## Currencies of External Balance Sheets — Working Paper No. WP/2023/237 (Content unit: wpiea2023237-print-pdf)

## Source details

**Canonical URL:** [Currencies of External Balance Sheets — Working Paper No. WP/2023/237 (Content unit: wpiea2023237-print-pdf)](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023237-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023237-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023237-print-pdf.pdf.json)

---

### Data: Actual, Synthetic, and Estimated
- Data sources and construction:
  - Actual data: IMF-administered survey to country authorities (survey solicited data from 1990 onward; voluntary with a 85% response rate for recent years; coverage diminishes for earlier periods) and the Coordinated Portfolio Investment Survey (CPIS), Table 2.
  - Synthetic data: BIS International Debt Issuance (IDS) for portfolio debt liabilities; BIS Locational Banking Statistics (LBS) for banking-related other investment (assets and liabilities).
  - Estimated data: estimates of currency weights rely on geographical distribution of holdings as predictor for currency composition for certain balance-sheet items (assumptions described in the Appendix).
  - Hierarchy for dataset construction: Use actual data whenever available → supplement with synthetic data where actual data are missing → fill remaining gaps with estimated currency weights (direct estimation and model-based methods; detailed methods in Appendix A).
  - Survey requested breakdowns by five SDR currencies (US dollar, euro, Japanese yen, pound sterling, renminbi), domestic currency, and “other currencies.”
  - Tables A1–A3 in the Appendix describe coverage of actual data for each country.
  - Data for Russia start in 1993; Czechia reported from 1993 onward.
- Component-specific treatments and assumptions:
  - Portfolio equity and FDI equity assumed denominated in the currency of the host country (exposure in domestic currency).
  - FDI: equity/debt split; equity shares from IMF IFS and currency weights from Lane and Shambaugh (1990–2008) and CDIS (2009–2020) when survey data unavailable; FDI debt uses actual survey data when available, otherwise proxy weights using portfolio debt assets currency weights.
  - Portfolio debt: actual data from IMF survey and CPIS Table 2; extended using geography-based estimation (Lane and Shambaugh (2010a)) and BIS IDS synthetic series.
  - Other investment: actual survey data; backward coverage extended via BIS LBS (banking assets largest component).
  - Reserves: currency composition 1990–2017 from Bénetrix et al. (2019); extended to 2020 using Central Bank/MoF publications, publicly available IMF COFER, and Ito and McCauley (2020) data.

### Foreign Assets
- Composition: portfolio equity, foreign direct investment (equity and debt), portfolio debt, other investment (mainly bank-related), and reserves.
- Data treatment highlights:
  - Portfolio equity assets: IMF survey and CPIS Table 2; gaps filled using CPIS geography-based method with assumption equity denominated in host currency.
  - FDI: equity/debt split with actual survey data where available; FDI equity currency weights estimated using Lane and Shambaugh (2010b) and CDIS (2009–2020).
  - Portfolio debt: IMF survey and CPIS Table 2; extended combining CPIS geography and host-country bond issuance currency.
  - Other investment: survey information supplemented with BIS LBS for backward extension.
  - Reserves: built on Bénetrix et al. (2019) and extended through 2020 using multiple public sources and interpolation where needed.

### Foreign Liabilities
- Composition: portfolio equity, FDI (equity and debt), portfolio debt, other investment.
- Key assumptions:
  - Portfolio equity and FDI equity liabilities treated as denominated in host-country currency (exposure in domestic currency).
  - FDI debt: IMF survey actuals where available; backward extension using proxy weights based on portfolio debt liabilities currency breakdown.
  - Portfolio debt: IMF survey actuals supplemented by BIS IDS synthetic currency breakdown (covers all debt securities issued by non-residents with comprehensive currency breakdown).
  - Other investment liabilities assembled analogously to other investment assets.

### Currencies over Time (IFI and currency dominance)
- IFI measure: sum of total foreign assets and foreign liabilities (in each currency) in percent of GDP; scaled by weighted average of each country’s GDP.
- Aggregate findings and numeric shares:
  - IFI doubled from the early 1990s to 2020, with a decline in some currencies after the global financial crisis.
  - As of 2020, around 50% of total cross-border holdings are denominated in US dollars or euros.
  - Pound sterling ≈ 4%, Japanese yen ≈ 3%, renminbi ≈ 2% — combined about 8%.
  - If only foreign currency holdings are considered, USD and euro shares rise to 77%.
  - USD represents around 28% of gross assets and 23% of gross liabilities; the euro accounts for 27% of gross assets and 21% of gross liabilities.
  - USD cross-border positions became predominant over the euro since 2014.
  - USD cross-border positions were three times larger than those denominated in euros by 2017, increasing further in 2020 after the peak of the COVID-19 crisis.
  - Excluding the US and the Euro Area, cross-border holdings in USD or euros in 2020 represent over 45% of the total.
- Drivers of the shift toward the USD and away from the euro:
  - Euro Area bank deleveraging since the global financial crisis.
  - Flight away from euro positions spurred by Euro Area sovereign debt crisis.
  - USD appreciation and high liquidity of dollar assets during the global financial crisis.
  - Relative lack of supply of safe euro-denominated assets compared to USD.
  - Further USD shift in 2020 may reflect flight to safety during the COVID-19 crisis.

### Evolution of Foreign Currency Exposures (FX_AGG)
- Definition:
  - FX_AGG_i,t = ω^A_i,t s^A_i,t − ω^L_i,t s^L_i,t = Σ_c ω^A,c_i,t s^A_i,t − Σ_c ω^L,c_i,t s^L_i,t.
  - ω^A_i,t: proportion of assets denominated in foreign currency; s^A_i,t: share of assets in external balance sheet (s^A_i,t = A_i,t / (A_i,t + L_i,t)). Analogous definitions for liabilities.
  - Positive FX_AGG implies net long foreign currency; negative implies net short foreign currency.
  - Index captures sensitivity to uniform appreciation/depreciation of home currency relative to all other currencies.
- Distributional changes 1990–2020:
  - 1990: 60% of economies held net negative foreign currency positions.
  - 2020: proportion with net negative positions declined to 14%.
  - Rightward shift toward long foreign currency positions since 1990; bulk of change occurred before the global financial crisis and then leveled off.
- Asset classes driving the shift:
  - Other investment (predominantly bank-related) and portfolio equity are principal contributors to the shift toward long foreign currency positions.
  - Accumulation of foreign exchange reserves also contributed, especially in the 1990s.
- Emerging market economies (EMEs):
  - In 1990, 20 of the 30 economies with short foreign currency positions were EMEs.
  - By 2020, this number reduced to 6 out of 7.
  - Contributing factors: current account surpluses increasing foreign assets relative to liabilities; shift in foreign liabilities from foreign-currency debt to equity-type liabilities; increase in local-currency lending via local affiliates after 1995.
- Remaining vulnerabilities:
  - Net negative positions remain concentrated in EMEs.
  - Decline in currency mismatches attributed to reforms since the 1990s, foreign asset accumulation in the 2000s, and stricter banking regulation.
  - Persistent vulnerabilities in portfolio debt due to foreign currency exposure.
  - Substantial worsening in Portfolio Debt exposures since the Global Financial Crisis (GFC) of 2007/2008 and the Taper Tantrum of 2013.

### Link with Aggregate Trade Invoicing
- Investigation: correlation between dominant currencies in trade invoicing and in financial liabilities/assets (simple macro correlations; no causal claims).
- Key patterns:
  - Economies with higher shares of exports invoiced in USD tend to have higher shares of USD-denominated assets and liabilities (similar for imports).
  - When considering net trade and net balance-sheet exposures:
    - Correlation between trade invoicing in USD and net balance-sheet USD exposure (FX_AGG) drops drastically.
    - Economies with predominant US dollar-based trade do not necessarily show larger net balance sheet exposures in USD at the aggregate level.
- Visual descriptions:
  - Left panel: share of exports denominated in USD vs. share of total liabilities denominated in USD (circle sizes proportional to GDP).
  - Right panel: FX_AGG in USD vs. net trade exposure in USD (FX_XM) — weak association in net metrics.

### Regression Analysis — Empirical strategy and main findings
- Empirical strategy:
  - Parsimonious panel regressions following Lane and Shambaugh (2010b) using full panel dataset with year fixed effects; robust standard errors reported.
  - Determinants: trade openness, inflation volatility, GDP volatility, NEER volatility, cov(GDP, NEER), exchange rate peg, log population, institutions, capital controls, EMU membership, log GDP per capita, reserve-currency issuer dummy.
  - Significance notation: ***, **, * denote p <0.01, p <0.05, p <0.1 respectively.
- Key coefficients for aggregate FX_AGG (selected coefficients across specifications):
  - Openness: 0.135*** (column (1)); 0.138*** (2); 0.108*** (3); 0.098*** (4); 0.111*** (5).
  - Inflation volatility: 0.179 (1); 0.165 (2); -0.089 (3); -0.139 (4); 0.034 (5).
  - GDP volatility: 1.094** (1); 0.873* (2); 0.712 (3); 1.094** (4); 0.539 (5).
  - NEER volatility: -0.237** (1); -0.221* (2); 0.008 (3); 0.035 (4); -0.092 (5).
  - cov(GDP, NEER): -4.226 (1); -7.744** (2); -9.840** (3); -10.238*** (4).
  - Log population: 0.052*** (1); 0.051*** (2); 0.049*** (3); 0.056*** (4); 0.069*** (5).
  - Institutions: 0.800*** (1); 0.841*** (2); 0.419*** (3); 0.526*** (4).
  - Capital controls: -0.163*** (1); -0.181*** (3); -0.094*** (4); -0.089*** (5).
  - Peg: -0.041*** (1); -0.001 (2); -0.024** (3).
  - EMU: -0.076*** (4); -0.129*** (5).
  - Log GDP per capita: 0.145*** (3); 0.112*** (4); 0.112*** (5).
  - Reserve currency: -0.494*** (5).
  - Observations and fit: Observations = 1,437 (cols (1),(2),(4),(5)), 1,442 (3); R-squared = 0.402 (1); 0.440 (2); 0.407 (3); 0.475 (4); 0.498 (5).
- Interpretation of aggregate results:
  - Trade openness is robustly and positively associated with FX_AGG across specifications.
  - NEER volatility generally negative in some specifications.
  - GDP volatility significant and positive in several specifications.
  - cov(GDP, NEER) significantly negative in several specifications.
  - Better institutional environment associated with higher FX_AGG.
  - Exchange rate peg coefficient negative in some specs but may become insignificant after controlling for GDP per capita.
  - FX_AGG strongly correlated with level of development: richer countries have higher foreign-currency exposure.
  - Country size (log population) positively and significantly associated.

### Asset-class regression patterns and selected coefficients
- General patterns:
  - Openness, level of development, and country size positively associated with foreign-currency exposures across most asset classes (FDI, portfolio equity (PEQ), portfolio debt (PD), other investment (OI), reserves).
  - Capital controls enter negatively for all asset classes except reserves.
  - Institutions: positive for FDI, PEQ, OI; negative for PD and reserves.
  - Volatility measures show heterogeneous signs across asset classes.
- Selected highlights (coefficients reported exactly as in source):
  - FDI (Table 2):
    - Openness: 0.038*** (1); 0.039*** (2); 0.030*** (3); 0.024*** (4); 0.029*** (5).
    - Inflation volatility: 0.211*** (1); 0.206*** (2); 0.128*** (3); 0.094*** (4); 0.129*** (5).
    - GDP volatility: -0.617*** (1); -0.689*** (2); -0.607*** (3); -0.607*** (4); -0.617*** (5).
    - NEER volatility: -0.155*** (1); -0.149*** (2); -0.082*** (3); -0.054** (4); -0.082*** (5).
    - Log population: 0.011*** (1); 0.010*** (2); 0.011*** (3); 0.012*** (4); 0.014*** (5).
    - Institutions: 0.186*** (1); 0.201*** (2); 0.045*** (3); 0.057*** (4); 0.185*** (5).
    - Capital controls: -0.031*** (1); -0.037*** (2); -0.006 (3); -0.007 (4); -0.018*** (5).
    - Peg: -0.013*** (1); 0.002 (2); -0.013*** (3); -0.020*** (4); -0.016*** (5).
    - Log GDP per capita: 0.037*** (3); 0.041*** (4); 0.035*** (5).
    - Reserve currency: -0.085*** (5).
    - Observations: 1,437 (1),(2),(4),(5); 1,442 (3); 1,484 for reserve-currency subsample. R-squared range: 0.294 to 0.421.
  - Portfolio Equity (Table 2):
    - Openness: -0.016*** (6); -0.015*** (7); -0.020*** (8); -0.018*** (9); -0.010*** (10).
    - Inflation volatility: -0.016 (6); -0.019 (7); -0.072*** (8); -0.048* (9); 0.023 (10).
    - GDP volatility: -0.405*** (6); -0.507*** (7); -0.496*** (8); -0.486*** (9); -0.586*** (10).
    - NEER volatility: 0.027 (6); 0.032 (7); 0.074*** (8); 0.056*** (9); -0.001 (10).
    - cov(GDP, NEER): 0.632 (6); -0.993* (7); -0.946** (8); -1.231** (9).
    - Log population: -0.001 (6); -0.002** (7); -0.003*** (8); -0.002** (9); -0.002* (10).
    - Institutions: 0.185*** (6); 0.200*** (7); 0.159*** (8); 0.160*** (9).
    - Capital controls: -0.025*** (6); -0.017*** (7); -0.016*** (8); -0.016*** (9).
    - R-squared range for PEQ regressions includes 0.319 to 0.469.
  - Portfolio Debt (Table 3):
    - Openness: 0.069*** (1); 0.069*** (2); 0.066*** (3); 0.066*** (4); 0.076*** (5).
    - Inflation volatility: -0.484*** (1); -0.481*** (2); -0.448*** (3); -0.506*** (4); -0.294*** (5).
    - GDP volatility: 0.762*** (1); 0.770*** (2); 0.561*** (3); 0.788*** (4); 0.169 (5).
    - NEER volatility: 0.377*** (1); 0.375*** (2); 0.365*** (3); 0.395*** (4); 0.230*** (5).
    - cov(GDP, NEER): -8.578*** (1); -8.446*** (2); -7.104*** (3); -8.650*** (4).
    - Institutions: -0.050*** (1); -0.054*** (2); -0.089*** (3); -0.052** (4).
    - Capital controls: -0.070*** (1); -0.068*** (2); -0.061*** (3); -0.054*** (4); -0.143*** (5).
    - Log GDP per capita: 0.012*** (1); 0.009** (2); 0.006 (3); 0.133*** (5).
    - R-squared range: 0.202 to 0.243 for PD columns (1)-(5).
  - Other Investment (Table 3):
    - Openness: 0.026*** (6); 0.034*** (7); 0.000 (8); 0.002 (9); -0.008 (10).
    - Inflation volatility: 0.753*** (6); 0.799*** (7); 0.502*** (8); 0.555*** (9); 0.422*** (10).
    - GDP volatility: 0.713* (6); 0.670* (7); 0.470 (8); 0.848** (9); 0.910*** (10).
    - NEER volatility: -0.716*** (6); -0.748*** (7); -0.489*** (8); -0.542*** (9); -0.425*** (10).
    - Institutions: 0.721*** (6); 0.665*** (7); 0.325*** (8); 0.351*** (9).
    - Capital controls: -0.132*** (6); -0.063*** (7); -0.054** (9).
    - R-squared range: 0.484–0.553 for OI columns (6)-(10).
  - Reserves (Table 3):
    - Openness: 0.023*** (11); 0.015*** (12); 0.034*** (13); 0.028*** (14); 0.026*** (15).
    - Inflation volatility: -0.273*** (11); -0.326*** (12); -0.198*** (13); -0.231*** (14); -0.251*** (15).
    - GDP volatility: 0.656*** (11); 0.637*** (12); 0.802*** (13); 0.568*** (14); 0.693*** (15).
    - NEER volatility: 0.223*** (11); 0.261*** (12); 0.141*** (13); 0.181*** (14); 0.194*** (15).
    - Reserve currency: 0.001 (11); -0.198*** (12); -0.196*** (13).
    - Institutions: -0.216*** (11); -0.146*** (12).
    - Capital controls: 0.096*** (11); 0.079*** (12); 0.052*** (13); 0.046*** (14).
    - Log GDP per capita: -0.056*** (11); -0.035*** (12); -0.033*** (13).
    - R-squared range: 0.431–0.540 for reserves columns.
- Summary of substantive regression findings:
  - Trade openness is consistently strong and positive determinant of foreign-currency exposure.
  - Level of development (log GDP per capita) and institutional quality matter: higher development and better institutions associated with greater foreign-currency exposure for several asset classes (institutions have mixed signs).
  - Country size (log population) positively associates with foreign-currency exposure.
  - Capital controls generally associated with lower foreign-currency exposure, except for reserves where relation can be positive.
  - Volatility measures heterogeneous across asset classes: inflation volatility negative for portfolio debt and reserves but positive for FDI and other investment; GDP volatility positive for portfolio debt, other investment, reserves and negative for FDI and portfolio equity; cov(GDP, NEER) notably negative for portfolio debt.

### Appendix B — Regression by country group, financial exchange rates, and valuation decomposition
- Regression results by country group (EME vs Advanced):
  - Positive association between openness and foreign currency exposures remains significant for both EMEs and advanced economies (Tables B5 and B8).
  - Inflation volatility associated with longer positions for both EMEs and advanced economies.
  - NEER volatility: positive for EMEs, negative in some specifications for advanced economies.
  - Institutions associated with longer positions (predominant in EMEs); capital controls associated with shorter positions (predominant in EMEs).
- Financial exchange rate indices — definitions:
  - Asset-weighted: I_A_{i,t} = I_A_{i,t−1} × (1 + Σ_j ω^A_{i,j,t} × ∆E_{i,j,t}).
  - Liability-weighted: I_L_{i,t} = I_L_{i,t−1} × (1 + Σ_j ω^L_{i,j,t} × ∆E_{i,j,t}).
  - Net financial exchange index: I_F_{i,t} = I_F_{i,t−1} × (1 + ∆I_A_{i,t} × s^A_{i,t} − ∆I_L_{i,t} × s^L_{i,t}).
  - ∆E_{i,j,t} is percentage change in bilateral exchange rate of major currencies (US dollar, euro, Japanese yen, Pound sterling, renminbi, other); increase in E defined as depreciation of domestic currency relative to currency j.
  - “Other” currencies: imputed using average change of major currencies when included.
- Statistical properties and correlations (1990-2020):
  - Pairwise medians:
    - Financial indices for gross assets and liabilities (A, L): close to 1 and positive.
    - Gross debt assets and debt liabilities (A_D, L_D): high correlations.
    - Net debt and net equity (N_D, N_EQ): overall pairwise correlation -0.71; Advanced economies -0.5; Emerging markets -0.76; Creditor economies 0.20; Debtor economies -0.77.
  - Median correlation between trade-weighted index changes and asset- and liability-based financial-weighted indices: around 0.5 over full sample.
  - Median correlation between net financial exchange rate and trade-weighted exchange rate: 0.15.
  - Advanced economies: positive correlation between net financial-weighted and trade-weighted indices; Emerging economies: negative correlation.
  - Debtor economies: no correlation between trade index and financial index.
  - Decomposition: positive correlations for equity positions between financial and trade indices; mostly negative correlations for net debt positions.
- Valuation decomposition methodology:
  - Valuation effect due to currency movements (VAL_XR): VAL_XR_{i,t} = ∆I_A_{i,t} × A_{i,t−1} − ∆I_L_{i,t} × L_{i,t−1}, where A_i and L_i are foreign assets and liabilities relative to GDP. “Other” currencies assumed to move as average of main SDR currencies.
  - Stock-flow reconciliation: ∆NIIP = FA_{i,t} + VAL_XR_{i,t} + VAL_OTH_{i,t}, where FA_{i,t} is financial account/current account balance, VAL_OTH_{i,t} is valuation change due to asset prices and other statistical changes.
- Episodes: GFC (2008) versus COVID-19 (2020) — stylized facts and numeric highlights (selected weighted-average numbers in percent of group (or economy) GDP):
  - COVID-19 (2020), Full sample:
    - FA: -0.1; ∆ NIIP: -1.4; VAL_XR Total: -0.3; Debt & FXR: -0.8; Equity: 0.6; VAL_OTH Total: -1.0; Debt & FXR: 0.5; Equity: -1.5; VAL_TOTAL: -1.3; Debt & FXR: -0.3; Equity: -1.0.
  - COVID-19 (2020), Advanced Economies:
    - FA: -0.5; ∆ NIIP: -2.8; VAL_XR Total: -0.8; Debt & FXR: -1.1; Equity: 0.3; VAL_OTH Total: -1.6; Debt & FXR: 0.5; Equity: -2.1; VAL_TOTAL: -2.3; Debt & FXR: -0.5; Equity: -1.8.
  - COVID-19 (2020), USA:
    - FA: -3.2; ∆ NIIP: -13.3; VAL_XR Total: 3.3; Debt & FXR: -1.4; Equity: 4.7; VAL_OTH Total: -13.3; Debt & FXR: 0.3; Equity: -13.6; VAL_TOTAL: -10.1; Debt & FXR: -1.1; Equity: -9.0.
  - COVID-19 (2020), Emerging Economies:
    - FA: 0.6; ∆ NIIP: 1.1; VAL_XR Total: 0.6; Debt & FXR: -0.4; Equity: 1.0; VAL_OTH Total: 0.2; Debt & FXR: 0.6; Equity: -0.6; VAL_TOTAL: 0.9; Debt & FXR: 0.1; Equity: 0.4.
  - COVID-19 (2020), Creditors Economies:
    - FA: 1.3; ∆ NIIP: 4.7; VAL_XR Total: -0.7; Debt & FXR: -0.6; Equity: -0.2; VAL_OTH Total: 4.2; Debt & FXR: 1.9; Equity: 2.3; VAL_TOTAL: 3.4; Debt & FXR: 1.3; Equity: 2.1.
  - COVID-19 (2020), Debtors Economies:
    - FA: -1.1; ∆ NIIP: -5.5; VAL_XR Total: 0.1; Debt & FXR: -1.0; Equity: 1.1; VAL_OTH Total: -4.7; Debt & FXR: -0.4; Equity: -4.2; VAL_TOTAL: -4.7; Debt & FXR: -1.5; Equity: -3.1.
  - GFC (2008), Full sample:
    - FA: -0.8; ∆ NIIP: 0.1; VAL_XR Total: 0.9; Debt & FXR: -1.9; Equity: 2.7; VAL_OTH Total: 0.6; Debt & FXR: 2.3; Equity: -2.3; VAL_TOTAL: 0.9; Debt & FXR: 0.3; Equity: 0.5.
  - GFC (2008), Advanced Economies:
    - FA: -2.4; ∆ NIIP: -4.3; VAL_XR Total: 1.0; Debt & FXR: -2.2; Equity: 3.1; VAL_OTH Total: -2.3; Debt & FXR: 3.2; Equity: -6.4; VAL_TOTAL: -2.0; Debt & FXR: 1.0; Equity: -3.3.
  - GFC (2008), USA:
    - FA: -5.4; ∆ NIIP: -19.5; VAL_XR Total: -2.1; Debt & FXR: 0.9; Equity: -3.0; VAL_OTH Total: -11.9; Debt & FXR: 1.3; Equity: -13.2; VAL_TOTAL: -14.1; Debt & FXR: 2.2; Equity: -16.3.
  - GFC (2008), Emerging Economies:
    - FA: 4.1; ∆ NIIP: 12.7; VAL_XR Total: 0.6; Debt & FXR: -1.0; Equity: 1.6; VAL_OTH Total: 9.3; Debt & FXR: -0.4; Equity: 8.9; VAL_TOTAL: 9.7; Debt & FXR: -1.7; Equity: 11.4.
- Episode-level interpretations:
  - GFC: originated in banking sector; generalized losses in stock markets; valuation changes due to other movements played a stabilizing role in cross-country NIIP distribution.
  - COVID-19: global health shock with shorter-lived exchange rate movements; large stock market gains in the US; valuation gains in EMEs and valuation losses in advanced economies; valuation changes due to asset prices increased global imbalances (“destabilizing”).
  - Both episodes induced large wealth transfers across countries; valuation gains for creditor economies and losses for debtors; US experienced valuation losses in both events.
  - Valuation changes induced by exchange rates systematically smaller than valuation due to other movements; no evidence of systematic offsetting behavior between VAL_XR and VAL_OTH.
- Dataset and applications:
  - Dataset assembled covers currency composition of external balance sheets and financial exchange rates for 50 economies over 1990–2020.
  - Stylized facts: USD and euro remain dominant; economies improved currency exposures over time; financial-weighted exchange rates weakly correlated with trade-weighted indices and can better measure wealth effects of currency movements.
  - Application illustrated via comparison of wealth transfers during COVID-19 and the GFC.
  - Future use: study unresolved open-economy macro issues and guide design of open economy models.

*Source: wpiea2023237-print-pdf (https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023237-print-pdf.pdf)*

### 2.1  Data: Actual, Synthetic, and Estimated

### 2.1 Data: Actual, Synthetic, and Estimated

### Data sources and construction
- Actual data were collected from:
  - an IMF-administered survey to country authorities (survey solicited data from 1990 onward; voluntary with a 85% response rate for recent years; coverage diminishes for earlier periods).
  - the Coordinated Portfolio Investment Survey (CPIS), Table 2.
- Synthetic data used to address gaps:
  - BIS International Debt Issuance (IDS) as synthetic data for portfolio debt liabilities.
  - BIS Locational Banking Statistics (LBS) as synthetic data for banking-related other investment (assets and liabilities).
- Estimated data:
  - Estimates of currency weights that rely on the geographical distribution of holdings as a predictor of currency composition for certain balance-sheet items (assumptions described in the Appendix).
- Hierarchy for dataset construction:
  - Use actual data whenever available.
  - Supplement with synthetic data where actual data are missing.
  - Fill remaining gaps with estimated currency weights (direct estimation and model-based methods; detailed methods in Appendix A).
- Coverage notes:
  - Survey requested breakdowns by five SDR currencies (US dollar, euro, Japanese yen, pound sterling, renminbi), domestic currency, and “other currencies.”
  - Tables A1–A3 in the Appendix describe coverage of actual data for each country.
  - Data for Russia start in 1993 due to lack of information. Czechia reported from 1993 onward.

### Key methodological assumptions
- Portfolio equity and FDI equity are assumed denominated in the currency of the host country (implying exposure in domestic currency).
- For FDI:
  - Equity/debt components split; equity shares from IMF International Financial Statistics (IFS) and currency weights from Lane and Shambaugh (1990–2008) and CDIS (2009–2020) when survey data unavailable.
  - FDI debt uses actual survey data when available; proxy weights use currency weights of portfolio debt assets when needed.
- Portfolio debt:
  - Actual data from IMF survey and CPIS Table 2.
  - Extended using Lane and Shambaugh (2010a) geography-based estimation and BIS IDS synthetic series.
- Other investment:
  - Actual data from IMF survey; backward coverage extended using BIS LBS (banking assets are largest component).
- Reserves:
  - Currency composition 1990–2017 sourced from Benetrix et al. (2019); extended to 2020 using Central Bank/MoF publications, publicly available IMF COFER, and Ito and McCauley (2020) data.

---

### 2.2 Foreign Assets

- International investment position assets comprise five main items: portfolio equity, foreign direct investment (equity and debt), portfolio debt, other investment (mainly bank-related), and reserves.
- Portfolio equity assets:
  - Actual data from IMF survey and CPIS Table 2.
  - Gaps filled using geography-based method (CPIS geographic locations; assumption equity is denominated in currency of host country; methodology in Appendix A).
- FDI:
  - Equity/debt split; actual survey data used when available.
  - Currency weights for FDI equity estimated using Lane and Shambaugh (2010b) and CDIS (2009–2020); assumed denominated in host currency.
  - FDI debt proxied by portfolio debt currency weights when actual data unavailable.
- Portfolio debt:
  - Actual data from IMF survey and CPIS Table 2.
  - Extended with Lane and Shambaugh (2010a) method combining CPIS geography and host-country bond issuance currency.
- Other investment:
  - Actual survey data; backward extension via BIS LBS to capture banking-dominated composition.
- Reserves:
  - 1990–2017 from Benetrix et al. (2019); extended through 2020 using multiple non-confidential public sources.

---

### 2.3 Foreign Liabilities

- Liabilities comprise portfolio equity, FDI (equity and debt), portfolio debt, and other investment.
- Assumptions and data treatment:
  - Portfolio equity and FDI equity liabilities treated as denominated in host-country currency (exposure in domestic currency).
  - FDI debt: use IMF survey actuals when available; extend backwards using proxy weights based on portfolio debt liabilities currency breakdown.
  - Portfolio debt: IMF survey actuals supplemented by BIS IDS synthetic currency breakdown (covers all debt securities issued by non-residents with comprehensive currency breakdown).
  - Other investment liabilities assembled analogously to other investment assets.

---

### 3 Currencies over Time (IFI and currency dominance)

- IFI measure defined as sum of total foreign assets and foreign liabilities (in each currency) in percent of GDP; scaled by the weighted average of each country’s GDP.
- Aggregate findings:
  - IFI doubled from the early 1990s to 2020, with a decline in some currencies after the global financial crisis.
  - As of 2020, around 50% of total cross-border holdings are denominated in US dollars or euros.
  - Pound sterling ≈ 4%, Japanese yen ≈ 3%, renminbi ≈ 2% — combined about 8%.
  - If only foreign currency holdings are considered, USD and euro shares rise to 77%.
  - USD represents around 28% of gross assets and 23% of gross liabilities; the euro accounts for 27% of gross assets and 21% of gross liabilities.
  - USD cross-border positions became predominant over the euro since 2014.
  - USD cross-border positions were three times larger than those denominated in euros by 2017, increasing further in 2020 after the peak of the COVID-19 crisis.
  - Excluding the US and the Euro Area, cross-border holdings in USD or euros in 2020 represent over 45% of the total.
- Drivers of the shift toward the USD and away from the euro:
  - Euro Area bank deleveraging from cross-border positions since the global financial crisis (cited studies: McCauley et al. (2019), Emter et al. (2019)).
  - Flight away from euro positions spurred by Euro Area sovereign debt crisis (Maggiori et al. (2019)).
  - USD appreciation and high liquidity of dollar assets during the global financial crisis.
  - Relative lack of supply of safe euro-denominated assets compared to USD (Ilzetzki et al. (2019)).
  - Further USD shift in 2020 may reflect flight to safety during the COVID-19 crisis (Hale and Juvenal, 2023).

---

### 3.1 Evolution of Foreign Currency Exposures (FX_AGG)

- FX_AGG definition:
  - FX_AGG_i,t = ω^A_i,t s^A_i,t − ω^L_i,t s^L_i,t = Σ_c ω^A,c_i,t s^A_i,t − Σ_c ω^L,c_i,t s^L_i,t
  - ω^A_i,t: proportion of assets denominated in foreign currency; s^A_i,t: share of assets in external balance sheet (s^A_i,t = A_i,t / (A_i,t + L_i,t)).
  - ω^L_i,t and s^L_i,t analogous for liabilities.
  - Positive FX_AGG implies net long foreign currency; negative implies net short foreign currency.
  - Index captures sensitivity to uniform appreciation/depreciation of home currency relative to all other currencies.
- Distributional changes 1990–2020:
  - 1990: 60% of economies held net negative foreign currency positions.
  - 2020: proportion with net negative positions declined to 14%.
  - Rightward shift toward long foreign currency positions since 1990; bulk of change occurred before the global financial crisis and then leveled off.
- Asset classes driving the shift:
  - Other investment (predominantly bank-related) and portfolio equity are the principal contributors to the shift toward long foreign currency positions.
  - Accumulation of foreign exchange reserves also contributed, especially in the 1990s.
- Emerging market economies (EMEs) specifics:
  - In 1990, 20 of the 30 economies with short foreign currency positions were EMEs.
  - By 2020, this number reduced to 6 out of 7.
  - Contributing factors for EMEs’ improvement:
    - Current account surpluses increasing foreign assets relative to liabilities.
    - Shift in foreign liabilities from foreign-currency debt to equity-type liabilities.
    - Change in international bank lending: post-1995 increase in local-currency lending via local affiliates, often after acquiring EME banks with local currency deposits (reduces currency mismatch risk; Chui et al., 2018).
- Remaining vulnerabilities:
  - Net negative positions remain concentrated in EMEs (“original sin” literature; Eichengreen et al., 2003).
  - Decline in currency mismatches attributed to reforms since the 1990s, foreign asset accumulation in the 2000s, and stricter banking regulation (Chui et al., 2016).
  - Persistent vulnerabilities in portfolio debt due to foreign currency exposure.
  - Noted substantial worsening in Portfolio Debt exposures since the Global Financial Crisis (GFC) of 2007/2008 and the Taper Tantrum of 2013.

---

### 3.2 Link with Aggregate Trade Invoicing

- Investigation focus:
  - Correlation between dominant currencies in trade invoicing and in financial liabilities/assets.
  - Simple macro correlations, not sector- or firm-level mismatch analysis; no causal claims.
- Key patterns:
  - Economies with higher shares of exports invoiced in USD tend to have higher shares of USD-denominated assets and liabilities (similar pattern for imports).
  - However, when considering net trade and net balance-sheet exposures:
    - Correlation between trade invoicing in USD and net balance-sheet USD exposure (FX_AGG) drops drastically.
    - Economies with predominant US dollar-based trade do not necessarily show larger net balance sheet exposures in USD at the aggregate level.
- Visual correlations (as described):
  - Left panel: share of exports denominated in USD vs. share of total liabilities denominated in USD (circle sizes proportional to GDP).
  - Right panel: FX_AGG in USD vs. net trade exposure in USD (FX_XM) — weak association in net metrics.

*Source: wpiea2023237-print-pdf - 2.1 Data: Actual, Synthetic, and Estimated*

### 3.3  Regression Analysis

### 3.3  Regression Analysis

### Empirical strategy and explanatory variables
- Approach follows Lane and Shambaugh (2010b): parsimonious regressions using the full panel dataset rather than four-year interval observations.
- Determinants analyzed: trade and financial openness, inflation volatility, GDP volatility, NEER (nominal effective exchange rate) volatility, cov(GDP, NEER), exchange rate regime (peg), country size (log population), institutions, capital controls, EMU membership, log GDP per capita, reserve-currency issuer dummy.
- Regressions include year fixed effects. Robust standard errors reported. ***, **, and * denote p <0.01, p <0.05, and p <0.1, respectively.
- Appendix A.4 (not reproduced here) details sources of macro variables.

### Determinants of aggregate foreign currency exposure (FX_AGG) — key coefficients (Table 1)
- Openness: 0.135*** (column (1)); 0.138*** (2); 0.108*** (3); 0.098*** (4); 0.111*** (5).
- Inflation volatility: 0.179 (1); 0.165 (2); -0.089 (3); -0.139 (4); 0.034 (5).
- GDP volatility: 1.094** (1); 0.873* (2); 0.712 (3); 1.094** (4); 0.539 (5).
- NEER volatility: -0.237** (1); -0.221* (2); 0.008 (3); 0.035 (4); -0.092 (5).
- cov(GDP, NEER): -4.226 (1); -7.744** (2); -9.840** (3); -10.238*** (4).
- Log population: 0.052*** (1); 0.051*** (2); 0.049*** (3); 0.056*** (4); 0.069*** (5).
- Institutions: 0.800*** (1); 0.841*** (2); 0.419*** (3); 0.526*** (4).
- Capital controls: -0.163*** (1); -0.181*** (3); -0.094*** (4); -0.089*** (5).
- Peg: -0.041*** (1); -0.001 (2); -0.024** (3).
- EMU: -0.076*** (4); -0.129*** (5).
- Log GDP per capita: 0.145*** (3); 0.112*** (4); 0.112*** (5).
- Reserve currency: -0.494*** (5).
- Observations and fit: Observations = 1,437 (cols (1),(2),(4),(5)), 1,442 (3); R-squared = 0.402 (1); 0.440 (2); 0.407 (3); 0.475 (4); 0.498 (5).

### Interpretation of aggregate results
- Trade openness is robustly and positively associated with FX_AGG across specifications.
- NEER volatility generally has a negative sign in columns (1) and (2).
- GDP volatility is significant and positive in columns (1), (2), and (4).
- cov(GDP, NEER) is significantly negative in columns (2), (3), and (4).
- Better institutional environment is associated with higher FX_AGG.
- Exchange rate peg coefficient is significantly negative in some specifications, but becomes insignificant after controlling for log GDP per capita in column (4).
- FX_AGG correlates strongly with level of development: richer countries have higher foreign-currency exposure (are longer in foreign currency).
- Country size (log population) coefficient is positive and significant.

### Asset-class results — overview and patterns
- Overall consistency: Positive association between openness, level of development, and country size with foreign-currency exposures holds across most asset classes (FDI, portfolio equity (PEQ), portfolio debt (PD), other investment (OI), and reserves).
- Capital controls enter negatively for all asset classes except reserves.
- Institutions:
  - Enter positively for FDI, portfolio equity, and other investment.
  - Enter negatively for portfolio debt and reserves.
- Inflation volatility:
  - Negative and significant for portfolio debt and foreign exchange reserves.
  - Positive for FDI and other investment.
  - Insignificant for portfolio equity except when controlling for GDP per capita in certain specifications.
- GDP volatility:
  - Positive for portfolio debt, other investment, and reserves.
  - Negative for FDI and portfolio equity.
- cov(GDP, NEER) is negative and significant only for portfolio debt.

### Selected coefficient highlights by asset class (Tables 2 and 3)

- FDI (Table 2, columns (1)-(5)):
  - Openness: 0.038*** (1); 0.039*** (2); 0.030*** (3); 0.024*** (4); 0.029*** (5).
  - Inflation volatility: 0.211*** (1); 0.206*** (2); 0.128*** (3); 0.094*** (4); 0.129*** (5).
  - GDP volatility: -0.617*** (1); -0.689*** (2); -0.607*** (3); -0.607*** (4); -0.617*** (5).
  - NEER volatility: -0.155*** (1); -0.149*** (2); -0.082*** (3); -0.054** (4); -0.082*** (5).
  - Log population: 0.011*** (1); 0.010*** (2); 0.011*** (3); 0.012*** (4); 0.014*** (5).
  - Institutions: 0.186*** (1); 0.201*** (2); 0.045*** (3); 0.057*** (4); 0.185*** (5).
  - Capital controls: -0.031*** (1); -0.037*** (2); -0.006 (3); -0.007 (4); -0.018*** (5).
  - Peg: -0.013*** (1); 0.002 (2); -0.013*** (3); -0.020*** (4); -0.016*** (5).
  - Log GDP per capita: 0.037*** (3); 0.041*** (4); 0.035*** (5).
  - Reserve currency: -0.085*** (5).
  - Observations: 1,437 (1),(2),(4),(5); 1,442 (3); 1,484 for reserve-currency subsample. R-squared range: 0.294 to 0.421.

- Portfolio Equity (Table 2, columns (6)-(10)):
  - Openness: -0.016*** (6); -0.015*** (7); -0.020*** (8); -0.018*** (9); -0.010*** (10).
  - Inflation volatility: -0.016 (6); -0.019 (7); -0.072*** (8); -0.048* (9); 0.023 (10).
  - GDP volatility: -0.405*** (6); -0.507*** (7); -0.496*** (8); -0.486*** (9); -0.586*** (10).
  - NEER volatility: 0.027 (6); 0.032 (7); 0.074*** (8); 0.056*** (9); -0.001 (10).
  - cov(GDP, NEER): 0.632 (6); -0.993* (7); -0.946** (8); -1.231** (9).
  - Log population: -0.001 (6); -0.002** (7); -0.003*** (8); -0.002** (9); -0.002* (10).
  - Institutions: 0.185*** (6); 0.200*** (7); 0.159*** (8); 0.160*** (9).
  - Capital controls: -0.025*** (6); -0.017*** (7); -0.016*** (8); -0.016*** (9).
  - Peg, EMU, Log GDP per capita, Reserve currency coefficients reported in Table 2 where applicable.
  - Observations and R-squared reported per column; R-squared range for PEQ regressions includes 0.319 to 0.469.

- Portfolio Debt, Other Investment, Reserves (Table 3, selected coefficients)
  - Portfolio Debt (columns (1)-(5)):
    - Openness: 0.069*** (1); 0.069*** (2); 0.066*** (3); 0.066*** (4); 0.076*** (5).
    - Inflation volatility: -0.484*** (1); -0.481*** (2); -0.448*** (3); -0.506*** (4); -0.294*** (5).
    - GDP volatility: 0.762*** (1); 0.770*** (2); 0.561*** (3); 0.788*** (4); 0.169 (5).
    - NEER volatility: 0.377*** (1); 0.375*** (2); 0.365*** (3); 0.395*** (4); 0.230*** (5).
    - cov(GDP, NEER): -8.578*** (1); -8.446*** (2); -7.104*** (3); -8.650*** (4).
    - Institutions: -0.050*** (1); -0.054*** (2); -0.089*** (3); -0.052** (4).
    - Capital controls: -0.070*** (1); -0.068*** (2); -0.061*** (3); -0.054*** (4); -0.143*** (5).
    - Log GDP per capita: 0.012*** (1); 0.009** (2); 0.006 (3); 0.133*** (5).
    - Observations: 1,437 (cols 1,2,4), 1,442 (3), 1,484 (5). R-squared range: 0.202 to 0.243 for PD columns (1)-(5).

  - Other Investment (columns (6)-(10)):
    - Openness: 0.026*** (6); 0.034*** (7); 0.000 (8); 0.002 (9); -0.008 (10).
    - Inflation volatility: 0.753*** (6); 0.799*** (7); 0.502*** (8); 0.555*** (9); 0.422*** (10).
    - GDP volatility: 0.713* (6); 0.670* (7); 0.470 (8); 0.848** (9); 0.910*** (10).
    - NEER volatility: -0.716*** (6); -0.748*** (7); -0.489*** (8); -0.542*** (9); -0.425*** (10).
    - Institutions: 0.721*** (6); 0.665*** (7); 0.325*** (8); 0.351*** (9).
    - Capital controls: -0.132*** (6); -0.063*** (7); -0.054** (9).
    - Observations and R-squared: Observations range 1,437–1,484; R-squared range 0.484–0.553 for OI columns (6)-(10).

  - Foreign Exchange Reserves (columns (11)-(15)):
    - Openness: 0.023*** (11); 0.015*** (12); 0.034*** (13); 0.028*** (14); 0.026*** (15).
    - Inflation volatility: -0.273*** (11); -0.326*** (12); -0.198*** (13); -0.231*** (14); -0.251*** (15).
    - GDP volatility: 0.656*** (11); 0.637*** (12); 0.802*** (13); 0.568*** (14); 0.693*** (15).
    - NEER volatility: 0.223*** (11); 0.261*** (12); 0.141*** (13); 0.181*** (14); 0.194*** (15).
    - Reserve currency: 0.001 (11); -0.198*** (12); -0.196*** (13).
    - Institutions: -0.216*** (11); -0.146*** (12).
    - Capital controls: 0.096*** (11); 0.079*** (12); 0.052*** (13); 0.046*** (14).
    - Log GDP per capita: -0.056*** (11); -0.035*** (12); -0.033*** (13).
    - Observations: 1,437 (11–14) and 1,484 (15). R-squared range: 0.431–0.540 for reserves columns.

### Summary of substantive findings
- Trade openness is a consistently strong and positive determinant of foreign-currency exposure across aggregate and asset-class regressions.
- Level of development (log GDP per capita) and institutional quality are important: higher development and better institutions are associated with greater foreign-currency exposure for several asset classes, though institutions have mixed signs across asset classes.
- Country size (log population) positively associates with foreign-currency exposure in aggregate and many asset classes.
- Capital controls are generally associated with lower foreign-currency exposure, except for reserves where the relation can be positive.
- Volatility measures show heterogeneous associations across asset classes:
  - Inflation volatility: negative for portfolio debt and reserves; positive for FDI and other investment.
  - GDP volatility: positive for portfolio debt, other investment, and reserves; negative for FDI and portfolio equity.
  - NEER volatility: negative for aggregate in some specs and shows asset-class heterogeneity (positive for PD and reserves, negative for OI in many specs).
  - cov(GDP, NEER): notably negative for aggregate in several specifications and especially negative and significant for portfolio debt.

*Source: wpiea2023237-print-pdf - 3.3  Regression Analysis*

### Appendix B reports the regression analysis by splitting the countries into emerging and

### Appendix B: Regression Analysis and Financial Exchange Rates and Valuation Changes

### Regression results by country group
- The regression analysis splits countries into emerging and advanced economies.
- Focusing on FX_AGG:
  - The positive association between openness and foreign currency exposures remains statistically significant both for emerging and advanced economies (Tables B5 and B8).
  - Inflation volatility is associated with longer positions both for EMEs and advanced economies.
  - The relationship between NEER volatility and FX_AGG is:
    - Positive for EMEs.
    - Negative in some specifications for advanced economies.
  - Interpretation: EMEs more prone to valuation losses due to exchange rate depreciations moved to longer positions.
- Institutions and capital controls:
  - Institutions are associated with longer positions (predominant in EMEs).
  - Capital controls are associated with shorter positions (predominant in EMEs).

### Financial exchange rate indices: definitions
- Asset-weighted and liability-weighted currency indices:
  - I_A_{i,t} = I_A_{i,t−1} × (1 + Σ_j ω^A_{i,j,t} × ∆E_{i,j,t})
  - I_L_{i,t} = I_L_{i,t−1} × (1 + Σ_j ω^L_{i,j,t} × ∆E_{i,j,t})
  - ∆E_{i,j,t} is the percentage change in the bilateral exchange rate of major currencies j in period t (US dollar, euro, Japanese yen, Pound sterling, renminbi, and other currencies).
  - An increase in E is defined as a depreciation of the domestic currency relative to currency j.
- Net financial exchange index:
  - I_F_{i,t} = I_F_{i,t−1} × (1 + ∆I_A_{i,t} × s^A_{i,t} − ∆I_L_{i,t} × s^L_{i,t})
- Handling “other” currencies:
  - Inclusive approach: incorporate “other” currency assets and liabilities in the weights and impute exchange rates using the average change of major currencies relative to the domestic currency.
  - This approach aligns measures more closely with published data for selected countries.

### Statistical properties and correlations (Table 4 summary)
- Data cover 1990-2020.
- Pairwise correlations (medians of within-country correlations between annual percentage changes in exchange rate indices):
  - Financial indices for gross assets and liabilities (A, L): close to 1 and positive across all groups (Table 4, column (1)).
  - Gross debt assets and debt liabilities (A_D, L_D): comparably high correlations (column (2)).
  - Net debt and net equity (N_D, N_EQ): overall pairwise correlation is -0.71 (column (3)).
    - Advanced economies: -0.5.
    - Emerging markets: -0.76.
    - Creditor economies: 0.20.
    - Debtor economies: -0.77.
- Comparison with trade-weighted indices (columns (4) to (6)):
  - Median correlation between changes in the trade-weighted index and the asset-based and liability-based financial-weighted indices: around 0.5 over the full sample.
  - Median correlation between the net financial exchange rate and the trade-weighted exchange rate: 0.15 (positive but much lower).
  - Across income groups (1990-2020):
    - Advanced economies: positive correlation between net financial-weighted and trade-weighted indices.
    - Emerging economies: negative correlation between net financial-weighted and trade-weighted indices.
  - Pre- vs post-GFC differences in correlations noted.
  - Debtor economies: no correlation between the trade index and the financial index.
  - Decomposition:
    - Positive correlations for equity positions between financial and trade indices.
    - Mostly negative correlations for net debt positions.

### Valuation decomposition: definitions and methodology
- Valuation effect due to currency movements (VAL_XR):
  - VAL_XR_{i,t} = ∆I_A_{i,t} × A_{i,t−1} − ∆I_L_{i,t} × L_{i,t−1}
  - A_i and L_i are foreign assets and liabilities relative to GDP.
  - Assumes “other” currencies move according to the average of the main SDR currencies.
- Stock-flow reconciliation of the net international balance sheet:
  - ∆NIIP = FA_{i,t} + VAL_XR_{i,t} + VAL_OTH_{i,t}
  - FA_{i,t} is the financial account (or equivalently the current account balance).
  - VAL_OTH_{i,t} is the change in valuation due to asset prices and other statistical changes.
- The decomposition allows analysis of currency movements’ contribution to overall NIIP changes.

### Episodes: GFC (2008) versus COVID-19 (2020) — key stylized facts and Table 5 highlights
- Nature and dynamics:
  - GFC originated in the banking sector; COVID-19 was a global health shock.
  - COVID-19: no generalized sudden stop of capital flows/current account adjustments as in the GFC.
  - Exchange rate movements during COVID-19 were short-lived (mostly early 2020) compared to longer-lasting changes during the GFC.
  - Stock market performance:
    - Large stock market price gains in the US during COVID-19.
    - Generalized losses during the GFC.
  - Largest valuation gains induced by exchange rate movements were registered in the US during COVID-19 after large but short-lived currency induced losses early in 2020.
- Common patterns across the two episodes:
  - Both episodes induced large wealth aggregate transfers across countries.
  - Valuation gains in Emerging markets and valuation losses in advanced economies in both episodes.
  - Creditor economies (defined on the eve of the GFC) experienced valuation gains; debtors faced losses.
  - The US experienced valuation losses during both events.
  - Valuation changes induced by exchange rates were systematically smaller in magnitude than valuation due to other movements.
  - No evidence of systematic offsetting behavior between valuation changes due to exchange rates and due to other movements.
  - Systematic offset observed between debt and equity within VAL_XR.
- Differences between episodes:
  - Valuation shifts during COVID-19 were generally of a lesser magnitude than during the GFC for different country groups.
  - US losses:
    - During the GFC, losses primarily due to a decline in the value of foreign equity assets exceeding decreases in foreign equity liabilities.
    - During COVID-19, substantial increase in the value of foreign liabilities surpassed those of foreign assets.
  - Global effect on imbalances:
    - COVID-19: valuation gains due to asset prices increased global imbalances (“destabilizing”).
    - GFC: valuation changes due to other movements played a stabilizing role for cross-country NIIP distribution.
  - These wealth transfers were not fully reversed in the short-term in 2021, similar to the GFC period.

- Selected numeric highlights from Table 5 (weighted average in percent of group (or economy) GDP):
  - COVID-19 (2020), Full sample:
    - FA: -0.1
    - ∆ NIIP: -1.4
    - VAL_XR Total: -0.3; Debt & FXR: -0.8; Equity: 0.6; VAL_OTH Total: -1.0; Debt & FXR: 0.5; Equity: -1.5; VAL_TOTAL: -1.3; Debt & FXR: -0.3; Equity: -1.0
  - COVID-19 (2020), Advanced Economies:
    - FA: -0.5
    - ∆ NIIP: -2.8
    - VAL_XR Total: -0.8; Debt & FXR: -1.1; Equity: 0.3; VAL_OTH Total: -1.6; Debt & FXR: 0.5; Equity: -2.1; VAL_TOTAL: -2.3; Debt & FXR: -0.5; Equity: -1.8
  - COVID-19 (2020), USA:
    - FA: -3.2
    - ∆ NIIP: -13.3
    - VAL_XR Total: 3.3; Debt & FXR: -1.4; Equity: 4.7; VAL_OTH Total: -13.3; Debt & FXR: 0.3; Equity: -13.6; VAL_TOTAL: -10.1; Debt & FXR: -1.1; Equity: -9.0
  - COVID-19 (2020), Emerging Economies:
    - FA: 0.6
    - ∆ NIIP: 1.1
    - VAL_XR Total: 0.6; Debt & FXR: -0.4; Equity: 1.0; VAL_OTH Total: 0.2; Debt & FXR: 0.6; Equity: -0.6; VAL_TOTAL: 0.9; Debt & FXR: 0.1; Equity: 0.4
  - COVID-19 (2020), Creditors Economies:
    - FA: 1.3
    - ∆ NIIP: 4.7
    - VAL_XR Total: -0.7; Debt & FXR: -0.6; Equity: -0.2; VAL_OTH Total: 4.2; Debt & FXR: 1.9; Equity: 2.3; VAL_TOTAL: 3.4; Debt & FXR: 1.3; Equity: 2.1
  - COVID-19 (2020), Debtors Economies:
    - FA: -1.1
    - ∆ NIIP: -5.5
    - VAL_XR Total: 0.1; Debt & FXR: -1.0; Equity: 1.1; VAL_OTH Total: -4.7; Debt & FXR: -0.4; Equity: -4.2; VAL_TOTAL: -4.7; Debt & FXR: -1.5; Equity: -3.1
  - GFC (2008), Full sample:
    - FA: -0.8
    - ∆ NIIP: 0.1
    - VAL_XR Total: 0.9; Debt & FXR: -1.9; Equity: 2.7; VAL_OTH Total: 0.6; Debt & FXR: 2.3; Equity: -2.3; VAL_TOTAL: 0.9; Debt & FXR: 0.3; Equity: 0.5
  - GFC (2008), Advanced Economies:
    - FA: -2.4
    - ∆ NIIP: -4.3
    - VAL_XR Total: 1.0; Debt & FXR: -2.2; Equity: 3.1; VAL_OTH Total: -2.3; Debt & FXR: 3.2; Equity: -6.4; VAL_TOTAL: -2.0; Debt & FXR: 1.0; Equity: -3.3
  - GFC (2008), USA:
    - FA: -5.4
    - ∆ NIIP: -19.5
    - VAL_XR Total: -2.1; Debt & FXR: 0.9; Equity: -3.0; VAL_OTH Total: -11.9; Debt & FXR: 1.3; Equity: -13.2; VAL_TOTAL: -14.1; Debt & FXR: 2.2; Equity: -16.3
  - GFC (2008), Emerging Economies:
    - FA: 4.1
    - ∆ NIIP: 12.7
    - VAL_XR Total: 0.6; Debt & FXR: -1.0; Equity: 1.6; VAL_OTH Total: 9.3; Debt & FXR: -0.4; Equity: 8.9; VAL_TOTAL: 9.7; Debt & FXR: -1.7; Equity: 11.4
  - GFC (2008), Creditors Economies:
    - FA: 4.6
    - ∆ NIIP: 10.2
    - VAL_XR Total: 0.3; Debt & FXR: -5.0; Equity: 5.3; VAL_OTH Total: 5.3; Debt & FXR: 4.7; Equity: 0.5; VAL_TOTAL: 5.5; Debt & FXR: -0.3; Equity: 5.8
  - GFC (2008), Debtors Economies:
    - FA: -4.2
    - ∆ NIIP: -5.6
    - VAL_XR Total: 1.2; Debt & FXR: -0.1; Equity: 1.3; VAL_OTH Total: -2.4; Debt & FXR: 0.9; Equity: -3.9; VAL_TOTAL: -2.0; Debt & FXR: 0.7; Equity: -2.6

### Implications and conclusion
- Dataset assembled: currency composition of external balance sheets and financial exchange rates for 50 economies over 1990–2020.
- Stylized facts:
  - The US dollar and the euro remain the dominant currencies for global holdings of assets and liabilities.
  - Economies have improved their currency exposures over time; emerging markets no longer display the textbook short foreign currency position.
  - Financially-weighted exchange rates have weak correlation with trade-weighted indices and can be a more appropriate metric for measuring wealth effects of currency movements.
- Application illustrated:
  - Comparison of wealth transfers during COVID-19 and the GFC shows substantial, broad-based shifts, with emerging markets experiencing valuation gains in both episodes.
  - COVID-19 valuation changes increased global imbalances (“destabilizing”), whereas the GFC saw reductions in global imbalances (“stabilizing”).
- Future use:
  - The dataset can be used to study unresolved issues in open-economy macroeconomics and guide the design of open economy models.

*Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023237-print-pdf.pdf*

### References

### References

### Core bibliographic sources
- Lists extensive references cited in the paper, including working papers, journal articles, NBER and CEPR papers, BIS and IMF research contributions, and books. Representative authors and works include Adler and Garcia-Macia (2018); Alfaro, Calani, and Varela (2021); Arslanalp, Eichengreen, and Simpson-Bell (2022); Bénetrix et al. (2015, 2019); Lane and Milesi-Ferretti (2003, 2007a, 2007b, 2008, 2012, 2018); Gourinchas and Rey (2007); Gopinath and Stein (2018); Maggiori, Neiman, and Schreger (2019, 2020); Ito and McCauley (2020); and many others cited in the source text.

### Data appendix (Section A): dataset scope, sources, and methodology
- Coverage and scope:
  - Dataset covers the period 1990 to 2020.
  - Data for Russia start in 1993.
  - Data for the Czechia are from 1993.
- Main data sources:
  - IMF survey to country authorities on currency of denomination of each IIP component (survey sent to authorities of 52 economies).
  - Coordinated Portfolio Investment Survey (CPIS).
  - BIS International Debt Issuance Statistics.
  - BIS Locational Banking Statistics (LBS).
  - IMF’s International Reserves and Foreign Currency Liquidity (IRFCL) and COFER shares for reserves estimates.
- Survey details:
  - Survey requested data from 1990 on IIP components broken down into five SDR currencies (US dollar, euro, Japanese yen, Pound sterling, and renminbi), domestic currency, and “other currencies.”
  - Response rate was 85%, with partial submissions accepted.
  - For recent years, around 60% of economies in the sample reported some data.
- Synthetic data and estimations:
  - “Synthetic data” used to fill gaps include BIS International Debt Issuance Statistics and BIS LBS, treated as proxies for portfolio debt liabilities and other investments.
  - Financial derivatives are excluded from analysis due to methodological and collection issues.
  - Estimation methodology builds on Lane and Shambaugh (2010a) and Bénetrix et al. (2015), with improvements and a novel decomposition of FDI into equity and debt components.
- Component-specific approaches:
  - Portfolio equity: primary source IMF Survey and CPIS Table 2; when unavailable, an improved Lane and Shambaugh (2010) geography-based method using CPIS geographical distribution; offshore financial centers excluded; gravity-based model estimates (Benetrix et al., 2019) used when CPIS pairs are missing.
  - FDI: equity and debt components estimated separately following Benetrix et al. (2019); CDIS used for 2009–2020; FDI debt can account for up to 40% of total FDI in some countries and is estimated separately using portfolio debt as proxy when needed.
  - Portfolio debt: combines CPIS bilateral positions with BIS currency distribution of debt issued by source countries; adjustments made to net out holdings by domestic residents in US dollars, euros, and yen using US Treasury, ECB, and Bank of Japan holdings data.
  - Other investment: survey information available for 23 economies; complemented for 2018–2020 with BIS LBS; earlier years sourced from Benetrix et al. (2019).
  - Reserves: built on Bénetrix et al. (2015) (1990–2012) and Bénetrix et al. (2019) (2012–2017), extended for 2017–2020 combining IMF Survey, IRFCL, central bank publications, Ito and McCauley (2020), and COFER; interpolation applied where needed (linear interpolation using five-year windows preceding updated data).
- Data labeling and confidentiality:
  - All data labeled public unless country authorities objected.

### Data combination and country coverage matrices (Tables A1–A3)
- Table A1: Assets — lists per-country year ranges for Actual Data, Estimated Data, and Synthetic Data across FDI Equity, FDI Debt, Portfolio Equity, Portfolio Debt, and Other Investment (1990–2020 coverage shown at country level; country names reported as ISO codes).
- Table A2: Reserves Assets — shows per-country Actual Data and Estimated Data year ranges (notes: * indicate partial information for some years; country names reported as ISO codes).
- Table A3: Liabilities — provides per-country breakdown for FDI Debt, Portfolio Debt, and Other Investment indicating Actual Data and Synthetic data year ranges; portfolio equity liabilities and FDI equity liabilities are denominated in host country currency and thus excluded from the table.

Key dataset facts preserved exactly:
- Survey sent to authorities of 52 economies representing over 90 percent of the world’s GDP.
- As of December 2022, 82 economies reported core data to CPIS; 57 economies reported CPIS Table 2 as of December 2022.
- Some explicit country-year entries (examples from tables): Russia data start in 1993; Czechia data from 1993.

### Regression variables and sources (Table A4)
- Macro variables and their data sources:
  - Openness: Openness [(Export+Import) / (2XGDP)] — World Economic Outlook.
  - Inflation volatility: CPI rolling standard deviation using 15-year window — Information Notice System (INS).
  - GDP volatility: Real GDP growth rolling standard deviation using 15-year window — World Economic Outlook.
  - NEER volatility: Change in NEER rolling standard deviation using 15-year window — Information Notice System (INS).
  - cov (GDP, NEER): Covariance between GDP growth and change in NEER — World Economic Outlook, INS.
  - Log population: UN World Population Prospects, 2019 Vintage.
  - Institutions: Political Risk Rating — International Country Risk Guide (ICRG).
  - Capital controls: Capital Controls Index, FARI — IMF (Baba et.al., 2022).
  - Peg: classification from Ilzetzki et al. (2021) (Peg if coarse classification 3, 4, 5; Peg otherwise).
  - EMU: Euro Area dummy variable (Dummy=1 if country belongs to Euro Area).
  - Log GDP per capita: Log of Real GDP in PPP terms, per capita — World Economic Outlook.
  - Reserve currency: Share of the country’s currency held as FX reserves by central banks worldwide — IMF, COFER.

### Additional results (Section B): regression output, decompositions, and event comparisons
- Regression tables (B5–B10) present determinants of FX aggregate exposure and component exposures (FDI, Portfolio Equity, Portfolio Debt, Other Investment, Reserves) separately for EMEs and Advanced economies. Examples of reported statistics include:
  - Table B5 (EME FX AGG): Observations 692, R-squared values across specifications include 0.468, 0.469, 0.413, 0.491, 0.489.
  - Table B6 (EMEs FDI and Portfolio Equity): Observations e.g., 1,437 and 1,484 across columns; R-squared values range from 0.294 to 0.469 depending on specification.
  - Table B7 (EMEs PD, OI, Reserves): Observations 692–715 across columns; R-squared values reported (example: Reserves R-squared 0.334, 0.367, 0.353, 0.377, 0.371 across columns).
  - Table B8 (Advanced FX AGG): Observations 745–769; R-squared values 0.311, 0.458, 0.360, 0.489, 0.513.
  - Table B9 and Table B10 provide analogous detailed outputs for Advanced economies across asset categories.
- Statistical significance notation: ***, **, and * denote, respectively p <0.01, p <0.05, and p <0.1. Regressions include year fixed effects and report robust standard errors in parentheses.
- Decomposition and event comparison tables (B11–B13) and figures:
  - Table B11: Financial and Trade Weighted Exchange Rate, Detailed Decomposition — reports medians of within-country correlations between annual percentage changes in components of the financial exchange rate and the trade weighted index; data cover 1990-2020. Example entries: Full Sample All (including FXR) 0.15; All (excluding FXR) -0.10; All FDI 0.32; All Portfolio -0.09; All Debt -0.35; All Equity 0.46; All debt (including FXR) -0.27.
  - Table B12: COVID-19 versus the Global Financial Crisis: Gross Assets — shows stock-flow reconciliation of asset side with weighted averages in percent of group (or economy) GDP. Example entries for COVID-19 (2020), Full sample: FA Assets 7.6, ∆ Assets 19.0, V AL_XR -0.1, V AL_OTH -0.6, Total VAL 0.61, Total VAL breakdowns include 12.1, 5.7, 6.2, 12.0, 5.1, 6.7 (as presented in table structure).
  - Table B13: COVID-19 versus the Global Financial Crisis: Gross Liabilities — analogous reconciliation for liabilities. Example entries for COVID-19 (2020), Full sample: FA Liabilities 7.6, ∆ Liabilities 20.4, V AL_XR 0.2, V AL_OTH 0.2, Total VAL 0.01, Total VAL breakdowns include 12.6, 4.9, 7.7, 12.8, 5.1, 7.7.
  - Figures B1 and B2: cross-country bar plots contrasting 2020 and 2008 valuation changes due to exchange rate movements (V AL_XR) and overall ∆ NIIP decompositions; examples include country-level labels plotted in the figures (e.g., IRL, NLD, CHE, SGP, JPN, HKG, USA among others).
- Notes on classification:
  - Creditor and Debtor economies classified according to being a debtor or creditor on the eve of the global financial crisis.
  - Table captions and notes preserve exact terminologies used in decompositions: FA (financial account), ∆ Assets / ∆ Liabilities, V AL_XR (valuation change induced by exchange rate movements), V AL_OTH (valuation change due to asset prices and other statistical changes), V AL_TOTAL (overall change in valuation).

*Currencies of External Balance Sheets — Working Paper No. WP/2023/237*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023237-print-pdf.pdf_