## wp1804

## Source details

**Canonical URL:** [wp1804](https://www.imf.org/-/media/files/publications/wp/2018/wp1804.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp1804.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp1804.pdf.json)

---

### Introduction and research objective
- First analysis, to the authors' knowledge, of the effects of uncertainty shocks on cross-border banking flows using bilateral BIS LBS data.
- Motivation: cross-border bank lending rose markedly from the early 1990s to the Global Financial Crisis (GFC); cross-border activity sharply slowed during the GFC alongside heightened uncertainty.
- Identification: exploits dyadic BIS LBS structure to include recipient country-time fixed effects to better separate supply-side (push) from demand-side (pull) factors.

### Data and sample
- Main dataset: Bank for International Settlements (BIS) Locational Banking Statistics (LBS) on an unconsolidated, residency basis.
- Coverage and sample construction:
  - 46 reporting countries in LBS; after dropping offshore financial centers (IMF classification) benchmark analysis focuses on 25 reporting countries with available uncertainty and macroeconomic controls.
  - After dropping offshore financial centers from recipient list, sample includes 50 recipient countries.
  - Observations with cross-border flows less than $5 million or with negative total outstanding claims are dropped.
  - Dependent variable winsorized at the one percent.
  - BIS LBS captures around 93 percent of all cross-border interbank business.
- Uncertainty measures:
  - Baseline: quarterly stock market realized volatility from Baker and Bloom (2013).
  - Alternative: Economic Policy Uncertainty (EPU) index (Baker et al., 2016) available for 15 countries.
  - Idiosyncratic stock market volatility: country-specific volatility purged of contemporaneous VIX via residuals.
  - Correlation between stock market volatility and EPU across 15 countries: average correlation 0.38; range 0.03 (Sweden) to 0.76 (Brazil).
- Controls (lagged one quarter): real GDP growth, stock market growth, inflation rate, monetary policy (policy or interbank) rate, nominal exchange rate growth, private credit growth, external debt to GDP ratio.
- Exchange-rate adjusted flows: authors reconstruct stocks and take log differences to obtain quarterly growth rates (∆L and ∆B).

### Methodology
- Core empirical specifications:
  - Claims (∆L): ∆L_{i,j,t} = α_{j,t} + β X_{i,t−1} + γ UNC_{i,t−1} + ε_{i,j,t}
  - Liabilities (∆B): ∆B_{i,j,t} = α_{j,t} + β X_{i,t−1} + γ UNC_{i,t−1} + ε_{i,j,t}
  - i = reporting (source) country; j = counterparty (recipient) country; α_{j,t} are recipient country-time fixed effects.
  - All explanatory variables lagged by one quarter; standard errors clustered at reporter and counterparty country-pair levels.
- Outcome interpretation:
  - Negative γ in claims equation → retrenchment (decrease in cross-border claims) when home uncertainty rises.
  - Negative γ in liabilities equation → stops (decrease in gross inflows) when home uncertainty rises.

### Main empirical findings
- Baseline results:
  - Higher uncertainty in a home country reduces both cross-border claims (retrenchment) and cross-border liabilities (stops); effects are economically and statistically significant.
  - Example magnitude: an increase in uncertainty from the historical median to the level observed during the GFC is associated with a reduction in cross-border claims of 1.6-2.8 percentage points.
- Selected baseline coefficients (Table 2):
  - Log of uncertainty (claims, columns I–III): -1.670** (0.824); -2.845** (1.150); -2.716** (1.211)
  - Log of uncertainty (liabilities, columns IV–VI): -2.369** (0.957); -2.734** (1.328); -2.302* (1.289)
  - Real GDP growth (claims): 0.852*** (0.290); 0.782** (0.402); 0.391 (0.422)
  - Policy rate (claims): 0.557*** (0.107); 0.718*** (0.136); 0.669*** (0.133)
  - Nominal exchange rate growth (claims): -0.094 (0.078); -0.123 (0.088); -0.153* (0.089)
  - External debt to GDP: -0.022*** (0.004); -0.018*** (0.006)
  - Observations and fit:
    - Claims: Obs 30,608; 17,462; 16,431 across columns; R-squared 0.13; 0.14; 0.15
    - Liabilities: Obs 29,889; 16,725; 15,416 across columns; R-squared 0.14; 0.15; 0.16
- Sample-specific patterns:
  - Emerging market economies: restricting sample yields larger negative coefficients—uncertainty shocks have larger negative effects on flows into emerging market economies than into advanced economies.
  - Spillovers from advanced economies: uncertainty in advanced economies reduces cross-border banking flows into emerging market economies.

### Robustness checks and subsample analyses
- Alternative uncertainty measures:
  - EPU index (Table 4): negative effects on growth of cross-border claims and liabilities remain significant and quantitatively similar to baseline.
    - Log of uncertainty (claims): -1.923** (0.930); -2.664** (1.211); -3.574*** (1.239)
    - Log of uncertainty (liabilities): -2.621** (1.207); -3.654** (1.720); -2.832* (1.681)
    - Observations: Claims Obs 21,564; 14,623; 13,715. Liabilities Obs 21,212; 14,009; 13,369.
  - Idiosyncratic stock market volatility (Table 5): negative effects persist.
    - Log of uncertainty (claims): -0.069* (0.039); -0.119** (0.050); -0.125** (0.055)
    - Log of uncertainty (liabilities): -0.100** (0.047); -0.117* (0.068); -0.106* (0.062)
- Before vs. after GFC (Table 6):
  - Uncertainty increases predict slowdown in cross-border claims and liabilities both before and after the GFC.
  - Log of uncertainty:
    - Before GFC (1995Q1–2007Q2) claims: -1.983** (1.026)
    - After GFC (2007Q3–2012Q4) claims: -2.841** (1.289)
    - Before GFC liabilities: -2.111** (1.052)
    - After GFC liabilities: -3.550* (1.910)
  - Observations: Claims Obs 18,846; 11,578. Liabilities Obs 18,808; 11,081.
- Euro area vs. non-euro area (Table 7):
  - Log of uncertainty (claims):
    - Euro area: -5.136** (2.444)
    - Non-euro area: -2.827* (1.511)
  - Observations: Euro area Obs 6,559; Non-euro area Obs 9,508. R-squared 0.32 and 0.22 (claims).

### Rebalancing channel of global banks (mechanism)
- Hypothesis: global banks rebalance portfolio toward relatively safer foreign borrowers when home uncertainty rises.
- Share proxy:
  - s_{i,j,t} = cross-border claims_{i,j,t} / (cross-border claims_{i,j,t} + local claims in foreign currencies_{i,t}) × 100
- Estimation: s_{i,j,t} = α_{j,t} + β X_{i,t−1} + γ UNC_{i,t−1} + ε_{i,j,t}
- Findings (Tables 8–9):
  - Uncertainty variable switches sign relative to absolute cross-border claims: higher home uncertainty associated with an increase in the share of cross-border claims (rebalancing toward foreign borrowers).
  - Baseline coefficients (Table 8) — Log of uncertainty: 0.283* (0.166); 0.465** (0.191); 0.359* (0.185)
  - Rebalancing is stronger since the GFC (post-GFC coefficients larger and more significant).
  - Rebalancing depends on perceived riskiness: operates for lending to advanced economies but not emerging market economies (evidence of “flight-to-quality”).
- Data limitation:
  - BIS LBS did not provide historical total domestic claims by global banks in local currencies prior to 2012; hence local claims in foreign currencies used as domestic comparator.
- Complementary exercise (Appendix B) using share of cross-border claims of global banks to total domestic claims yields qualitatively similar results.

### Instrumental variable approach (addressing endogeneity)
- IV strategy: use natural disasters (extreme weather and geological events as defined by CRED) scaled by increase in media mentions in 15-days after shock vs 15-days before as instrument for uncertainty.
- Estimation: two-stage least squares (2SLS).
  - First stage: measures of uncertainty regressed on instruments.
    - Cragg-Donald Wald F-statistics far above Stock and Yogo (2005) critical values → “strong instruments.”
  - Second stage: re-estimate rebalancing equation using fitted uncertainty from first stage.
    - IV results confirm OLS rebalancing results but with substantially larger coefficient on uncertainty (OLS biased downwards).
    - IV baseline (Table 10) — Log of uncertainty: 1.513* (0.884); 1.427* (0.827); 1.735* (0.980)
    - Cragg-Donald Wald F-statistic: 207.18; 321.86; 256.47. Stock-Yogo 5% critical value 16.38 in each column.
    - Observations: 24,420; 14,690; 14,101. R-squared 0.48; 0.48; 0.50.

### Safe vs. risky borrowers (interaction evidence)
- Specification: s_{i,j,t} = α_{j,t} + β X_{i,t−1} + γ UNC_{i,t−1} + δ EM_{j} UNC_{i,t−1} + ε_{i,j,t}
  - EM_{j} = 1 if counterparty j is an emerging market economy; 0 otherwise.
- Findings (Table 11):
  - Log of uncertainty: 0.497* (0.262); 0.852*** (0.322); 0.734** (0.299)
  - Interaction (Log uncertainty × counterparty EM dummy): -0.427* (0.256); -0.909*** (0.345); -0.901*** (0.336)
  - Interpretation: rebalancing toward safer foreign borrowers occurs only toward advanced economies, not emerging market economies — consistent with flight-to-quality.

### Key descriptive statistics (Table 1)
- Growth of cross-border claims (q-o-q): Obs 30,608; Mean 3.136; Median 1.225; Standard deviation 40.751
- Growth of cross-border liabilities (q-o-q): Obs 29,889; Mean 2.998; Median 1.359; Standard deviation 50.930
- Stock market volatility: Obs 30,608; Mean 19.943; Median 17.211; Standard deviation 10.238
- Economic policy uncertainty: Obs 24,901; Mean 105.921; Median 97.412; Standard deviation 44.417
- Real GDP growth (q-o-q): Obs 30,608; Mean 0.600; Median 0.645; Standard deviation 1.059
- Growth of stock market (q-o-q): Obs 30,608; Mean 1.254; Median 2.214; Standard deviation 9.826
- Inflation rate (q-o-q): Obs 30,608; Mean 0.609; Median 0.573; Standard deviation 0.626
- Policy rate: Obs 30,608; Mean 3.666; Median 3.370; Standard deviation 2.653
- Growth of nominal exchange rate (q-o-q): Obs 30,608; Mean -0.127; Median 0.000; Standard deviation 4.243
- Growth of private credit (q-o-q): Obs 17,973; Mean 1.506; Median 1.372; Standard deviation 2.143
- External debt to GDP ratio: Obs 19,605; Mean 80.704; Median 74.052; Standard deviation 53.674
- Note: “Growth rates are calculated quarter-over-quarter. All variables are in percentage points.”

### Appendix highlights
- Appendix A: country-specific uncertainty index panels and event classification; list of countries in final sample.
- Appendix A.3: Before/after GFC results using Economic Policy Uncertainty show:
  - Log of uncertainty (example): (I) -2.841** (1.289); (IV) -6.026*** (2.225)
- Appendix A.4 and A.5: rebalancing results with idiosyncratic volatility and IV with alternative measures:
  - Table A.4 Log of uncertainty: (I) 0.012* (0.007); (II) 0.018** (0.008); (III) 0.015** (0.007)
  - Table A.5 IV — Log of uncertainty (EPU): (I) 1.194* (0.683); (II) 1.499* (0.798); (III) 1.392* (0.817)
  - Table A.5 IV — Log of uncertainty (idiosyncratic): (IV) 0.102* (0.060); (V) 0.098* (0.057); (VI) 0.124* (0.067)
- Appendix B: alternative rebalancing measure using cross-border claims as share of total domestic claims (s̃_{i,j,t}) — results qualitatively similar and show:
  - Log of uncertainty (stock market volatility): (I) 3.012*** (0.670); (II) 9.125*** (1.833); (III) 4.979*** (1.108)
  - Interaction Log uncertainty × counterparty EM dummy: -4.830*** (1.247) and -2.149*** (0.808) in respective specifications.

### Conclusion and key findings
- Contribution: uses dyadic BIS LBS structure with recipient country-time fixed effects to identify country-specific uncertainty shocks as both push and pull factors of cross-border banking flows.
- Core conclusions:
  - Higher country-specific uncertainty reduces cross-border banking flows (both claims and liabilities).
  - Global banks rebalance lending composition toward foreign borrowers when local uncertainty increases.
  - Rebalancing constitutes flight-to-quality: observed toward advanced economies but not emerging market economies.
- Robustness: results hold with alternative uncertainty measures (EPU, idiosyncratic volatility), across subperiods, country groups, and under IV estimation using natural disasters scaled by media attention.

*Source: wp1804 (IMF Working Paper content unit).*

### References 24

### wp1804 - References 24

### Introduction and research objective
- Paper provides a first analysis, to the authors' knowledge, of the effects of uncertainty shocks on cross-border banking flows using bilateral BIS LBS data.
- Motivation: cross-border bank lending rose markedly from the early 1990s to the Global Financial Crisis (GFC); cross-border activity sharply slowed during the GFC alongside heightened uncertainty.
- Identification strategy exploits the dyadic structure of BIS LBS data to control for recipient country-time fixed effects and better separate supply-side (push) from demand-side (pull) factors.

### Data and sample
- Main dataset: Bank for International Settlements (BIS) Locational Banking Statistics (LBS), capturing outstanding claims and liabilities on an unconsolidated, residency basis.
- Coverage and sample construction:
  - 46 reporting countries in LBS; after dropping offshore financial centers (IMF classification) the benchmark analysis focuses on 25 reporting countries with available uncertainty and macroeconomic controls.
  - After dropping offshore financial centers from recipient list, sample includes 50 recipient countries.
  - Observations with cross-border flows less than $5 million or with negative total outstanding claims are dropped.
  - Dependent variable winsorized at the one percent.
  - BIS LBS captures around 93 percent of all cross-border interbank business.
- Uncertainty measures:
  - Baseline: quarterly stock market realized volatility from Baker and Bloom (2013).
  - Alternative: Economic Policy Uncertainty (EPU) index (Baker et al., 2016) available for 15 countries in the sample.
  - Idiosyncratic stock market volatility: country-specific volatility purged of contemporaneous VIX via residuals.
  - Correlation between stock market volatility and EPU across 15 countries: average correlation 0.38; range 0.03 (Sweden) to 0.76 (Brazil).
- Controls included (lagged one quarter): real GDP growth, stock market growth, inflation rate, monetary policy (policy or interbank) rate, nominal exchange rate growth, private credit growth, external debt to GDP ratio.
- Exchange-rate adjusted flows: BIS LBS provides exchange-rate adjusted flows; authors reconstruct stocks and take log differences to obtain quarterly growth rates (∆L and ∆B).

### Methodology
- Core empirical specifications:
  - For cross-border claims (∆L): ∆L_{i,j,t} = α_{j,t} + β X_{i,t−1} + γ UNC_{i,t−1} + ε_{i,j,t}
  - For cross-border liabilities (∆B): ∆B_{i,j,t} = α_{j,t} + β X_{i,t−1} + γ UNC_{i,t−1} + ε_{i,j,t}
  - i = reporting (source) country; j = counterparty (recipient) country; α_{j,t} are recipient country-time fixed effects.
  - All explanatory variables lagged by one quarter; standard errors clustered at reporter and counterparty country-pair levels.
- Outcome interpretation:
  - Negative γ in claims equation indicates retrenchment (decrease in cross-border claims) when home uncertainty rises.
  - Negative γ in liabilities equation indicates stops (decrease in gross inflows) when home uncertainty rises.

### Main empirical findings
- Baseline results (Table 2 summary):
  - Higher uncertainty in a home country reduces both cross-border claims (retrenchment) and cross-border liabilities (stops); effects are economically and statistically significant.
  - Example magnitude: an increase in uncertainty from the historical median to the level observed during the GFC is associated with a reduction in cross-border claims of 1.6-2.8 percentage points.
  - Control variable patterns:
    - Higher domestic real GDP growth → greater cross-border lending by home banks.
    - Domestic monetary policy tightening has a positive effect on gross cross-border claims in baseline (consistent with Correa et al., 2017).
    - Depreciation of local currency (increase in nominal exchange rate) associated with slowdown in cross-border lending, but effect not robust in all specifications.
    - Higher external debt to GDP ratio is negatively related to cross-border lending (highly statistically significant).
- Sample-specific comparisons:
  - Emerging market economies: restricting sample to emerging market counterparties yields larger negative coefficients—i.e., uncertainty shocks have larger negative effects on cross-border banking flows into emerging market economies than into advanced economies.
  - Spillovers from advanced economies: uncertainty in advanced economies reduces cross-border banking flows into emerging market economies (consistent with Gauvin et al., 2014).

### Robustness checks and subsample analyses
- Alternative uncertainty measures:
  - EPU index: negative effects on growth of cross-border claims and liabilities remain significant and quantitatively similar to baseline (Table 4).
  - Idiosyncratic stock market volatility (purged of VIX): negative effects persist (Table 5).
- Before and after GFC:
  - Sample split into pre-GFC (1995Q1-2007Q2) and post-GFC (2007Q3-2012Q4).
  - Uncertainty increases predict slowdown in cross-border claims and liabilities both before and after the GFC (Table 6; EPU results in Table A.3).
  - Observed decrease in the size and significance of policy rate effects after the GFC.
- Euro area vs. non-euro area:
  - Reporting countries split between euro and non-euro area.
  - Negative effect of domestic uncertainty on cross-border banking flows exists in both groups; effect is larger and more precisely estimated in euro area economies (Table 7).

### Rebalancing channel of global banks (mechanism)
- Hypothesis: global banks may rebalance their portfolio toward relatively safer foreign borrowers when home-country uncertainty rises.
- Proxy for rebalancing: share of cross-border claims in total claims approximated by
  s_{i,j,t} = cross-border claims_{i,j,t} / (cross-border claims_{i,j,t} + local claims in foreign currencies_{i,t}) × 100
  (both numerator and denominator are BIS LBS series converted to U.S. dollars).
- Estimation: s_{i,j,t} = α_{j,t} + β X_{i,t−1} + γ UNC_{i,t−1} + ε_{i,j,t} (equation (4)).
- Findings (Table 9):
  - Uncertainty variable switches sign relative to absolute cross-border claims: higher home uncertainty is associated with an increase in the share of cross-border claims, implying rebalancing toward foreign borrowers.
  - Interpretation: although global banks reduce cross-border lending in absolute terms under higher uncertainty, they reduce local lending in foreign currencies by more, effectively increasing the cross-border share.
  - Rebalancing is stronger since the GFC (the channel becomes more economically and statistically significant in the post-GFC sample).
  - Rebalancing depends on perceived riskiness of recipient countries: channel operates for lending to advanced economies but not emerging market economies (evidence of “flight-to-quality”).
- Data limitations noted:
  - BIS LBS did not provide historical total domestic claims by global banks in local currencies prior to 2012; authors therefore use local claims in foreign currencies as the domestic comparator.
  - Complementary analysis using fraction of cross-border claims of global banks to total domestic claims (including all depository institutions) reported in Appendix B, qualitatively similar.

### Additional methodological and empirical notes
- Identification addresses demand-side confounders by including recipient country-time fixed effects, thereby controlling for any macroeconomic shock affecting credit demand in recipient countries.
- All explanatory variables are lagged by one quarter to reduce reverse causality concerns.
- Winsorization at the one percent for the dependent variable to control for extreme outliers.
- Observations with cross-border flows less than $5 million or with negative outstanding claims removed.

*Source: wp1804 - References 24 (internal content of the chapter/section).*

### appendix shows that these results still hold when we use the idiosyncratic uncertainty

### wp1804 - appendix shows that these results still hold when we use the idiosyncratic uncertainty

### Instrumental variable approach
- Endogeneity concern: unobserved factors might drive uncertainty and macroeconomic conditions in a home country simultaneously; controlling for GDP growth and stock market growth in a reporting country mitigates this concern but may not eliminate it.
- IV strategy: in the spirit of Baker and Bloom (2013), use natural disasters (extreme weather and geological events as defined by the Center for Research on the Epidemiology of Disasters (CRED)) as instruments, scaled by the increase in media mentions of the country in the 15-days after the shock compared to the 15-days before the shock.
- Estimation procedure: two-stage least squares (2SLS).
  - First stage: regress measures of uncertainty on the instruments.
    - Empirical diagnostics: Cragg-Donald Wald F-statistics are far above the Stock and Yogo (2005) critical values for weak instruments in all cases, indicating “strong instruments.”
    - Hansen’s J statistics for valid instruments are not reported because the equation is exactly identified (only one instrument variable).
  - Second stage: re-estimate equation (4) using the exogenous part of stock market volatility driven by the instrument (the fitted value from the first step).
    - Result: the IV estimates confirm the OLS results in Table 8, but the size of the coefficient on uncertainty increases substantially, implying that the OLS estimates are biased downwards.
    - Robustness: Table A.5 in the appendix confirms that the results from an IV approach still hold when using alternative measures of uncertainty.

### Safe vs. risky borrowers
- Hypothesis: if higher uncertainty in a local economy encourages global banks to switch lending toward relatively safer foreign borrowers, this mechanism could be weaker when banks lend to borrowers who are genuinely perceived as risky (flight-to-quality).
- Empirical test: interact the main independent variable of uncertainty, 푈푁퐶 푖,푡−1, with a dummy variable 퐸푀 푗 equal to one if counterparty country j is an emerging market economy and zero otherwise, in the following specification:
  - 푠 푖,푗,푡 = 훼 푗,푡 + 훽 푋 푖,푡−1 + 훾 푈푁퐶 푖,푡−1 + 훿 퐸푀 푗 푈푁퐶 푖,푡−1 + 휀 푖,𝑗,𝑡  (equation (5))
- Findings:
  - The interaction term (훿) is negative and statistically significant.
  - Interpretation: global banks switch their lending toward relatively safer foreign borrowers when facing higher uncertainty in a local economy, but this rebalancing occurs only for lending toward advanced economies, not emerging market economies.
  - Consistency with literature: aligns with flight-to-quality behavior observed during episodes such as the Asian Financial Crisis, 9/11, the collapse of Lehman Brothers, and the Taper Tantrum (Beber et al., 2008; Caballero and Krishnamurthy, 2008; De Bock and Filho, 2015) and with literature showing larger effects of uncertainty shocks in emerging markets (Carrière-Swallow and Céspedes, 2013; Choi, forthcoming).

### Conclusion and key findings
- Contribution: uses the dyadic structure of BIS LBS data to control for shocks affecting recipient countries and better identify country-specific uncertainty shocks as both push and pull factors of cross-border banking flows.
- Main empirical results:
  - Higher uncertainty—measured by country-specific stock market volatility—in a local economy reduces cross-border banking flows from/into this economy.
  - Global banks rebalance their lending composition toward foreign borrowers when local uncertainty increases.
  - The rebalancing is observed only toward advanced economies, not emerging market economies, indicating flight-to-quality behavior by global banks.
- Robustness:
  - Findings are robust to alternative measures of uncertainty, including economic policy uncertainty and idiosyncratic stock market volatility purged by the VIX.
  - Instrumenting uncertainty using exogenous events yields consistent results.
  - Various sample split exercises do not overturn the main conclusions.

*Source: wp1804 - appendix shows that these results still hold when we use the idiosyncratic uncertainty*

### References

### References

### Literature coverage
- Cites empirical and theoretical contributions on capital flows, cross-border banking, uncertainty, and financial crises, including works by Ahmed and Zlate; Avdjiev and Takáts; Baker, Bloom, and Davis; Bekaert, Hoerova, and Lo Duca; Bloom; Bruno and Shin; Cerutti et al.; Gourinchas and Rey; Ivashina, Scharfstein, and Stein; Rey; Shin; and many others as listed.

### Notes on figures
- Uncertainty measure: “Uncertainty is measured by stock market volatility in a reporter (source) country.”
- Figures present time series of cross-border claims and liabilities (billion of USD) and stock market volatility (right axis) for reporters including U.S., Germany, and Brazil, and bilateral reporter–counterparty pairs (U.S.—Germany, U.S.—Brazil, Germany—Brazil) across dates such as 1995q1, 1999q3, 2004q1, 2008q3, 2013q1 and 2002q3, 2005q1, 2007q3, 2010q1, 2012q3.

### Key descriptive statistics (Table 1)
- Growth of cross-border claims (q-o-q): Obs 30,608; Mean 3.136; Median 1.225; Standard deviation 40.751
- Growth of cross-border liabilities (q-o-q): Obs 29,889; Mean 2.998; Median 1.359; Standard deviation 50.930
- Stock market volatility: Obs 30,608; Mean 19.943; Median 17.211; Standard deviation 10.238
- Economic policy uncertainty: Obs 24,901; Mean 105.921; Median 97.412; Standard deviation 44.417
- Real GDP growth (q-o-q): Obs 30,608; Mean 0.600; Median 0.645; Standard deviation 1.059
- Growth of stock market (q-o-q): Obs 30,608; Mean 1.254; Median 2.214; Standard deviation 9.826
- Inflation rate (q-o-q): Obs 30,608; Mean 0.609; Median 0.573; Standard deviation 0.626
- Policy rate: Obs 30,608; Mean 3.666; Median 3.370; Standard deviation 2.653
- Growth of nominal exchange rate (q-o-q): Obs 30,608; Mean -0.127; Median 0.000; Standard deviation 4.243
- Growth of private credit (q-o-q): Obs 17,973; Mean 1.506; Median 1.372; Standard deviation 2.143
- External debt to GDP ratio: Obs 19,605; Mean 80.704; Median 74.052; Standard deviation 53.674
- Note: “Growth rates are calculated quarter-over-quarter. All variables are in percentage points.”

### Main empirical results — Baseline analysis (Table 2)
- Dependent variables: growth rate of exchange rate‑adjusted cross-border claims (outflows) and liabilities (inflows). All independent variables lagged by one period. Heteroskedasticity‑robust standard errors clustered at reporting–counterparty levels.
- Log of uncertainty:
  - Claims (columns I–III): -1.670** (0.824); -2.845** (1.150); -2.716** (1.211)
  - Liabilities (columns IV–VI): -2.369** (0.957); -2.734** (1.328); -2.302* (1.289)
- Real GDP growth:
  - Claims: 0.852*** (0.290); 0.782** (0.402); 0.391 (0.422)
  - Liabilities: 0.856** (0.393); 0.207 (0.629); -0.598 (0.604)
- Policy rate:
  - Claims: 0.557*** (0.107); 0.718*** (0.136); 0.669*** (0.133)
  - Liabilities: 0.147 (0.114); 0.244 (0.166); 0.062 (0.150)
- Nominal exchange rate growth:
  - Claims: -0.094 (0.078); -0.123 (0.088); -0.153* (0.089)
  - Liabilities: 0.003 (0.11); -0.093 (0.138); -0.217** (0.109)
- External debt to GDP (included in some specifications): -0.022*** (0.004); -0.018*** (0.006)
- Observations and fit:
  - Claims: Obs 30,608; 17,462; 16,431 across columns; R-squared 0.13; 0.14; 0.15
  - Liabilities: Obs 29,889; 16,725; 15,416 across columns; R-squared 0.14; 0.15; 0.16

### Emerging market economies sample (Table 3)
- Reporter and counterparty restricted to emerging market economies in separate panels.
- Log of uncertainty (claims):
  - Reporter EM only: -8.126* (4.114); -14.287* (7.727); -16.928** (7.821)
- Log of uncertainty (liabilities):
  - Counterparty EM only: -2.177* (1.223); -3.695** (1.792); -4.016** (1.797)
- Observations:
  - Claims: Obs 2,671; 1,400; 1,358 with R-squared 0.37; 0.41; 0.42
  - Liabilities: Obs 13,685; 7,694; 7,249 with R-squared 0.15; 0.17; 0.18

### Robustness: Alternative uncertainty measures
- Economic policy uncertainty (Table 4):
  - Log of uncertainty:
    - Claims: -1.923** (0.930); -2.664** (1.211); -3.574*** (1.239)
    - Liabilities: -2.621** (1.207); -3.654** (1.720); -2.832* (1.681)
  - External debt to GDP: -0.022*** (0.005); -0.028*** (0.007)
  - Observations: Claims Obs 21,564; 14,623; 13,715. Liabilities Obs 21,212; 14,009; 13,369. R-squared range 0.15–0.18.
- Idiosyncratic stock market volatility (Table 5):
  - Log of uncertainty:
    - Claims: -0.069* (0.039); -0.119** (0.050); -0.125** (0.055)
    - Liabilities: -0.100** (0.047); -0.117* (0.068); -0.106* (0.062)
  - External debt to GDP: -0.023*** (0.005); -0.021*** (0.006)
  - Observations: Claims Obs 30,608; 17,462; 16,431. Liabilities Obs 29,889; 16,725; 15,416. R-squared range 0.13–0.16.

### Robustness: Subperiods and country groups
- Before vs. after the Global Financial Crisis (Table 6):
  - Log of uncertainty:
    - Before GFC (1995Q1–2007Q2) claims: -1.983** (1.026)
    - After GFC (2007Q3–2012Q4) claims: -2.841** (1.289)
    - Before GFC liabilities: -2.111** (1.052)
    - After GFC liabilities: -3.550* (1.910)
  - Observations: Claims Obs 18,846; 11,578. Liabilities Obs 18,808; 11,081. R-squared range 0.119–0.16.
- Euro area vs. non‑euro area (Table 7):
  - Log of uncertainty (claims):
    - Euro area: -5.136** (2.444)
    - Non-euro area: -2.827* (1.511)
  - Observations: Euro area Obs 6,559; Non-euro area Obs 9,508. R-squared 0.32 and 0.22 (claims).

### Rebalancing between local and cross-border claims
- Baseline (Table 8):
  - Dependent variable: ratio of exchange rate‑adjusted cross‑border claims to sum of cross‑border claims and local claims in foreign currencies.
  - Log of uncertainty: 0.283* (0.166); 0.465** (0.191); 0.359* (0.185)
  - Real GDP growth: -0.084*** (0.020); -0.060* (0.031); -0.012 (0.025)
  - External debt to GDP: 0.004*** (0.001) in one specification
  - Observations: 24,420; 14,690; 14,101. R-squared 0.51; 0.49; 0.50.
- Before vs. after GFC (Table 9):
  - Log of uncertainty:
    - Before GFC: 0.237 (0.199); 0.319* (0.191); 0.268 (0.180)
    - After GFC: 0.447** (0.182); 0.571** (0.241); 0.465* (0.237)
  - Observations: Before GFC Obs 13,594; 6,092; 5,503. After GFC Obs 10,826; 8,598; 8,598. R-squared range 0.48–0.53.
- IV approach (Table 10):
  - Log of uncertainty: 1.513* (0.884); 1.427* (0.827); 1.735* (0.980)
  - Policy rate: -0.071*** (0.024); -0.080** (0.032); -0.081* (0.047)
  - Nominal exchange rate growth: -0.006** (0.003); -0.007* (0.004); -0.007* (0.004)
  - Weak identification tests: Cragg-Donald Wald F-statistic 207.18; 321.86; 256.47. Stock-Yogo 5% critical value 16.38 in each column.
  - Observations: 24,420; 14,690; 14,101. R-squared 0.48; 0.48; 0.50.
- Safe vs. risky borrowers (Table 11):
  - Log of uncertainty: 0.497* (0.262); 0.852*** (0.322); 0.734** (0.299)
  - Interaction (Log of uncertainty X counterparty EM dummy): -0.427* (0.256); -0.909*** (0.345); -0.901*** (0.336)
  - External debt to GDP: 0.004*** (0.001)
  - Observations: 24,420; 14,690; 14,101. R-squared 0.50; 0.49; 0.49.

*Source: wp1804 - References (IMF Working Paper content unit).*

### Appendix A. Additional Figures and Tables

### Appendix A. Additional Figures and Tables (wp1804 - Appendix A. Additional Figures and Tables)

### Country-specific uncertainty index (Figure A.1)
- Countries shown: Australia, Austria, Belgium, Brazil, Canada, Chile, Denmark, Finland, France, Germany, Greece, India, Indonesia, Italy, Japan, Korea, Mexico, Netherlands, Portugal, South Africa, Spain, Sweden, Taiwan Province of China, United Kingdom, United States.
- Two measures depicted: Stock market volatility; Economic policy uncertainty.
- Time markers on panels: 1995q1, 2002q3, 2010q1.

### Event classification (Table A.1)
- An increase: Gross inflows = Surge; Gross outflows = Flight.
- A decrease: Gross inflows = Stops; Gross outflows = Retrenchment.

### List of countries in the final sample (Table A.2)
- Source countries (with = 1 if advanced economy where listed): Australia 1, Austria 1, Belgium 1, Brazil 0, Canada 1, Chile 0, Denmark 1, Finland 1, France 1, Germany 1, Greece 1, India 0, Indonesia 0, Italy 1, Japan 1, Korea 1, Mexico 0, Netherlands 1, Portugal 1, South Africa 0, Spain 1, Sweden 1, Taiwan 1, United Kingdom 1, United States 1, Lithuania 0, Malaysia 0, Mexico 0, Netherlands 1, New Zealand 1, Norway 1, Pakistan 0, Peru 0, Philippines 0, Poland 0, Portugal 1, Romania 0, Russia 0, Slovak Republic 1, Slovenia 1, South Africa 0, Spain 1, Sweden 1, Taiwan 1, Thailand 0, Turkey 0, Ukraine 0, United Kingdom 1, United States 1, Venezuela 0.
- Recipient countries (with = 1 if advanced economy where listed): Argentina 0, Australia 1, Austria 1, Belgium 1, Brazil 0, Bulgaria 0, Canada 1, Chile 0, China 0, Colombia 0, Czech Republic 1, Denmark 1, Estonia 1, Finland 1, France 1, Germany 1, Greece 1, Hungary 0, India 0, Indonesia 0, Israel 1, Italy 1, Japan 1, Korea 1, Latvia 0, Lithuania 0, Malaysia 0, Mexico 0, Netherlands 1, New Zealand 1, Norway 1, Pakistan 0, Peru 0, Philippines 0, Poland 0, Portugal 1, Romania 0, Russia 0, Slovak Republic 1, Slovenia 1, South Africa 0, Spain 1, Sweden 1, Taiwan 1, Thailand 0, Turkey 0, Ukraine 0, United Kingdom 1, United States 1, Venezuela 0.

### Before and after the Global Financial Crisis: Economic policy uncertainty (Table A.3)
- Dependent variables: growth rate of exchange rate-adjusted cross-border claims (columns I–II) and growth rate of exchange rate-adjusted cross-border liabilities (columns III–IV).
- Sample periods:
  - Before the GFC (1995Q1-2007Q2)
  - After the GFC (2007Q3-2012Q4)
- Key coefficients and statistics:
  - Log of uncertainty: (I) -2.841** (1.289); (II) -2.616* (1.521); (III) -0.827 (1.348); (IV) -6.026*** (2.225)
  - Real GDP growth: (I) 0.909*** (0.297); (II) 1.080** (0.473); (III) 0.593 (0.552); (IV) 0.820 (0.647)
  - Stock market growth: (I) 0.008 (0.033); (II) 0.062 (0.045); (III) -0.142** (0.068); (IV) 0.187*** (0.072)
  - CPI inflation: (I) -0.183 (0.536); (II) 0.078 (0.694); (III) -0.202 (1.056); (IV) 0.311 (0.748)
  - Policy rate: (I) 0.177 (0.179); (II) 0.438** (0.214); (III) 0.213 (0.141); (IV) 0.057 (0.226)
  - Nominal exchange rate growth: (I) 0.003 (0.079); (II) -0.063 (0.105); (III) 0.259 (0.167); (IV) 0.057 (0.147)
  - Obs: (I) 11,578; (II) 9,559; (III) 11,126; (IV) 8,998
  - R-squared: (I) 0.11; (II) 0.13; (III) 0.12; (IV) 0.14
- Note: All independent variables are lagged by one period; heteroskedasticity-robust standard errors clustered at reporting-counterparty country levels. Significance: *** 1%, ** 5%, * 10%.

### Rebalancing between local and cross-border claims: Alternative measure of uncertainty (idiosyncratic stock market volatility) (Table A.4)
- Dependent variable: ratio of exchange rate-adjusted cross-border claims to the sum of exchange rate-adjusted cross-border claims and local claims in foreign currencies.
- Key coefficients and statistics:
  - Log of uncertainty: (I) 0.012* (0.007); (II) 0.018** (0.008); (III) 0.015** (0.007)
  - Real GDP growth: (I) -0.083*** (0.020); (II) -0.060* (0.031); (III) -0.009 (0.025)
  - Stock market growth: (I) -0.002 (0.001); (II) -0.001 (0.002); (III) -0.001 (0.002)
  - CPI inflation: (I) -0.065* (0.039); (II) -0.135** (0.059); (III) -0.099 (0.062)
  - Policy rate: (I) -0.045** (0.020); (II) -0.043* (0.024); (III) -0.034 (0.026)
  - Nominal exchange rate growth: (I) -0.008** (0.004); (II) -0.010** (0.005); (III) -0.007 (0.005)
  - Private credit growth: (I) -0.029 (0.020); (II) 0.001 (0.019)
  - External debt to GDP: (III) 0.004*** (0.001)
  - Obs: (I) 24,420; (II) 14,690; (III) 14,101
  - R-squared: (I) 0.51; (II) 0.50; (III) 0.50
- Note: All independent variables are lagged by one period; heteroskedasticity-robust standard errors clustered at reporting-counterparty country levels. Significance: *** 1%, ** 5%, * 10%.

### Rebalancing between local and cross-border claims: IV approach with alternative measures of uncertainty (Table A.5)
- Dependent variable: ratio of exchange rate-adjusted cross-border claims to the sum of exchange rate-adjusted cross-border claims and local claims in foreign currencies.
- Columns (I)–(III): Economic policy uncertainty; (IV)–(VI): Idiosyncratic stock market volatility.
- Key coefficients and statistics:
  - Log of uncertainty:
    - (I) 1.194* (0.683)
    - (II) 1.499* (0.798)
    - (III) 1.392* (0.817)
    - (IV) 0.102* (0.060)
    - (V) 0.098* (0.057)
    - (VI) 0.124* (0.067)
  - Real GDP growth:
    - (I) -0.103*** (0.024)
    - (II) -0.160*** (0.048)
    - (III) -0.085** (0.039)
    - (IV) -0.073*** (0.018)
    - (V) 0.032 (0.058)
    - (VI) 0.069 (0.061)
  - Stock market growth: (I) -0.002 (0.002); (II) 0.002 (0.003); (III) 0.002 (0.003); (IV) 0.009 (0.007); (V) 0.007 (0.006); (VI) 0.009 (0.007)
  - CPI inflation: (I) -0.083** (0.036); (II) -0.146** (0.072); (III) -0.125 (0.079); (IV) -0.099* (0.052); (V) 0.004 (0.079); (VI) 0.033 (0.095)
  - Policy rate: (I) -0.046** (0.020); (II) -0.050** (0.024); (III) -0.044* (0.025); (IV) -0.072*** (0.025); (V) -0.090** (0.037); (VI) -0.098* (0.055)
  - Nominal exchange rate growth: (I) -0.009** (0.004); (II) -0.015** (0.006); (III) -0.012* (0.006); (IV) -0.009** (0.004); (V) -0.008* (0.005); (VI) -0.011* (0.006)
  - Private credit growth: (II) -0.003 (0.023); (III) 0.024 (0.021); (IV) -0.021* (0.019); (V) -0.009 (0.020)
  - External debt to GDP: (II) 0.004*** (0.001); (V) 0.003 (0.002)
  - Cragg-Donald Wald F-statistic: (I) 518.58; (II) 459.28; (III) 425.07; (IV) 105.84; (V) 150.77; (VI) 112.59
  - Stock-Yogo weak identification test 5% critical values: 16.38 for all columns
  - Obs: (I) 19,617; (II) 13,271; (III) 13,057; (IV) 24,420; (V) 14,690; (VI) 14,101
  - R-squared: (I) 0.50; (II) 0.48; (III) 0.49; (IV) 0.50; (V) 0.50; (VI) 0.51
- Note: All independent variables are lagged by one period; heteroskedasticity-robust standard errors clustered at reporting-counterparty country levels. Significance: *** 1%, ** 5%, * 10%.

*Appendix B. Additional exercise on the rebalancing channel of global banks (summary of methods and results)*

### Motivation and alternative share measure
- BIS LBS provides local claims by global banks in local currencies only after 2012; to maintain consistency the main analysis used local claims in foreign currencies only when constructing s_{i,j,t} in equation (3).
- Complementary analysis constructs a new variable measuring the share of cross-border claims of global banks to total domestic claims of the banking system (using Domestic bank claims, line 32, from IMF IFS Depository Corporations Survey).
- Cross-border claims converted into local currencies using nominal exchange rate i,t before computing:
  - s̃_{i,j,t} = (cross-border claims_{i,j,t} × nominal exchange rate_{i,t}) / total domestic claims_{i,t} × 100.     (6)
- Baseline regression specification:
  - s̃_{i,j,t} = α_{j,t} + β X_{i,t−1} + γ UNC_{i,t−1} + ε_{i,j,t}.          (7)
- For most countries, total domestic claims quarterly available from IFS since 2001; baseline analysis repeated using data since 2001 with main findings hardly changing.

### Rebalancing between domestic and cross-border lending: alternative share results (Table B.1)
- Dependent variable: ratio of exchange rate-adjusted cross-border claims of global banks to the claims of the domestic banking system.
- Columns and approaches:
  - (I) Stock market volatility — OLS
  - (II) Stock market volatility — IV
  - (III) Stock market volatility — OLS interaction
  - (IV) Economic policy uncertainty — OLS
  - (V) Economic policy uncertainty — IV
  - (VI) Economic policy uncertainty — OLS interaction
- Key coefficients and statistics:
  - Log of uncertainty (stock market volatility): (I) 3.012*** (0.670); (II) 9.125*** (1.833); (III) 4.979*** (1.108)
  - Interaction term Log of uncertainty × counterparty EM dummy: -4.830*** (1.247) in stock market volatility IV/interaction columns and -2.149*** (0.808) in economic policy uncertainty columns (placement noted in table)
  - Log of uncertainty (economic policy uncertainty): (IV) 2.201*** (0.604); (V) 2.341** (1.036); (VI) 3.082*** (0.909)
  - Real GDP growth: (I) -0.207* (0.121); (II) -0.248** (0.119); (III) -0.225* (0.120); (IV) -0.675*** (0.160); (V) -0.449*** (0.116); (VI) -0.675*** (0.159)
  - Stock market growth: (I) 0.019*** (0.006); (II) 0.010** (0.005); (III) 0.018*** (0.006); (IV) 0.016*** (0.006); (V) 0.003)0.016*** (table shows a formatting issue); (VI) 0.016*** (0.006)
  - CPI inflation: (I) -2.975*** (0.573); (II) -2.845*** (0.555); (III) -2.953*** (0.567); (IV) -3.082*** (0.730); (V) -3.067*** (0.765); (VI) -3.058*** (0.723)
  - Policy rate: (I) -0.482*** (0.112); (II) -0.688*** (0.139); (III) -0.516*** (0.120); (IV) -0.256*** (0.082); (V) -0.309*** (0.096); (VI) -0.266*** (0.085)
  - Nominal exchange rate growth: (I) -0.150*** (0.031); (II) -0.131*** (0.027); (III) -0.151*** (0.031); (IV) -0.162*** (0.037); (V) -0.138*** (0.032); (VI) -0.162*** (0.037)
  - Private credit growth: (I) -0.085 (0.059); (II) -0.065 (0.058); (III) -0.09 (0.059); (IV) -0.185*** (0.048); (V) -0.136*** (0.037); (VI) -0.185*** (0.048)
  - External debt to GDP: (I) -0.018*** (0.005); (II) -0.023*** (0.005); (III) -0.019*** (0.005); (IV) -0.016*** (0.005); (V) -0.013*** (0.004); (VI) -0.016*** (0.005)
  - Obs: (I) 17,029; (II) 17,029; (III) 17,029; (IV) 14,213; (V) 14,213; (VI) 14,213
  - R-squared: (I) 0.20; (II) 0.22; (III) 0.20; (IV) 0.23; (V) 0.19; (VI) 0.23
- Note: All independent variables are lagged by one period; heteroskedasticity-robust standard errors clustered at reporting-counterparty country levels. Significance: *** 1%, ** 5%, * 10%.

*Appendix A and Appendix B tables and figures as presented in the source PDF.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp1804.pdf_