## wp18217

## Source details

**Canonical URL:** [wp18217](https://www.imf.org/-/media/files/publications/wp/2018/wp18217.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2018/wp18217.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2018/wp18217.pdf.json)

---

### I. Introduction — motivation and data scope
- Monetary policy alone is insufficient to achieve financial stability and/or avoid undesirable consequences linked to surges in capital inflows; Macroprudential Policy Measures (MPM) and Capital Control Measures (CCM) are increasingly used.
- Global usage:
  - MPM and CCM usage increased especially after the Global Financial Crisis of 2008 (GFC).
  - All advanced economies (AEs) now have at least one MPM; EMDEs use more CCM than AEs.
- Research gap and objective:
  - Most studies analyze CCM or MPM separately; few consider simultaneous effects and policy interactions, and fewer analyze cross-border banking flows specifically.
  - Objective: analyze together the potential impact of CCM and MPM on worldwide cross-border banking flows, distinguishing direct cross-border lending and lending through local affiliates using BIS Consolidated Banking Statistics (CBS).
- Key datasets and coverage:
  - Cerutti, Claessens and Laeven (2017) dataset: 12 macroprudential measures for 160 countries during 2000-17.
  - Fernandez et al (2015) CCM dataset: 100 countries from 1995 to 2015 (extended to lenders in sample).
  - Barth, Caprio and Levine (2013) dataset: proxy for bank supervision and restrictions on non-core bank activities.
  - BIS Consolidated Banking Statistics (CBS): bilateral consolidated cross-border banking flows.
- Final empirical sample for main estimations: 29 BIS reporting lenders and 86 borrowers, 2006 to 2015.

### II. Empirical strategy and identification
- Framework:
  - Two-step structural gravity framework building on Helpman, Melitz and Rubinstein (2008) and Cerutti and Zhou (2018), correcting for selection bias from zero bilateral banking flows.
- First stage (Probit):
  - Estimates probability of new positive net lending connection T_ijt based on lender and borrower characteristics (r_it, r_jt), time-invariant bilateral gravity factors (d_ij), and first-stage instruments ζ_ijt (fixed-cost shifters).
- Second stage (non-linear maximum likelihood):
  - Estimates quantity Y_ijt of net new positive flows conditional on selection, includes inverse Mills ratio η_ijt and nonlinear term ln{exp[δ(z_ijt+η_ijt)]−1}.
  - Dependent variable: Y_ijt = max(0, X_ijt − X_ij(t−1)); T_ijt = I(Y_ijt > 0).
- Treatment of zeroes and selection:
  - Accounts for zeros driven by unobserved bank productivity heterogeneity; estimator exploits truncated Pareto assumption for bank productivity (Appendix A1).
- Aggregation and frequency:
  - Quarterly BIS data collapsed to annual frequency using end-of-year observations; flows computed as year-on-year differences.
- Policy variables and instruments:
  - Overall macroprudential index (MPI) based on 12 MPM types.
  - Individual MPM: leverage ratio requirement, loan-to-value (LTV) limit, interbank exposure limit, foreign currency loan limit.
  - CCM: overall capital outflow (inflow) restriction indices averaging ten binary transaction-type dummies; individual CCM: restrictions on bond investment, commercial credit, foreign credit.
  - Bank activity restriction from Barth, Caprio and Levine (2013) as proxy for microprudential intensity.
- Timing and controls:
  - Policy variables lagged by one year (r_kt−1).
  - Controls: lender and borrower real GDP growth, nominal credit-to-GDP, policy-related interest rate/discount rate, Fraser Institute property right index, log geographical distance, common official language dummy.
- First-stage instruments for selection:
  - Synthetic indicator of high banks’ overhead cost to total assets (value one if both lender and borrower above median).
  - For local affiliate lending, additional indicators: free-trade agreement; high costs and long time to set up firms.

### III. Main findings from two-stage regressions and counterfactuals
- Gravity controls and selection:
  - Distance elasticity: around -0.5 for direct cross-border lending and -1.5 for local affiliate lending.
  - Lenders’ and borrowers’ domestic credit to GDP and common language positively associated with both probability and quantity of net new lending.
  - Borrower GDP growth positively associated with new direct cross-border and local banking linkages.
  - Institutional quality for borrowers positively correlated with higher direct cross-border flows in both stages.
  - First-stage instruments (high overhead cost and free-trade agreement) enter with statistical significance; productivity heterogeneity structural parameter highly significant.
- Aggregate MPM effects (controlling for CCM and micro-prudential restrictions):
  - Lenders’ overall MPM usage:
    - Increases probability of forming a new direct cross-border banking connection (first stage).
    - Decreases quantity of direct cross-border banking flows (second stage; negative elasticity).
    - Associated with higher lending through banks’ local affiliates (strong positive association).
  - Borrowers’ overall MPM usage:
    - Positive and statistically significant impact on direct cross-border banking inflows (second stage).
    - Expected negative but statistically insignificant effect on lending through local affiliates.
  - Robustness: results hold after adding domestic credit booms and regional cross-border general inflows.
- Aggregate CCM effects:
  - Lenders’ capital outflow restrictions strongly associated with higher local affiliate lending, particularly when restrictions target outward bond investments.
  - Borrowers’ CCM on inflows lead to higher borrowing through local affiliates.
  - Impact of CCM on direct cross-border lending is sometimes statistically significant (e.g., lenders’ bond outflows restrictions increase direct cross-border lending to EMDEs) but usually small in magnitude.
- Counterfactual exercise (removing all MPM or CCM):
  - MPM counterfactuals: lenders’ MPM reduce direct cross-border lending and raise local affiliate lending in counterfactual predictions; borrowers’ MPM associated with higher direct cross-border inflows in counterfactuals.
  - CCM counterfactuals: show spillovers from CCM on local affiliate lending; overall quantitative impact of CCM on direct cross-border lending is small.

### IV. Individual instruments, heterogeneity, and circumvention
- Macroprudential instruments (selected):
  - Lenders’ leverage ratio requirement is a key driver of the negative effect on direct cross-border lending.
  - Borrowers’ individual MPM (leverage ratio, interbank exposure limits, foreign currency loan limits) associated with higher direct cross-border inflows — consistent with circumvention motives.
- Tests for alternative explanations:
  - Adding change of domestic credit to GDP and regional weighted inflows does not nullify the positive link between borrowers’ MPM and direct cross-border flows, supporting circumvention interpretation.
- Heterogeneity by borrower type:
  - Lenders’ overall MPM positive effect on local affiliate flows primarily associated with AE borrowers.
  - Lenders’ MPM produce significantly negative effects on direct cross-border lending to EMDE borrowers across overall usage and individual instruments.
- CCM instrument heterogeneity:
  - Lenders’ outflow restrictions, especially bond outflow restrictions, partly substitute toward higher local affiliate lending.
  - Borrowers’ restrictions on lending by nonresidents associated with higher inflows through local affiliates.
  - Positive coefficients for some credit outflow restrictions on direct cross-border lending may reflect that some credit controls target non-bank institutions rather than banks.

### V. Policy interactions: do CCM mitigate or amplify MPM spillovers?
- Interaction analysis (Figure 4):
  - Interaction effects are heterogeneous across instruments and not uniformly offsetting.
  - Example: lenders’ leverage ratio negative spillovers increase in the simultaneous presence of overall capital outflow restrictions (CCM).
  - Little consistent evidence that borrowers’ CCM mitigate circumvention of borrowers’ MPM; only borrowers’ interbank exposure limits show some evidence of offsetting circumvention in the sample.
  - Overall: borrowers’ CCM do not appear large or uniformly effective at preventing circumvention of borrowers’ MPM; borrowers’ CCM may be non-binding or their effects captured by other controls.

### VI. Robustness checks
- Outlier control:
  - Truncating year-to-year growth rates above the 95th percentile yields similar second-stage results for both MPM and CCM (Table 10).
- Post-GFC subsample (2011-2015):
  - Results for MPM similar to baseline (2006-2015); lenders’ MPM continue to show negative spillovers on direct cross-border flows and positive associations with local affiliate lending.
  - Borrowers’ MPM continue to trigger increased direct cross-border flows, with statistical significance for overall index, leverage ratio, foreign currency loan limits, and LTVs (for the first time in this subsample).
- Alternative dependent variable definitions and estimators:
  - Alternative definitions (dropping negative flows; using log differences) and estimators (OLS, PPML) display broadly similar findings for MPM; PPML results differ in some cases but may be sensitive to data-generating process and do not correct selection bias from structural zeroes.
- Conclusion on robustness: baseline two-stage framework designed to address selection due to many zero flows; robustness analyses generally support core conclusions, especially on MPM effects.

### VII. Conclusions and policy implications
- Empirical conclusions:
  - Substantial policy spillovers exist from MPM and CCM on cross-border banking flows; MPM spillovers are generally more prevalent and larger than CCM spillovers.
  - Lenders’ MPM:
    - Reduce direct cross-border lending.
    - Increase local affiliate lending (circumvention via affiliates), especially toward AEs.
    - Reduce direct cross-border lending to EMDE borrowers.
  - Borrowers’ MPM:
    - Often trigger circumvention reflected in increased direct cross-border inflows.
    - Do not see consistent offsetting effects from borrowers’ CCM in aggregate.
  - CCM:
    - More associated with substitution toward local affiliate lending (especially when outflow restrictions target bond investments).
    - Limited aggregate effect on direct cross-border lending.
- Policy implications:
  - Agents react to CCM and especially to MPM by adjusting international lending/borrowing decisions, generating international spillovers.
  - Need to consider banks’ foreign affiliates when assessing the cross-border effects of home-country MPM; risk-taking and connectedness assessments should include affiliate behavior.
  - Borrower-country policymakers should be aware that MPM implementation can trigger unintended cross-border spillovers and circumvention; borrower CCMs do not appear uniformly effective at preventing such circumvention.
  - Findings point toward potential benefits from cross-border coordination and cooperation between regulators, particularly when foreign affiliates and economically significant spillovers are involved, to reduce unintended spillovers and improve risk-sharing.

### VIII. Key statistics (selected exact values from Table 2 and core estimates)
- Summary statistics (selected):
  - Direct Cross-Border Connection: N 55970; Mean 0.23; SD 0.42; Min 0; Max 1; Country/Pair 5597
  - Local Affiliate Connection: N 55970; Mean 0.07; SD 0.25; Min 0; Max 1; Country/Pair 5597
  - Log Direct Cross-Border Flow: N 12772; Mean -2.92; SD 2.77; Min -13.81; Max 6.55; Country/Pair 3216
  - Log Local Affiliate Flow: N 3763; Mean -3.55; SD 3.47; Min -19.1; Max 5.55; Country/Pair 1234
  - Log Distance: N 5597; Mean 8.7; SD 0.82; Min 4.18; Max 9.9; Country/Pair 5597
  - Overall Prudential Index: Lender: N 290; Mean 2.26; SD 1.57; Min 0; Max 7; Country/Pair 29
  - Overall Prudential Index: Borrower: N 1570; Mean 1.9; SD 1.6; Min 0; Max 9; Country/Pair 157
- First-stage (selected average marginal effects, Table 3):
  - Overall macroprudential index (Lender): 0.0225*** (0.00373) — Direct Cross-Border
  - Overall macroprudential index (Borrower): -0.0143*** (0.00390) — Direct Cross-Border
  - Overall outflow restriction (Lender): 0.0516* (0.0298) — Direct Cross-Border
  - Overall inflow restriction (Borrower): 0.0863*** (0.0306) — Local Affiliate
  - Real GDP Growth (Borrower): 0.00333*** (0.000744) — Direct Cross-Border
  - Credit to GDP (Borrower): 0.000581*** (0.000187) — Direct Cross-Border
  - Log Distance: -0.0541*** (0.00535) — Direct Cross-Border
  - Common Language: 0.0700*** (0.0116) — Direct Cross-Border
- Second-stage (selected structural estimates, Table 4/5/8):
  - Overall macroprudential index:
    - Lender: -0.119* (0.0702) — Direct Cross-Border
    - Borrower: 0.151*** (0.0463) — Direct Cross-Border
  - Leverage Ratio (Lender): -0.907*** (0.300) — Direct Cross-Border
  - Overall outflow restriction (Lender): -0.116 (0.324) — Direct Cross-Border; 2.532*** (0.834) — Local Affiliate
  - Bond (Lender Outflow): 0.262 (0.165) — Direct Cross-Border; 1.600*** (0.587) — Local Affiliate
  - Foreign Currency Loan (Borrower): 0.493*** (0.184) — Direct Cross-Border
- Heterogeneity (AE vs EMDE, Table 7 & 9):
  - AE borrower (Direct Cross-Border): Overall Borrower: 0.141** (0.0573); Leverage Ratio (Lender): -0.800** (0.353)
  - EMDE borrower (Direct Cross-Border): Overall Lender: -0.377*** (0.113); Leverage Ratio (Lender): -1.309*** (0.427)
- Robustness (selected):
  - Right-censoring (Table 10): Macroprudential Overall (Lender): -0.136* (0.0773) — Direct Cross-Border; 0.296*** (0.107) — Local Affiliate
  - 2011–2015 sample (Table 11): Macroprudential Overall (Lender): -0.214** (0.0917) — Direct Cross-Border; 0.453*** (0.151) — Local Affiliate; Leverage Ratio (Lender, 2011–2015): -1.584*** (0.429)

*Source: Appendix A.2. Additional Tables and Counterfactual Figures — wp18217 (sections I–VII, empirical framework, main results, robustness, and conclusions).*

### References .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .

### wp18217 - References .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .

### Figures
- Figure 1:Global usage of macroprudential instruments and capital controls    .  .  .  .  .  .  .  .  .    23
- Figure 2:Model prediction and counterfactual banking flows:  Overall macroprudential policy  .  .  .  .  .  .  .  .  .  .  .  .  .    24
- Figure 3:Model prediction and counterfactual banking flows:  Overall capital flow re-strictions    .  .  .  .  .  .  .  .  .  .  .  .  .  .    25
- Figure 4:Interaction between macroprudential policy, capital control and direct cross-border banking   .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    26

### Tables
- Table 1:Variable Definitions .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    27
- Table 2:Summary Statistics  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    28
- Table 3:Macroprudential policy, capital control and cross-border lending – first stage overall estimates   .  .  .  .  .  .  .  .  .  .  .  .    29
- Table 4:Macroprudential policy, capital control and cross-border lending – second stage overall estimates   .  .  .  .  .  .  .  .  .  .  .    30
- Table 5:Macroprudential policy and cross-border banking – second-stage specific estimates   .  .  .  .  .  .  .  .  .  .  .  .    31
- Table 6:Borrowers’ macroprudential policy and cross-border banking: Interaction with credit cycles .  .  .  .  .  .  .  .  .  .  .  .  .    32
- Table 7:Macroprudential policy and cross-border banking – second-stage specific estimates, AE/EMDE breakdown    .  .  .  .  .  .  .  .    33
- Table 8:Capital control and cross-border banking – second-stage specific estimates .  .  .  .  .  .  .  .    34
- Table 9:Capital control measures and cross-border banking – second-stage specific estimates, AE/EM breakdown    .  .  .  .  .  .  .    35
- Table 10:MPM, CCM and cross-border banking: robustness to right censoring .  .  .  .  .  .  .  .  .    36
- Table 11:MPM, CCM and cross-border banking: 2011-2015 sample estimates  .  .  .  .  .  .  .  .  .    37

### Appendix
- Appendix A.1. Micro-found the empirical framework and counterfactual analysis   .  .  .  .  .  .  .    38

*https://www.imf.org/-/media/files/publications/wp/2018/wp18217.pdf*

### Appendix A.2. Additional Tables and Counterfactual Figures  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .  .    41

### Appendix A.2. Additional Tables and Counterfactual Figures

### I. Introduction — motivation and data scope
- Monetary policy alone is insufficient to achieve financial stability and/or avoid undesirable consequences linked to surges in capital inflows; Macroprudential Policy Measures (MPM) and Capital Control Measures (CCM) are increasingly used.
- Global usage: MPM and CCM usage increased especially after the Global Financial Crisis of 2008 (GFC); all advanced economies (AEs) now have at least one MPM; EMDEs use more CCM than AEs.
- Research gap: most studies analyze CCM or MPM separately; few consider simultaneous effects and policy interactions, and fewer analyze cross-border banking flows specifically.
- Objective: analyze together the potential impact of CCM and MPM on worldwide cross-border banking flows, distinguishing direct cross-border lending and lending through local affiliates using BIS Consolidated Banking Statistics (CBS).
- Key datasets and coverage:
  - Cerutti, Claessens and Laeven (2017) dataset capturing 12 macroprudential measures for 160 countries during 2000-17.
  - Fernandez et al (2015) CCM dataset covering 100 countries from 1995 to 2015 (extended to cover lenders in sample).
  - Barth, Caprio and Levine (2013) dataset to proxy bank supervision and restrictions on non-core bank activities.
  - BIS Consolidated Banking Statistics (CBS) for bilateral consolidated cross-border banking flows.
- Final empirical sample used in main estimations: 29 BIS reporting lenders and 86 borrowers, 2006 to 2015.

### II. Empirical strategy and identification
- Method: two-step structural gravity framework building on Helpman, Melitz and Rubinstein (2008) and Cerutti and Zhou (2018), correcting for selection bias from zero bilateral banking flows.
- First stage (Probit): estimates probability of new positive net lending connection T_ijt based on lender and borrower characteristics (r_it, r_jt), time-invariant bilateral gravity factors (d_ij), and exclusive first-stage instruments ζ_ijt (fixed-cost shifters).
- Second stage (non-linear maximum likelihood): estimates quantity Y_ijt of net new positive flows conditional on selection, including an inverse Mills ratio η_ijt and non-linear term ln{exp[δ(z_ijt+η_ijt)]−1}.
- Dependent variables defined from annualized flows: Y_ijt = max(0, X_ijt − X_ij(t−1)); T_ijt = I(Y_ijt > 0).
- Treatment of zeroes and selection: framework accounts for zeros driven by unobserved bank productivity heterogeneity; estimator exploits truncated Pareto assumption for bank productivity (see Appendix A1).
- Aggregation and frequency: quarterly BIS data collapsed to annual frequency using end-of-year observations; flows computed as year-on-year differences.
- Policy variables and instruments:
  - Overall macroprudential index (MPI) based on 12 MPM types.
  - Individual MPM: leverage ratio requirement, loan-to-value (LTV) limit, interbank exposure limit, foreign currency loan limit.
  - CCM: overall capital outflow (inflow) restriction indices averaging ten binary transaction-type dummies; individual CCM focus on restrictions on bond investment, commercial credit, foreign credit.
  - Bank activity restriction from Barth, Caprio and Levine (2013) as proxy for microprudential intensity.
- Timing: policy variables lagged by one year to address simultaneity (r_kt−1).
- Controls: lender and borrower real GDP growth, nominal credit-to-GDP, policy-related interest rate/discount rate, Fraser Institute property right index, log geographical distance, common official language dummy.
- First-stage instruments for selection:
  - Synthetic indicator of high banks’ overhead cost to total assets (value one if both lender and borrower above median).
  - For local affiliate lending, additional indicators: free-trade agreement; high costs and long time to set up firms.

### III. Main findings from two-stage regressions and counterfactuals
- Gravity controls:
  - Distance elasticity: around -0.5 for direct cross-border lending and -1.5 for local affiliate lending.
  - Lenders’ and borrowers’ domestic credit to GDP and common language positively associated with both probability and quantity of net new lending.
  - Borrower GDP growth positively associated with new direct cross-border and local banking linkages.
  - Institutional quality for borrowers positively correlated with higher direct cross-border flows in both stages.
- Selection and identification:
  - High overhead cost and free-trade agreement indicators enter first stage with statistical significance; structural parameter for productivity heterogeneity is highly significant.
- Aggregate MPM effects (controlling for CCM and micro-prudential restrictions):
  - Lenders’ overall MPM usage:
    - Increases probability of forming a new direct cross-border banking connection (first stage).
    - Decreases quantity of direct cross-border banking flows (second stage; negative elasticity).
    - Associated with higher lending through banks’ local affiliates (strong positive association).
  - Borrowers’ overall MPM usage:
    - Positive and statistically significant impact on direct cross-border banking inflows (second stage).
    - Yields an expected negative but statistically insignificant effect on lending through local affiliates.
  - Results robust to adding domestic credit booms and regional cross-border general inflows.
- Aggregate CCM effects:
  - Lenders’ capital outflow restrictions strongly associated with higher local affiliate lending, particularly when restrictions target outward bond investments.
  - Borrowers’ CCM on inflows lead to higher borrowing through local affiliates.
  - Impact of CCM on direct cross-border lending is sometimes statistically significant (e.g., lenders’ bond outflows restrictions increase direct cross-border lending to EMDEs) but usually small in magnitude.
- Counterfactual exercise (removing all MPM or CCM):
  - For MPM: counterfactuals suggest negative spillovers from lenders’ MPM on direct cross-border lending; lenders’ MPM raises local affiliate lending in counterfactual predictions; borrowers’ MPM associated with higher direct cross-border inflows in counterfactuals.
  - For CCM: counterfactuals show spillovers from CCM on local affiliate lending; overall quantitative impact of CCM on direct cross-border lending is small.

### IV. Individual instruments, heterogeneity, and circumvention
- Macroprudential instruments (Table 5 summaries):
  - Lenders’ leverage ratio requirement is a key driver of the negative effect on direct cross-border lending.
  - Borrowers’ individual MPM (leverage ratio, interbank exposure limits, foreign currency loan limits) associated with higher direct cross-border inflows — consistent with circumvention motives.
- Tests for alternative explanations (Table 6 augmentations):
  - Adding change of domestic credit to GDP and regional weighted inflows does not nullify the positive link between borrowers’ MPM and direct cross-border flows, supporting circumvention interpretation.
- Heterogeneity by borrower type (Table 7):
  - Lenders’ overall MPM positive effect on local affiliate flows primarily associated with AE borrowers.
  - Lenders’ MPM produce significantly negative effects on direct cross-border lending to EMDE borrowers across overall usage and individual instruments.
- CCM instrument heterogeneity (Table 8 summaries):
  - Lenders’ outflow restrictions, especially bond outflow restrictions, partly substitute toward higher local affiliate lending.
  - Borrowers’ restrictions on lending by nonresidents associated with higher inflows through local affiliates.
  - Positive coefficients for some credit outflow restrictions on direct cross-border lending may reflect that some credit controls target non-bank institutions rather than banks.

### V. Policy interactions: do CCM mitigate or amplify MPM spillovers?
- Interaction analysis (Figure 4):
  - Interaction effects are heterogeneous across instruments and not uniformly offsetting.
  - Example: lenders’ leverage ratio negative spillovers increase in the simultaneous presence of overall capital outflow restrictions (CCM).
  - Little consistent evidence that borrowers’ CCM mitigate circumvention of borrowers’ MPM; only borrowers’ interbank exposure limits show some evidence of offsetting circumvention in the sample.
  - Overall: borrowers’ CCM do not appear large or uniformly effective at preventing circumvention of borrowers’ MPM; borrowers’ CCM may be non-binding or their effects captured by other controls.

### VI. Robustness checks
- Outlier control (Table 10): truncating year-to-year growth rates above the 95th percentile yields similar second-stage results for both MPM and CCM.
- Post-GFC subsample (2011-2015) (Table 11):
  - Results for MPM similar to baseline (2006-2015); lenders’ MPM continue to show negative spillovers on direct cross-border flows and positive associations with local affiliate lending.
  - Borrowers’ MPM continue to trigger increased direct cross-border flows, with statistical significance for overall index, leverage ratio, foreign currency loan limits, and LTVs (for the first time in this subsample).
- Alternative dependent variable definitions and estimators (Appendix A2, Table A2(d)):
  - Alternative definitions (dropping negative flows; using log differences) and estimators (OLS, PPML) display broadly similar findings for MPM; PPML results differ in some cases but may be sensitive to data-generating process and do not correct selection bias from structural zeroes.
- Conclusion on robustness: baseline two-stage framework designed to address selection due to many zero flows; robustness analyses generally support core conclusions, especially on MPM effects.

### VII. Conclusions and policy implications
- Summary of empirical conclusions:
  - Substantial policy spillovers exist from MPM and CCM on cross-border banking flows; MPM spillovers are generally more prevalent and larger than CCM spillovers.
  - Lenders’ MPM:
    - Reduce direct cross-border lending.
    - Increase local affiliate lending (circumvention via affiliates), especially toward AEs.
    - Reduce direct cross-border lending to EMDE borrowers.
  - Borrowers’ MPM:
    - Often trigger circumvention reflected in increased direct cross-border inflows.
    - Do not see consistent offsetting effects from borrowers’ CCM in aggregate.
  - CCM:
    - More associated with substitution toward local affiliate lending (especially when outflow restrictions target bond investments).
    - Limited aggregate effect on direct cross-border lending.
- Policy implications:
  - Agents react to CCM and especially to MPM by adjusting international lending/borrowing decisions, generating international spillovers.
  - Need to consider banks’ foreign affiliates when assessing the cross-border effects of home-country MPM; risk-taking and connectedness assessments should include affiliate behavior.
  - Borrower-country policymakers should be aware that MPM implementation can trigger unintended cross-border spillovers and circumvention; borrower CCMs do not appear uniformly effective at preventing such circumvention.
  - Findings point toward potential benefits from cross-border coordination and cooperation between regulators, particularly when foreign affiliates and economically significant spillovers are involved, to reduce unintended spillovers and improve risk-sharing.

*Source: Appendix A.2. Additional Tables and Counterfactual Figures — wp18217 (sections I–VII, empirical framework, main results, robustness, and conclusions).*

### REFERENCES

### REFERENCES

### Empirical framework and micro-foundation (Appendix A.1)
- Bank optimization problem (intensive margin):
  - max_{L^j, L^{CB}_{ij}, L^{S}_{ij}, D, M} Π^j + Π^{CB}_{ij} + Π^{S}_{ij} + r_f M − r_D(D) D
  - subject to balance-sheet constraint: E + D ≥ L^j + L^{CB}_{ij} + L^{S}_{ij} + M
- Profit functions for lending:
  - Π^{CB}_{ij} = τ^{CB}_{ij} r^L_{ij}(L^{CB}_{ij}) L^{CB}_{ij} − C_j(a) L^{CB}_{ij} − c_j f^{CB}_{ij}
  - Π^{S}_{ij} = τ^{S}_{ij} r^L_{ij}(L^{S}_{ij}) L^{S}_{ij} − C_i(a) L^{S}_{ij} − c_j f^{S}_{ij}
  - Assumption: C_j(a) = a c_j − r_f, a c_j > r_f ∀j
- Loan demand (constant-elasticity):
  - L^{CB}_{ij}(r^{CB}_{L_{ij}}) = r^{−ε_{CB}}_{L_{ij}} A^{CB}_i
- Aggregation and log-linear outcome equation for direct cross-border flows:
  - cb_{ij} = β^{CB}_0 + λ^{CB}_j + χ^{CB}_i + γ d_{ij} + w^{CB}_{ij} − u^{CB}_{ij}
  - d_{ij} is log symmetric distance; λ^{CB}_j and χ^{CB}_i are lender and borrower fixed effects
- Extensive margin (Probit) representation (entry/connection probability):
  - ρ^{CB}_{ij} = Φ(γ^{CB}_0 + ξ^{CB}_j + ζ^{CB}_i + γ^{CB} d_{ij} − κ^{CB} φ^{CB}_{ij})
  - z^{CB}_{ij} = Φ^{−1}( ̃ρ^{CB}_{ij} ), η^{CB}_{ij} = φ(z^{CB}_{ij})/Φ(z^{CB}_{ij}) (inverse Mills ratio)
- Two-step estimation: Probit first stage to construct z and η; second-stage log-linear ML using ln{exp[δ(z+η)]−1} term to capture extensive margin

### Counterfactual exercise (Appendix A.1)
- Procedure (direct cross-border illustration):
  1. Generate predicted latent variable z^{CB}_{ij} and inverse Mills ratio η^{CB}_{ij} from true data. For non-connected pairs (T^{CB}_{ij}=0) define
     - η^{CB}_{ij} = −φ(z^{CB}_{ij})/(1−Φ(z^{CB}_{ij})) if T^{CB}_{ij}=0
     - η^{CB}_{ij} = φ(z^{CB}_{ij})/Φ(z^{CB}_{ij}) if T^{CB}_{ij}=1
  2. Switch lender regulatory barrier r_i to r'_i. Obtain counterfactual latent z'^{CB}_{ij} from original first-stage estimates. Generate predicted counterfactual flows for pairs with positive z'^{CB}_{ij}+η^{CB}_{ij} using:
     - cb^c_{ij} = β_0 + λ_i + ξ_j + β_2 r'_i + θ_2 r_j + γ_2 d_{ij} + ln{exp[δ(z'^{CB}_{ij}+η^{CB}_{ij})−1]} + β η^{CB}_{ij}
     - Note: η^{CB}_{ij} uses original z^{CB}_{ij}, not z'^{CB}_{ij}
  3. Generate predicted flows using true data cb^p_{ij} with original second-stage estimates. Aggregate both cb^c_{ij} and cb^p_{ij} by year and compare

### Summary statistics (Table 2 — exact values)
- Dependent Variables:
  - Direct Cross-Border Connection: N 55970; Mean 0.23; SD 0.42; Min 0; Max 1; Country/Pair 5597
  - Local Affiliate Connection: N 55970; Mean 0.07; SD 0.25; Min 0; Max 1; Country/Pair 5597
  - Log Direct Cross-Border Flow: N 12772; Mean -2.92; SD 2.77; Min -13.81; Max 6.55; Country/Pair 3216
  - Log Local Affiliate Flow: N 3763; Mean -3.55; SD 3.47; Min -19.1; Max 5.55; Country/Pair 1234
- Gravity Factors and Instruments (selected):
  - Log Distance: N 5597; Mean 8.7; SD 0.82; Min 4.18; Max 9.9; Country/Pair 5597
  - Common Official Language: N 5539; Mean 0.14; SD 0.35; Min 0; Max 1; Country/Pair 5539
  - Free Trade Agreement (WTO): N 54230; Mean 0.22; SD 0.41; Min 0; Max 1; Country/Pair 5423
- Control Variables (selected; lender / borrower as indicated):
  - Policy-related / Short-term Interest Rate: Lender: N 290; Mean 2.8; SD 3.16; Min 0.001; Max 18; Country/Pair 29
  - Policy-related / Short-term Interest Rate: Borrower: N 1253; Mean 6.37; SD 5.63; Min 0.02; Max 70; Country/Pair 144
  - Legal System and Property Rights: Lender: N 290; Mean 7.18; SD 1.24; Min 4.22; Max 9.14; Country/Pair 29
  - Legal System and Property Rights: Borrower: N 1468; Mean 5.33; SD 1.6; Min 1.43; Max 9.14; Country/Pair 156
  - Credit to GDP: Lender: N 290; Mean 105.11; SD 43.12; Min 16.51; Max 233.21; Country/Pair 29
  - Credit to GDP: Borrower: N 1596; Mean 53.23; SD 44.58; Min 1.8; Max 312.12; Country/Pair 160
  - Real GDP Growth: Lender: N 290; Mean 2.23; SD 3.78; Min -9.13; Max 26.28; Country/Pair 29
  - Real GDP Growth: Borrower: N 1858; Mean 3.79; SD 4.76; Min -62.1; Max 34.5; Country/Pair 188
- Regulatory measures (selected):
  - Overall Prudential Index: Lender: N 290; Mean 2.26; SD 1.57; Min 0; Max 7; Country/Pair 29
  - Overall Prudential Index: Borrower: N 1570; Mean 1.9; SD 1.6; Min 0; Max 9; Country/Pair 157
  - Overall Outflow Control: Lender: N 290; Mean 0.22; SD 0.26; Min 0; Max 1; Country/Pair 29
  - Overall Inflow Control: Borrower: N 1020; Mean 0.34; SD 0.31; Min 0; Max 1; Country/Pair 102
  - Non-Core Activities Restriction: Lender: N 272; Mean 6.54; SD 1.97; Min 3; Max 10; Country/Pair 29
  - Non-Core Activities Restriction: Borrower: N 1280; Mean 7.71; SD 1.93; Min 3; Max 12; Country/Pair 152

### First-stage (Probit) average marginal effects — selected coefficients (Table 3 and A2)
- Table 3 (Overall MPM / CCM, average marginal effects; standard errors in parentheses)
  - Overall macroprudential index (Lender): 0.0225*** (0.00373) — Direct Cross-Border (column 1)
  - Overall macroprudential index (Borrower): -0.0143*** (0.00390) — Direct Cross-Border (column 1)
  - Overall outflow restriction (Lender): 0.0516* (0.0298) — Direct Cross-Border (column 2)
  - Overall inflow restriction (Borrower): 0.0863*** (0.0306) — Local Affiliate (column 7)
  - Real GDP Growth (Borrower): 0.00333*** (0.000744) — Direct Cross-Border (column 1)
  - Credit to GDP (Borrower): 0.000581*** (0.000187) — Direct Cross-Border (column 1)
  - Log Distance: -0.0541*** (0.00535) — Direct Cross-Border (column 1)
  - Common Language: 0.0700*** (0.0116) — Direct Cross-Border (column 1)
  - Sample sizes: N 33314 (column 1); N 24783 (column 5)
- Table A2(a) / A2(b) (first-stage specific measures, selected):
  - Overall (Lender): Direct Cross-Border 0.0337*** (0.00554); Local Affiliate -0.00280 (0.00494); N 19998 / 15973; Pseudo R-sq 0.154 / 0.225
  - LTV (Lender): Direct Cross-Border 0.0885*** (0.0158); Local Affiliate -0.0155 (0.0126)
  - Overall outflow restriction (Lender): Direct Cross-Border 0.0944*** (0.0353); Local Affiliate -0.00990 (0.0323)
  - Foreign Credit (Lender): Direct Cross-Border -0.0363** (0.0158); Local Affiliate -0.0306** (0.0131)

### Second-stage (ML) structural estimates — selected coefficients (Table 4, 5, 8)
- Table 4 (Overall MPM & CCM; structural parameters; standard errors in parentheses)
  - Overall macroprudential index:
    - Lender: -0.119* (0.0702) — Direct Cross-Border (column 2, All specification)
    - Borrower: 0.151*** (0.0463) — Direct Cross-Border (column 2, All specification)
  - Overall outflow restriction (Lender): 1.417* (0.773) — Direct Cross-Border (column 3, CCM)
  - Overall inflow restriction (Borrower): 0.568 (0.847) — Direct Cross-Border (column 3, CCM)
  - Structural parameters (Gravity specification, Direct Cross-Border):
    - δ: 1.845*** (0.214)
    - σ: 1.643*** (0.0168)
    - β: 0.642 (0.504)
  - Sample sizes: Direct Cross-Border N 9170 (Gravity) to 6754 (All); Local Affiliate N 2952 to 2379
- Table 5 (Second-stage specific MPM estimates; coefficients of interest only; standard errors in parentheses)
  - Overall (Lender): -0.119* (0.0702) — Direct Cross-Border; (Lender) 0.279*** (0.106) — Local Affiliate; N 6754 / 2379
  - LTV (Borrower): 0.146 (0.107) — Direct Cross-Border; 0.111 (0.249) — Local Affiliate; N 6754 / 2379
  - Leverage Ratio (Lender): -0.907*** (0.300) — Direct Cross-Border; 0.683 (0.460) — Local Affiliate
  - Foreign Currency Loan (Borrower): 0.493*** (0.184) — Direct Cross-Border; -0.326 (0.328) — Local Affiliate
- Table 8 (Capital control specific second-stage estimates; standard errors in parentheses)
  - Overall (Lender Outflow): -0.116 (0.324) — Direct Cross-Border; 2.532*** (0.834) — Local Affiliate; N 6754 / 2379
  - Bond (Lender Outflow): 0.262 (0.165) — Direct Cross-Border; 1.600*** (0.587) — Local Affiliate
  - Commercial Credit (Lender Outflow): 0.520** (0.252) — Direct Cross-Border
  - Foreign Credit (Borrower Inflow): 0.0843 (0.106) — Direct Cross-Border; 0.839** (0.424) — Local Affiliate

### Heterogeneity and robustness (selected)
- AE vs EMDE borrower breakdown (Table 7 & 9 — selected coefficients, standard errors in parentheses)
  - AE borrower sample (Direct Cross-Border):
    - Overall Borrower: 0.141** (0.0573)
    - Leverage Ratio (Lender): -0.800** (0.353)
  - EMDE borrower sample (Direct Cross-Border):
    - Overall Lender: -0.377*** (0.113)
    - Leverage Ratio (Lender): -1.309*** (0.427)
  - Capital control heterogeneity (Table 9):
    - Commercial Credit (AE Lender Outflow): 1.003** (0.394) — Direct Cross-Border; Borrower (AE) Commercial Credit: 1.963** (0.925)
    - Bond (AE Lender Outflow): -0.00381 (0.207) — Direct Cross-Border; 1.515*** (0.558) — Local Affiliate
- Robustness (Table A2(d) and Tables 10–11; selected entries)
  - Right-censoring robustness (Table 10):
    - Macroprudential Overall (Lender): -0.136* (0.0773) — Direct Cross-Border; 0.296*** (0.107) — Local Affiliate; N 6502 / 2292
    - Capital Control Overall (Lender Outflow): -0.277 (0.333) — Direct Cross-Border; 3.043*** (0.858) — Local Affiliate; N 6502 / 2292
  - 2011–2015 sample (Table 11):
    - Macroprudential Overall (Lender): -0.214** (0.0917) — Direct Cross-Border; 0.453*** (0.151) — Local Affiliate; N 2931 / 1047
    - Leverage Ratio (Lender, 2011–2015): -1.584*** (0.429) — Direct Cross-Border

### Additional notes on figures and counterfactual aggregates
- Counterfactual exercise implementation details:
  - “Model prediction” uses second-stage equation with parameters estimated from true data
  - “Counterfactual” sets the policy index to zero (switches off existing measures) and generates counterfactual z' and resulting cb^c_{ij} as described in Appendix A.1
  - For each policy instrument, model prediction and counterfactual calculations are generated based on the sample of lenders/borrowers that have ever adopted that instrument; for each year, net positive increases in direct cross-border and local affiliate exposure are predicted for each country pair and summed to global level
- Figures referenced (Figure 1–4, Figure A2(a)) present:
  - Time-series evolution of global usage of macroprudential instruments and capital controls (Figure 1)
  - Model prediction vs counterfactual banking flows for overall macroprudential policy and overall capital flow restrictions (Figures 2–3)
  - Interaction coefficients between macroprudential policy, capital control and direct cross-border banking (Figure 4)
  - Additional counterfactual figures by specific macroprudential and capital control instruments (Figure A2(a))

*Source: wp18217 - REFERENCES (IMF Working Paper content unit).*

### Appendix A1.  For each policy instrument,  model prediction and counterfactual calculation are

### Appendix A1.  For each policy instrument,  model prediction and counterfactual calculation are

### Sample selection
- Generated based on a sample of lenders/borrowers that have ever adopted this instrument.

### Prediction methodology
- For each year, the magnitude of net positive increase in direct cross-border and local affiliate exposure is predicted for each country pair in the sample.

### Aggregation / Counterfactual calculation
- Predicted magnitudes are summed to the global level.

*Source: wp18217 - Appendix A1.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2018/wp18217.pdf_