Ozkan, and Volosovych, 2008; Acemoglu, Johnson and Robinson, 2005) and to research on capital inflow episodes (Calvo, Leiderman, and Reinhart, 1993 and 1996; Ghosh and others, 2014; Fratzscher, 2012; Forbes and Warnock, 2012; Reinhart and Reinhart, 2009).

*Source: wpiea2024196-print-pdf - Introduction*

### 3. Results

### 3. Results

### Baseline regressions: geographic and gravity controls
- Dependent variables: cross-border equity allocation (columns 1–3) and cross-border bond allocation (columns 4–6).
- Geopolitical distance measure: (negative of) Signorino and Ritter’s S score, lagged one period.
- Estimation: Poisson Pseudo Maximum Likelihood (Santos Silva and Tenreyro, 2006); standard errors clustered at source-recipient country level.
- Specification differences:
  - Columns 1 and 4: bilateral geopolitical distance only.
  - Columns 2 and 5: add source and recipient country time effects (absorb time-varying country-specific push/pull factors, expected returns).
  - Columns 3 and 6: add gravity controls (geographic distance, common language, common colonial origin, etc.).
- Key findings:
  - Investment funds allocate smaller shares of cross-border equity and bond investments to recipient countries that are geopolitically more distant.
  - Inclusion of gravity-type controls reduces estimated effects of geopolitical distance on equity and bond allocations by almost ½.
  - Gravity controls largely have expected signs: lower geographic distance, common language, and common colonial origin imply higher cross-border portfolio allocation.
  - Common religion affects bond (but not equity) allocations; contiguity is not robustly positive once geographic distance is included.
- Economic magnitude (baseline):
  - A one standard deviation increase in geopolitical distance is associated with a reduction in both equity and bond portfolio allocations by about 25 percent.
  - Numerical example: if source-country funds allocated 5 percent of foreign investments to a recipient, a one standard deviation increase in geopolitical distance reduces allocation to 3.8 percent.

### Robustness and extended specifications (Table 3)
- Column 1: reproduces baseline (source and recipient time effects + gravity).
- Column 2: adds Source country’s geopolitical distance to others.
  - Positive and significant coefficient on Source country’s geopolitical distance to others indicates cross-border investment diversion: recipient countries attract additional investments when their source countries become more geopolitically distant from other financial partners.
  - A one standard deviation increase in Source country’s geopolitical distance to others doubles the investment allocation to the recipient country (economic effect reported).
  - Inclusion of Source country’s geopolitical distance to others can render Geopolitical distance insignificant in some specifications (partly due to endogeneity).
- Column 3: interaction of geopolitical distance with (lagged) institutional quality.
  - Funds reduce allocations more strongly in recipient countries with lower institutional quality when geopolitical distance rises.
- Column 4: includes bilateral trade (goods and services), excluding gravity controls.
  - Inclusion of bilateral trade increases (absolute) magnitude of geopolitical distance coefficient while preserving statistical significance.
  - Bilateral trade coefficient is positive.
- Columns 5–6: use alternative geopolitical distance measures (Häge (2011)  and Bailey et al. (2017) “ideal point distance” (IPD)).
  - Effect sizes of a one standard deviation increase in geopolitical distance: equity investments range between 25–29 percent; bond investments range between 26–36 percent.
- Columns 7–8: exclude the United States or international financial centers (IFCs) from source countries.
  - Main result that geopolitical distance matters holds when excluding the United States or IFCs.

### Endogeneity concerns and instrumental variable approach
- Potential endogeneity: investments could influence geopolitical distance (investments may reduce incentives for bilateral disputes); recipient and source country incentives may bias estimates.
- Instrumental variables considered (trade and conflict literature):
  - Measures based on Composite Indicator of National Capabilities (CINC): Log capabilities of the larger economy in the dyad, Larger economy’s share of total capabilities.
  - Democracy measures: Democracy score (lower within the dyad) and Democracy score (higher within the dyad) from Boix, Miller, and Rosato (2012) (binary scores).
  - Other considered but excluded in baseline: Expected real GDP growth (lower within dyad) and Capital inflow controls (higher within dyad).
- First-stage results (Table 4, panel A):
  - Column 1 (instruments used in trade/conflict studies) replicates coefficient signs found when dependent variable is military conflict, but has limitations (lost observations; time fixed effects do not vary across recipient/source).
  - Column 2 excludes capital controls and expected real GDP growth, and controls for source- and recipient-country time fixed effects and gravity controls.
    - CINC-related instruments become insignificant; lower democracy score within dyad remains significant with expected sign.
  - First-stage regressions 3–8 generate instrumented Geopolitical distance variables for second-stage estimation.
- Second-stage IV estimates (Table 4, panels B and C), Poisson ML with manual bootstrap (clustered by source-recipient, 500 replications):
  - Instrumented geopolitical distance has contemporaneous negative and sizable effects on equity and bond allocations after correcting for endogeneity.
  - Quantitative IV results (Table 4, panel B, column 1 and panel C, column 1):
    - A one standard deviation increase in (instrumented) geopolitical distance reduces cross-border equity allocation by 42 percent (vs. 25 percent in baseline).
    - A one standard deviation increase in (instrumented) geopolitical distance reduces cross-border bond allocation by 60 percent (vs. 25 percent in baseline).
    - Under IV, bond investments are substantially more sensitive than equity investments.
  - Including Source country’s geopolitical distance to others (columns 2):
    - Reduces instrumented Geopolitical distance coefficients by 55 percent for equity allocation and 18 percent for bond allocation; effects remain negative and significant for bond allocation.
  - Robustness in IV setting:
    - Results remain robust to inclusion of interaction with institutional quality, bilateral trade, alternative geopolitical distance measures, and exclusion of the United States or IFCs.

### Distributional sensitivity and illustrative percentiles
- Geopolitical distance distribution: bimodal and highly skewed (Appendix 2, Figure A2).
- Sensitivity example using Table 4, columns 2:
  - Increase from 25th percentile to median (0.4 standard deviation increase):
    - Equity allocation declines by 7 percent.
    - Bond allocation declines by 18 percent.
  - Increase from median to 75th percentile (1.7 standard deviation increase):
    - Equity allocation declines by 33 percent.
    - Bond allocation declines by 83 percent.

### Control function estimation and overall comparison
- Table 5: control function estimation includes first-stage residuals (and interactions with institutional quality when applicable) in second stage to account for endogeneity.
- Wooldridge (2010) point: IV yields more consistent but less efficient estimates than control function in non-linear models; both approaches used.
- Main message: all primary results are robust to estimation method (IV and control function).

### Summary conclusions from endogeneity-corrected analysis
- Correcting for endogeneity yields higher estimated sensitivities of cross-border equity and bond investments to geopolitical distance than baseline estimates.
  - IV estimates: equity sensitivity increases from ≈25 percent to 42 percent per one standard deviation increase; bond sensitivity increases from ≈25 percent to 60 percent.
- Bond investments are substantially more sensitive to geopolitical distance than equity investments after correcting for endogeneity.
- Results robust to alternative specifications, instruments (including bilateral arms trade in Appendix robustness checks), frequency (annual data in Table A1), and exclusion of major source countries or IFCs.

_Italic: Source: wpiea2024196-print-pdf - 3. Results_

### 4. Conclusion

### 4. Conclusion

### Main empirical findings
- Bilateral geopolitical distance between countries exerts a significant and economically meaningful influence on cross-border portfolio equity and bond allocation of investment funds—beyond the effects of multilateral push and pull factors and gravity variables documented in previous studies.
- Countries with lower institutional quality are more vulnerable to geopolitical shifts.
- Geopolitical tensions can generate cross-border investment diversion, whereby a recipient country attracts additional investments when the geopolitical distance between its source countries and their financial partner countries increases.
- Geopolitical tensions can lead to sudden reversals of cross-border capital flows and thereby exacerbate financial stability risks.

### Policy relevance and suggested applications
- The framework presented is policy relevant and can be extended to provide more direct and detailed evidence about transmission of bilateral geopolitical tensions to domestic financial systems through capital flows—a promising direction for future research in light of recent analysis (IMF, 2023).
- With appropriate extensions, the framework could be used for:
  - Design of stress test scenarios that capture the effects of geopolitical shocks.
  - Assessment of the adequacy of policy and regulatory buffers (e.g., international reserves and bank capital requirements).
  - Improvement of the design of safety nets (IMF, 2023).

*Source: wpiea2024196-print-pdf - 4. Conclusion*

### Appendix 1. Data Sources and Transformations

### Appendix 1. Data Sources and Transformations

### List of countries included — Equity
- Source countries. AE (32): Australia, Austria, Belgium, Canada, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Iceland, Ireland, Israel, Italy, Japan, Korea, Luxembourg, Malta, Netherlands, New Zealand, Norway, Portugal, Singapore, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, United Kingdom, United States.
- Source countries. EMDE (19): Bahamas, Bahrain, Brazil, Bulgaria, Chile, China, Colombia, India, Indonesia, Malaysia, Mauritius, Mexico, Poland, Romania, Russia, South Africa, Thailand, Turkey, United Arab Emirates.
- Recipient countries. AE (31): Australia, Austria, Belgium, Canada, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Iceland, Ireland, Israel, Italy, Japan, Korea, Latvia, Netherlands, New Zealand, Norway, Portugal, Singapore, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, United Kingdom, United States.
- Recipient countries. EMDE (75): Angola, Argentina, Bahrain, Bangladesh, Belarus, Bolivia, Botswana, Brazil, Bulgaria, Cambodia, Chile, China, Colombia, Costa Rica, Croatia, Côte d'Ivoire, Dominican Republic, Ecuador, Egypt, El Salvador, Georgia, Ghana, Guatemala, Hungary, India, Indonesia, Iran, Iraq, Jordan, Kazakhstan, Kenya, Kuwait, Lebanon, Liberia, Madagascar, Malawi, Malaysia, Mauritius, Mexico, Mongolia, Morocco, Mozambique, Myanmar, Namibia, Nepal, Nigeria, Oman, Pakistan, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Poland, Qatar, Romania, Russia, Rwanda, Saudi Arabia, South Africa, Sri Lanka, Swaziland, Tanzania, Thailand, Tunisia, Turkey, Turkmenistan, Uganda, Ukraine, United Arab Emirates, Uruguay, Venezuela, Vietnam, Zambia, Zimbabwe.

### Variables, descriptions, sources, and frequency — Cross-border portfolio allocation (Equity and Bond)
- Dependent variables:
  - Cross-border portfolio allocation — Equity (percent) X_{c,c',t}
  - Cross-border portfolio allocation — Bond (percent) X_{c,c',t}
- Geopolitical distance measures:
  - Baseline: Signorino and Ritter's (1999) S_{c,c',t}
  - Häge's (2011) π_{c,c',t}
  - Bailey et al.'s (2017) IPD_{c,c',t}
- Gravity controls:
  - Distance (c,c'): (Log of) geographical distance (in kilometers) between the most populated city of each country.
  - Common language (c,c'): Dummy variable that takes the value 1 if the countries share a common language (spoken by at least 9 percent of the population), and 0 otherwise.
  - Common colonial history (c,c'): Dummy variable that takes the value 1 if the countries share a common colonizer after 1945, and 0 otherwise.
  - Common religion (c,c'): Religious proximity index bounded between 0 and 1 that increases when the countries share a common religion practiced by large shares of their populations.
  - Contiguity (c,c'): Dummy variable that takes the value 1 if the countries share a common border, and 0 otherwise.
- Other controls:
  - Bilateral trade (c,c',t): Imports plus exports divided by the square root of the product of countries' nominal GDPs. Source: IMF Direction of Trade Statistics and IMF WEO. Frequency: Annual.
  - Lender's distance to others (c,c',t): Average geopolitical distance of source country c' to other recipient countries (excluding c), weighted by portfolio allocation values. Frequency: Annual.
  - Institutional quality (of recipient country) (c,t): Average of International Country Risk Guide indicators. Source: The International Country Risk Guide Database. Frequency: Monthly.
- Instrumental variables and additional controls (annual unless noted):
  - Log capabilities of the larger economy (c,c',t)
  - Larger economy's share of total capabilities (c,c',t)
  - Capital inflow controls (higher within the dyad) (c,c',t)
  - Expected real GDP growth (lower within the dyad) (c,c',t)
  - Democracy score (lower within the dyad) (c,c',t)
  - Democracy score (higher within the dyad) (c,c',t)
  - Bilateral arms trade: The total volume of bilateral arms trade (imports plus exports) normalized by the geometric mean of the GDP of the source and recipient country. Sources: UN COMTRADE and IMF WEO. Frequency: Annual.
  - CEPII Gravity database (Conte, Cotterlaz and Mayer, 2022). Frequency: NA.
  - Correlates of War and IMF WEO. (See "Correlates of War" database.)
  - Share of recipient country c in the total cross-border portfolio allocation of investment funds domiciled in country c' at time t. Source: EPFR Global. Frequency: Monthly.
  - Foreign policy disagreement based on countries' voting behavior in the UN General Assembly (see main text in Appendix I). Sources: Signorino and Ritter (1999); Häge (2011); Bailey et al. (2017). Frequency: Annual.

### List of countries included — Bond
- Source countries. AE (29): Australia, Austria, Belgium, Canada, Cyprus, Czech Republic, Denmark, Finland, France, Germany, Greece, Ireland, Italy, Japan, Korea, Luxembourg, Malta, Netherlands, New Zealand, Norway, Portugal, Singapore, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, United Kingdom, United States.
- Source countries. EMDE (14): Bahamas, Brazil, Bulgaria, Chile, Hungary, India, Malaysia, Mauritius, Poland, Romania, Russia, South Africa, Thailand, Turkey.
- Recipient countries. AE (31): Australia, Austria, Belgium, Canada, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Iceland, Ireland, Israel, Italy, Japan, Korea, Latvia, Netherlands, New Zealand, Norway, Portugal, Singapore, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, United Kingdom, United States.
- Recipient countries. EMDE (86): Albania, Algeria, Angola, Argentina, Armenia, Azerbaijan, Bahrain, Bangladesh, Belarus, Bolivia, Bosnia Herzegovina, Botswana, Brazil, Bulgaria, Cambodia, Chile, China, Colombia, Costa Rica, Croatia, Côte d'Ivoire, Republic of the Congo, Dominican Republic, Ecuador, Egypt, El Salvador, Eswatini, Ethiopia, Gabon, Georgia, Ghana, Guatemala, Honduras, Hungary, India, Indonesia, Iran, Iraq, Jamaica, Jordan, Kazakhstan, Kenya, Kuwait, Lebanon, Liberia, Malaysia, Mauritius, Mexico, Moldova, Mongolia, Morocco, Mozambique, Namibia, Nicaragua, Nigeria, North Macedonia, Oman, Pakistan, Panama, Papua New Guinea, Paraguay, Peru, Philippines, Poland, Qatar, Romania, Russia, Rwanda, Saudi Arabia, South Africa, Sri Lanka, Suriname, Tajikistan, Tanzania, Thailand, Trinidad and Tobago, Tunisia, Turkey, Uganda, Ukraine, United Arab Emirates, Uruguay, Uzbekistan, Venezuela, Vietnam, Zambia.

### International financial centers note
- International financial centers included: Ireland, Luxembourg, Netherlands, and Singapore.
- Note: Other known international financial centers, such as Bermuda, British Virgin Islands, Cayman Islands, and Hong Kong SAR are not included as source countries in the regressions.

### Figures and notes
- Figure A1. Geopolitical Distance Measures Based on U.N. Voting Behavior:
  - Panels labeled: 1. United States vs. Russia; 2. United Kingdom vs. Russia; 3. United States vs. China; 4. United States vs. United Kingdom.
  - Sources: Häge (2011); Bailey and others (2017); and IMF staff calculations.
  - Note: Higher values indicate greater geopolitical distance. IPD = Ideal Point Distance of Bailey and others (2017).
  - Time series ticks shown for years including 1946, 1951, 1956, 1961, 1966, 1971, 1976, 1981, 1986, 1991, 1996, 2001, 2006, 2011, 2016, 2021 in various panels.
- Figure A2. Distribution of Geopolitical Distance Measures across Countries (Years 2012 and 2022).

### Appendix 2 — Regression methods and notes (selected)
- Table A1. Baseline Regressions with Annual Frequency Data based on Control Function and Instrumental Variable Methods (second stage):
  - Columns (1) to (2) and (3) to (4) show regression results for cross-border equity and bond portfolio allocation of investment funds, respectively.
  - Dependent variable: the share of recipient country c in the total cross-border portfolio allocation of investment funds domiciled in source country c’ at time t.
  - Estimation: Poisson Pseudo Maximum Likelihood.
  - Columns (1) and (3): control function methods; Columns (2) and (4): two-stages least squared methods. Both use democracy score as instrumental variable.
  - The geopolitical distance is the Signorino and Ritter’s (1999) S measure.
  - 휖̂_{c,c',t} represents the residual of the first stage regression in the control function method.
  - “FE” denotes fixed effects and “Yes” indicates inclusion in the specification.
  - Bootstrap standard errors in parentheses; significance at the 1, 5, and 10 percent levels denoted by ***,**, and *, respectively.
  - Selected coefficient estimates (second stage reported):
    - Geopolitical distance_{c,c',t}: -2.204**; -2.415**; -2.907**; -2.895** (standard errors reported in parentheses).
    - Distance_{c,c'}: -0.231***; -0.217**; -0.227**; -0.240**.
    - Common language_{c,c'}: 0.209***; 0.228***; 0.256**; 0.263**.
    - Common colonial history_{c,c'}: 1.072**; 1.198**; 0.340; 0.381.
    - Common religion_{c,c'}: 0.098; 0.112; 0.333*; 0.338**.
    - Contiguity_{c,c'}: -0.229*; -0.211; -0.174; -0.178.
    - Source country x month FE: Yes (all columns).
    - Recipient country x month FE: Yes (all columns).
    - Observations: 28,295; 28,295; 25,339; 25,339 (columns (1) to (4) respectively).
- Table A2. Baseline Regressions with an Alternative Instrument: Bilateral Arms Trade:
  - Instrument: bilateral arms trade (total volume of bilateral arms trade normalized by the geometric mean of the GDP of the source and recipient countries).
  - Columns (1) to (3) and (5) to (7) show baseline regression results for cross-border equity and bond portfolio allocation estimated with Poisson Pseudo Maximum Likelihood.
  - Columns (4) and (8) show the first stage regressions based on the baseline geopolitical distance measure (S measure); results are similar for other geopolitical distance measures.
  - “FE” denotes fixed effects and “Yes” indicates inclusion.
  - Bootstrap standard errors in parentheses; significance at the 15, 10, 5, and 1 percent levels denoted by †, ***,**, and *, respectively.

### References cited in appendices (selection)
- Signorino, Curtis and Jeffery Ritter. 1999. “Tau-b or Not Tau-b: Measuring the Similarity of Foreign Policy Positions.” International Studies Quarterly 43 (1): 115–44.
- Häge, Frank. 2011. “Choice or Circumstance? Adjusting Measures of Foreign Policy Similarity for Chance Agreement.” Political Analysis 19 (3): 287–305.
- Bailey, Michael, Anton Strezhnev, and Erik Voeten. 2017. “Estimating Dynamic State Preferences from United Nations Voting Data.” Journal of Conflict Resolution 61 (2): 430–56.
- EPFR Global; IMF Direction of Trade Statistics; IMF WEO; UN COMTRADE; The International Country Risk Guide Database; CEPII Gravity database; Correlates of War.

*Source: Appendix 1. Data Sources and Transformations, wpiea2024196-print-pdf*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2024/english/wpiea2024196-print-pdf.pdf_