## Capital Account Liberalization and Bank-Level Effects in LIDCs (wpiea2021237-print-pdf excerpt) ## Source details **Canonical URL:** [Capital Account Liberalization and Bank-Level Effects in LIDCs (wpiea2021237-print-pdf excerpt)](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021237-print-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021237-print-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021237-print-pdf.pdf.json) --- ### II. OPENING UP: CAPITAL ACCOUNT LIBERALIZATION IN LIDCs — Measurement and patterns - Measurement and data - Updated the Wang-Jahan Index of capital account openness until 2017, following the same methodology as Jahan and Wang (2016) (Appendix I, Section B). The Wang-Jahan Index measures de jure capital account openness based on survey data from the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER, IMF, 2019a). - The Wang-Jahan Index provides a simple average over 12 subcategories, ranging from 0 (fully closed) to 1 (unrestricted in all 12 sub-categories). The index offers granular breakdown across twelve types of cross-border transactions. - Key findings on capital account openness - Many LIDCs have gradually liberalized their capital accounts in the past decade. - Average capital account openness in frontier market LIDCs has surpassed average EME openness during the global financial crisis, when certain EMEs tightened capital controls. - Non-frontier LIDCs have also gradually moved toward greater capital account openness, although average openness for overall LIDCs is still significantly lower than openness in EMEs. - LIDCs have opened up fairly evenly across all types of cross-border financial transactions, but interquartile ranges for LIDCs indicate a high degree of heterogeneity across countries. - Small and fragile countries are excluded from the figure. EMEs: emerging market economies; LIDCs: low-income developing countries. - Capital inflows accompanying liberalization - De jure capital account liberalization in LIDCs was accompanied by an increase in gross capital inflows, particularly to frontier market LIDCs. - In addition to FDI flows, LIDCs have received growing amounts of portfolio and other flows. “Other flows” mainly comprise direct loans and trade credit. - Private non-FDI flows exhibit volatile and pro-cyclical dynamics similar to those experienced by EMEs. - Aggregate co-movement with financial deepening - At the country level, domestic credit and gross capital inflows to the private sector move in tandem in LIDCs on average. - After being relatively constant for more than two decades, the average level of domestic credit to GDP has substantially increased in the early 2000s, coinciding with the documented capital inflow surge. ### III. FINANCIAL SECTOR DYNAMICS UNDER LARGE CAPITAL INFLOWS — Data and empirical strategy - Bank-level micro data - Bank-level balance sheet data from Fitch Fundamental Financial Data covering banks in LIDCs over 1989 to 2019. - Baseline dataset: a total of 584 banks from 32 LIDCs (number varies by regression specification). - Large share of banks from frontier market LIDCs (examples: Vietnam, Kenya, Nigeria). Most represented countries are not heavily reliant on commodity exports. - Micro-data sample is not representative of all LIDCs; it over-represents more open frontier-market banks that receive more non-FDI private capital flows. - Capital flow data from IMF’s Financial Flows Analytics database (FFA, IMF, 2021a). Other macro variables from IFS (IMF, 2021b) and WEO (IMF, 2021c). - Empirical strategy - Estimated panel regressions of bank-level variable Y_{j,i,t} (e.g., gross loans) on capital inflows K_{i,t}, with lags, macro and bank controls, time dummies, and bank fixed effects: - Y_{j,i,t} = β K_{i,t} + φ X_{i,t−1} + ψ B_{j,i,t−1} + D_{t} + α_{j} + ε_{j,t} - X_{i,t} includes macro controls (including lagged capital inflows K_{i,t−1}); B_{j,i,t} includes bank-level controls (including lagged dependent variable Y_{i,j,t−1}); D_{t} are year dummies; α_{j} are bank fixed effects. - Standard errors: Driscoll and Kraay (1998) to account for arbitrary cross-sectional dependence and some autocorrelation. - Methodological caveats: reverse causality (addressed in Section IV methods), Nickell bias from lagged dependent variable, and small number of clusters affecting inference. - Robustness checks: GMM dynamic panels, panel-split jackknife, sample restrictions to minimum observations per bank (4 and 8 years); these indicate Nickell bias still has non-negligible influence on OLS results. ### Main empirical findings: credit growth, funding, and balance-sheet responses - Credit growth and capital inflows (baseline OLS) - A one percent increase in capital flows associated with about 6 basis point increase in gross loans at banks on average. - A one (within) standard deviation increase in capital inflows associated with an increase of 5 percent in annual loan growth. - By comparison, the corresponding impact of real GDP growth is 2.6 percent (based on the sum of effects from both the level and growth terms). - Replacing time fixed effects by global control variables leads to only a marginal change in results. - Bank balance-sheet responses (baseline OLS and tables) - Domestic credit, bank deposits, and wholesale funding are positively associated with capital inflows. - A one percentage point increase in gross capital inflows is associated with: - an increase in gross loans of 6 basis points in the baseline OLS (Table 1: Total gross capital inflows (log) 0.12*** in column (1) and coefficients around 0.06***–0.07*** in other specifications); - increases in deposits and wholesale funding (Table 3: Total gross capital inflows (log) 0.07** for Wholesale funding and 0.05*** for Deposits). - Regulatory capital ratio and leverage ratio changes are only weakly significant, indicating some build-up of financial vulnerabilities but not strongly so. - Asset quality, profitability, and liquidity - Nonperforming loans (NPLs) and return on assets show no significant average change (Table 4, columns (1) and (2): Total gross capital inflows (log) -0.35 and 0.04, respectively, not statistically significant). - Net interest margin shows no significant change (Table 4, column (3): -0.01). - Liquid asset to short-term funding ratio shows a statistically significant decline (Table 4, column (4): -0.79**). - Business dynamics and financial inclusion signals - Number of employees increases significantly with capital inflows (Table 4, column (6): Total gross capital inflows (log) 0.02***). - Personnel expenses per overhead increase insignificantly (Table 4, column (5): 0.01). - Interpretation: labor expansion implies lending may extend to new customers, suggesting contributions to financial inclusion. - Flow composition and funding mix - Results similar when focusing on capital flows to the private sector only. - Both FDI and non-FDI inflows associated with higher loan growth, although associations are weak (Table 2, column 5). - Noted that a sizable portion of FDI to LIDCs are debt instruments; FDI may enter via foreign-owned banks or produce second-round spillovers easing credit constraints. - Wholesale funding increased in response to capital inflows, and bank deposits also grew (Table 3). - With both loans and deposits increasing, loan-to-deposit ratio saw only a moderate increase (text ends here). ### Distributional, ownership, and size heterogeneity - Distributional heterogeneity and quantile regression - Capital inflows increase skewness of the banking distribution for loans and loan-to-deposit ratio. - Quantile estimates: - A one percentage point increase in gross capital inflows associated with an increase in loans of 6.4 basis points for the highest decile and 4.7 basis points for the lowest decile. - Deposits: lowest decile exhibits 6.3 basis points correlation with capital inflows; highest decile 3.1 basis points. - Loan-to-deposit ratio estimates increase from the lowest decile to the highest decile within a narrow range of 0.73-0.75. - Wholesale funding shows a decreasing pattern from 8.7 basis points to 6.1 basis points across quantiles. - Differences across quantiles are not statistically significant but dynamics suggest some banks respond especially strongly and may face tail risks. - Ownership heterogeneity (Table 5) - Domestic banks expand loans more than foreign banks: Domestically owned banks 0.088*** vs Foreign owned banks 0.043*** for Gross loans (interaction coefficients). - The association between foreign banks’ loans and capital inflows is less than a half of that for domestic banks (difference significant at the 1 percent level). - NPLs decline significantly for domestic banks (Panel B: Domestically owned banks -0.934***), difference versus foreign banks significant at the 5 percent level. - Domestic banks show significant increases in personnel expenses per overhead (0.023*) and number of employees (0.055***). - Size heterogeneity (Table 6) - Smaller banks (below 25th percentile) extend more loans when receiving capital inflows (0.065*** interaction). - Interaction coefficients for Gross loans decline monotonically as asset size percentile increases (0.065***, 0.062*, 0.050**, 0.040 for increasing size quartiles), though differences are not statistically significant. - Some increased financial vulnerability indicators appear for lower middle-sized banks, but caution advised due to possible sample splicing effects. ### Causality assessment: methods, diagnostics, and results - Approaches to address reverse causality - Instrumental variable (regional capital-flow instrument) following Blanchard et al. methodology. - Augmented inverse-probability weighted (AIPW) estimator with quantile binning of treatment into three levels ([0.15, 0.85] baseline). - Instrumental variable (Table 7) diagnostics and estimates - With year fixed effects, 2SLS indicates little causal effect: coefficient reduces to 2 basis points (from baseline 6 basis points) and loses statistical significance. - First-stage F value: 11.80 (below non-i.i.d. benchmark of 23 with 10 percent worst-case bias), indicating weak instruments. - Replacing year fixed effects with global control variables increases instrument strength: F value 61.64 and 2SLS estimate of 14 basis points, but concerns about exogeneity remain. - Overall: IV approach does not strongly indicate causal effect from capital inflows to bank lending due to weak region-specific variation and potential endogeneity. - AIPW estimator (Table 8) results - Baseline [15, 85] thresholds: estimated ATE for Gross loans of 0.03 per one ppt of capital inflows (3 basis points) for both low and high levels; for Deposits high levels 0.04*** (4 basis points). - Results are not robust across alternative percentile thresholds ([20, 80], [25, 75]) and vary in significance and magnitude. - Interpretation: AIPW yields at best weak and unstable evidence of causal impacts, possibly due to sample size relative to parameter dimensionality and instability of AIPW. - Conclusion on causality - Both IV and AIPW methods provide only weak evidence of a causal effect from capital inflows to bank lending. - Domestic “pull” factors and strengthened fundamentals may explain much of the observed correlation. - OLS likely overstates causal magnitude under two-way causality and simultaneity (Appendix III formalization). ### Robustness checks, biases, and data limitations - Nickell bias (Appendix II.A) - Lagged dependent variable with fixed effects generates Nickell bias; sample average length 9 years and minimum 4 years. - Panel GMM (Arellano-Bond style): results broadly similar to baseline OLS; coefficient on capital flows on loans becomes slightly larger than OLS, significance lost for wholesale funding and deposits. - Split-panel jackknife bias correction: - Baseline result of 6.7 basis points changes slightly to 7.5 basis points under split-panel jackknife likelihood estimator, but point estimate no longer statistically significant. - Estimated size of the Nickell bias is 12 percent of the size of the coefficient; estimation precision declines when correcting for Nickell bias. - Table A5 summaries: - Baseline (Table 1, col. 3, Table 3): Gross loans = 0.055***; Wholesale funding = 0.073**; Deposits = 0.047***; Loan-to-deposit ratio = 0.739**. - Half-panel jackknifed likelihood: Gross loans = 0.057; Wholesale funding = 0.087; Deposits = 0.054; Loan-to-deposit ratio = 0.543. - Half-panel jackknifed likelihood, excluding observations with any data gaps: Gross loans = 0.061; Wholesale funding = 0.083; Deposits = 0.078; Loan-to-deposit ratio = 0.709. - Half-panel parameter jackknife: Gross loans = 0.061*; Wholesale funding = 0.121*; Deposits = 0.058**; Loan-to-deposit ratio = 0.322*. - Conclusion: Nickell bias non-negligible; limited long bank-level time series in LIDCs remains a primary challenge. - Small number of clusters (Appendix II.B) - Driscoll-Kraay standard errors cluster by year; number of sample periods (years) is 31 in the unbalanced panel—potentially insufficient. - Small number of clusters can lead to over-rejection and spurious significance due to leveraged observations producing mechanically small standard errors. - No consensus on bootstrap corrections for Driscoll-Kraay in this context. - Capital-flow surges and non-average responses (Appendix II.C) - Surge identification: 10 capital-inflow surge episodes identified following Caballero (2016) in 481 country-year observations using HP-filtered gap higher than two standard errors. - Bonanza/surge results directionally similar to log-level results but sensitive to surge-identification method and not robust across methods. - Selected surge table highlights: - Table A6 coefficients include 0.04, 0.08*, 0.07**, 0.12*** across specifications for total inflows bonanza – two S.D. - Table A7: Liquid asset to short-term funding ratio = 2.51***; Number of employees = 0.09** under bonanza specification. - Table A8: Alternative definitions produce varying coefficients (e.g., Total inflows bonanza – two S.D. = 0.07** (0.03); Total inflows surge – top 30 percent = 0.04* (0.02)). - Data gaps and omitted tests - Lack of data on foreign currency liabilities and loans prevents testing for currency mismatches—a key omission given historical crisis channels under fixed exchange rate regimes. - Potential time lags between inflows and observable vulnerabilities can hinder ex ante detection of vulnerabilities. ### Policy implications and recommendations - Macroprudential and policy framework priorities - Need for a sound macroprudential framework when capital accounts are liberalized, given that implied credit fluctuations associated with capital flows are larger than those from domestic real activity. - Loan expansions funded by deposit increases may reflect healthy financial deepening; key question is whether expansion exceeds fundamentals. - Policy responses must balance promoting financial deepening and addressing financial vulnerabilities; careful judgment required to assess whether expansion is excessive relative to economic fundamentals. - Given data gaps (notably currency composition of assets/liabilities), further micro- and macro-prudential surveillance and continued research are policy priorities. - Areas for active policy debate and further guidance - When capital controls are justifiable. - Monetary policy autonomy with open capital accounts. - Use and implications of macroprudential policies. - Foreign exchange interventions and their domestic financial sector implications. - The IMF’s “Integrated Policy Framework” work is relevant for guidance under these policy interactions. ### Selected exact statistics and diagnostics (preserved verbatim) - Sample and sample periods - Sample: 32 countries and 584 banks (Table 1 note). - Table 1 sample period 1989-2019 (columns (1)–(3)) and 1991-2019 (column (4)). - Table 3 sample periods include 1992-2019, 1989-2019, 1993-2018, etc. - Key coefficients and significance (verbatim) - Table 1: Total gross capital inflows (log): 0.12*** (0.03) in column (1). - Table 1: Gross loans (log), lag: 0.68*** (0.02) in column (1). - Table 1: Real GDP (log), lag: 0.37*** (0.10) in column (3). - Table 1: US inflation, lag: 6.11*** (2.10) in column (4). - Table 3: Wholesale funding lag coefficient: 0.38*** (0.05) (column (1)). - Table 3: Loan-to-deposit ratio lag (winsorized): 0.48*** (0.05) (column (3)). - Table 3: Regulatory capital ratio lag (winsorized): 0.44*** (0.05) (column (4)). - Table 4: Total gross capital inflows (log) effects: Liquid asset to short-term funding ratio -0.79** (0.35); Number of employees 0.02*** (0.01). - Instrument diagnostics (Table 7) - First-stage F: 11.80 (with year fixed effects) and F: 61.64 (without year fixed effects when using global control variables). - 2SLS estimates: Gross loans 0.02 (column (1) 2SLS) and 0.14** (column (4) 2SLS with global controls). - AIPW per-unit ATEs (Table 8) relative to medium level per one ppt of capital inflows - Gross loans low levels 0.03 (0.05) in [15, 85]; high levels 0.03** (0.01) in [15, 85]. - Deposits high levels 0.04*** (0.01) in [15, 85]. *Source: wpiea2021237-print-pdf (excerpt provided).* ### Section IV turns to the identif ication of causal ef f ects. Section V concludes with policy ### II. OPENING UP: CAPITAL ACCOUNT LIBERALIZATION IN LIDCS ### Measurement and data - Updated the Wang-Jahan Index of capital account openness until 2017, following the same methodology as Jahan and Wang (2016) (Appendix I, Section B). The Wang-Jahan Index measures de jure capital account openness based on survey data from the IMF’s Annual Report on Exchange Arrangements and Exchange Restrictions (AREAER, IMF, 2019a). - The Wang-Jahan Index provides a simple average over 12 subcategories, ranging from 0 (fully closed) to 1 (unrestricted in all 12 sub-categories). The index offers granular breakdown across twelve types of cross-border transactions. - Several indices are based on the AREAER; for comparisons see Cerdeiro and Komaromi (2019) (as noted in the source). ### Key findings on capital account openness - Many LIDCs have gradually liberalized their capital accounts in the past decade. - Average capital account openness in frontier market LIDCs has surpassed average EME openness during the global financial crisis, when certain EMEs tightened capital controls (Figure 1, Panel A). - Non-frontier LIDCs have also gradually moved toward greater capital account openness, although average openness for overall LIDCs is still significantly lower than openness in EMEs. - LIDCs have opened up fairly evenly across all types of cross-border financial transactions, but interquartile ranges for LIDCs indicate a high degree of heterogeneity across countries (Figure 1, Panel B). - Small and fragile countries are excluded from the figure. EMEs: emerging market economies; LIDCs: low-income developing countries. ### Capital inflows accompanying liberalization - De jure capital account liberalization in LIDCs was accompanied by an increase in gross capital inflows, particularly to frontier market LIDCs (Figure 2). - In addition to FDI flows, LIDCs have received growing amounts of portfolio and other flows. “Other flows” mainly comprise direct loans and trade credit. - Private non-FDI flows exhibit volatile and pro-cyclical dynamics similar to those experienced by EMEs (references in source: Abidi, Hacibedel, and Mwanza, 2016; Araujo and others, 2017b). ### Aggregate co-movement with financial deepening - At the country level, domestic credit and gross capital inflows to the private sector move in tandem in LIDCs on average (Figure 3). - After being relatively constant for more than two decades, the average level of domestic credit to GDP has substantially increased in the early 2000s, coinciding with the documented capital inflow surge. - Figure 2 source: Financial Flows Analytics (FFA, IMF, 2021a). Note: Figure shows total gross capital flows to private recipients in LIDCs in percent of the total GDP of LIDCs. - Figure 3 sources: World Bank’s Global Financial Development database and Financial Flows Analytics (FFA, IMF, 2021a). Note: Figure shows sample averages over 36 LIDCs based on aggregate macro data. ### III. FINANCIAL SECTOR DYNAMICS UNDER LARGE CAPITAL INFLOWS ### Data: bank-level micro data - Used bank-level balance sheet data from Fitch Fundamental Financial Data covering banks in LIDCs over the period 1989 to 2019. - Baseline dataset contains a total of 584 banks from 32 LIDCs, though the number varies depending on variables used in regression estimation (Appendix I, Section C; Appendix III). - A large share of banks in the sample is from frontier market LIDCs (examples: Vietnam, Kenya, Nigeria). Most represented countries are not heavily reliant on commodity exports. - The micro-data sample is not representative of all LIDCs; due to large share of frontier market banks, it represents those more open than the LIDC average and that receive more non-FDI private capital flows. - Capital flow data are from the IMF’s Financial Flows Analytics database (FFA, IMF, 2021a). Other macro variables are from IFS (IMF, 2021b) and WEO (IMF, 2021c). Table A2 in Appendix I lists all variables and data sources. ### Empirical strategy - Estimated panel regressions of bank-level variable Y_{j,i,t} (e.g., gross loans) on capital inflows K_{i,t}, with lags, macro-level and bank-level controls, time dummies, and bank fixed effects: - Y_{j,i,t} = β K_{i,t} + φ X_{i,t−1} + ψ B_{j,i,t−1} + D_{t} + α_{j} + ε_{j,t} - Where X_{i,t} includes macro controls (including lagged capital inflows K_{i,t−1}); B_{j,i,t} includes bank-level controls (including lagged dependent variable Y_{i,j,t−1}); D_{t} are year dummies; α_{j} are bank fixed effects. - Standard errors: Driscoll and Kraay (1998) to account for arbitrary cross-sectional dependence and some autocorrelation, reflecting within-country correlation from capital flows, cross-country synchronicity, and persistence over time. - Methodological caveats noted: reverse causality (addressed further in Section IV), Nickell bias from lagged dependent variable, and small number of clusters which could affect inference. - Robustness checks (Appendix II and III): GMM estimation of dynamic panel models, Nickell bias correction via panel-split jackknife, and sample restrictions to minimum observations per bank (4 and 8 years). These analyses indicate the Nickell bias still has non-negligible influence on OLS results. ### Main empirical findings: credit growth and capital inflows - OLS results show capital flows are associated with sizeable increases in bank loans in LIDCs (Table 1). - A one percent increase in capital flows was associated with about 6 basis point increase in gross loans at banks on average. - A one (within) standard deviation increase in capital inflows is associated with an increase of 5 percent in annual loan growth. - By comparison, the corresponding impact of real GDP growth is 2.6 percent (based on the sum of effects from both the level and growth terms). - Replacing time fixed effects by global control variables leads to only a marginal change in results (Table 1, column 4). ### Controls and intuitive associations - Higher lagged domestic interest rates (funding side) are associated with lower loan growth. - Higher levels of real GDP and stronger institutional quality are associated with higher loan growth. - Positive correlation with export prices may reflect more investment opportunities and greater financing needs for higher imports. - A banking crisis is associated with lower loans; association with a currency crisis is not significant. ### Flow composition and banking sector implications - Analysis by flow type: - Results do not change much if focusing on capital flows to the private sector only, possibly reflecting substitution of government domestic borrowing by official flows and unlocked bank resources for private lending. - Both FDI and non-FDI inflows are associated with higher loan growth, although associations are weak (Table 2, column 5). - Noted that a sizable portion of FDI flows to LIDCs are debt instruments, which can reconcile the FDI–credit linkage; FDI may also enter via foreign-owned banks or have second-round spillovers easing credit constraints. - Funding composition responses: - Regression results show wholesale funding increased in response to capital inflows, and bank deposits also grew (Table 3). - With both loans and deposits increasing, the loan-to-deposit ratio—a typical indicator for financial vulnerability—saw only a moderate increase (text ends here). *Italic line: Source: wpiea2021237-print-pdf (excerpt provided).* ### 0.74 percent. The changes in the regulatory capital ratio and the leverage ratio (measured as ### wpiea2021237-print-pdf - 0.74 percent. The changes in the regulatory capital ratio and the leverage ratio (measured as ### Main empirical findings on bank balance sheets - Domestic credit, bank deposits, and wholesale funding are positively associated with capital inflows. - A one percentage point increase in gross capital inflows is associated with: - an increase in gross loans of 6 basis points in the baseline OLS (Table 1: To t a l gross ca pital inflows (log) 0.12*** in column (1) and coefficients around 0.06***–0.07*** in other specifications); - increases in deposits and wholesale funding (Table 3: To t a l gross ca pital inflows (log) 0.07** for Wholesale funding and 0.05*** for Deposits). - Regulatory capital ratio and leverage ratio (total assets to equity) changes are only weakly significant; they indicate some build-up of financial vulnerabilities but not strongly so (text: "only weakly significant"). - Asset quality and profitability: - Nonperforming loans (NPLs) and return on assets show no significant average change (Table 4, columns (1) and (2): Total gross capital inflows (log) -0.35 and 0.04, respectively, not statistically significant). - Net interest margin shows no significant change (Table 4, column (3): -0.01). - Liquid asset to short-term funding ratio shows a statistically significant decline (Table 4, column (4): -0.79**). - Bank business dynamics and financial inclusion: - Number of employees increases significantly with capital inflows (Table 4, column (6): Total gross capital inflows (log) 0.02***). - Personnel expenses per overhead increase insignificantly (Table 4, column (5): 0.01). - Interpretation: labor expansion implies lending may extend to new customers, suggesting contributions to financial inclusion. ### Distributional heterogeneity and quantile regression results - Capital inflows increase skewness of the banking distribution for loans and loan-to-deposit ratio. - Quantile estimates (text): - A one percentage point increase in gross capital inflows associated with an increase in loans of 6.4 basis points for the highest decile and 4.7 basis points for the lowest decile. - Deposits: lowest decile exhibits 6.3 basis points correlation with capital inflows; highest decile 3.1 basis points. - Loan-to-deposit ratio estimates increase from the lowest decile to the highest decile within a narrow range of 0.73-0.75. - Wholesale funding shows a decreasing pattern from 8.7 basis points to 6.1 basis points across quantiles. - Difference in estimated coefficients across quantiles are not statistically significant, but dynamics suggest some banks respond especially strongly and may face tail risks. ### Ownership and size heterogeneity - Ownership (Table 5): - Domestic banks expand loans more than foreign banks. Interaction coefficients: Domestically owned banks 0.088*** vs Foreign owned banks 0.043*** for Gross loans. - The association between foreign banks’ loans and capital inflows is less than a half of that for domestic banks (difference significant at the 1 percent level). - NPLs decline significantly for domestic banks (Panel B: Domestically owned banks -0.934***), difference versus foreign banks significant at the 5 percent level. - Domestic banks show significant increases in personnel expenses per overhead (0.023*) and number of employees (0.055***). - Size (Table 6): - Smaller banks (below 25th percentile) extend more loans when receiving capital inflows (0.065*** interaction). - Interaction coefficients for Gross loans decline monotonically as asset size percentile increases (0.065***, 0.062*, 0.050**, 0.040 for increasing size quartiles), though differences are not statistically significant. - Some increased financial vulnerability indicators appear for lower middle-sized banks, but caution advised due to possible sample splicing effects. ### Causality assessment: methods and results - Two complementary approaches used to address reverse causality: - Instrumental variable (regional capital-flow instrument) following Blanchard et al. methodology. - Augmented inverse-probability weighted (AIPW) estimator with quantile binning of treatment into three levels ([0.15, 0.85] baseline). - Instrument variable estimation (Table 7): - With year fixed effects, 2SLS indicates little causal effect: coefficient reduces to 2 basis points (from baseline 6 basis points) and loses statistical significance. - First-stage F value: 11.80 (below non-i.i.d. benchmark of 23 with 10 percent worst-case bias), indicating weak instruments. - Replacing year fixed effects with global control variables increases instrument strength: F value 61.64 and 2SLS estimate of 14 basis points, but concerns about exogeneity remain. - Overall: IV approach does not strongly indicate causal effect from capital inflows to bank lending due to weak region-specific variation and potential endogeneity. - AIPW estimator (Table 8): - Baseline [15, 85] thresholds: estimated ATE for Gross loans of 0.03 per one ppt of capital inflows (3 basis points) for both low and high levels; for Deposits high levels 0.04*** (4 basis points). - Results are not robust across alternative percentile thresholds ([20, 80], [25, 75]) and vary in significance and magnitude. - Interpretation: AIPW yields at best weak and unstable evidence of causal impacts, possibly due to sample size relative to parameter dimensionality and instability of AIPW. - Conclusion on causality: both methods provide only weak evidence of a causal effect; domestic “pull” factors and strengthened fundamentals may explain much of the correlation. ### Financial stability assessment and data limitations - Overall empirical analysis finds little evidence for significant increases in financial vulnerabilities in response to capital inflows. - Caveats: - Potential time lags between inflows and observable vulnerabilities (profitability declines, regulatory capital reductions, NPL increases) can hinder ex ante detection. - Lack of data on foreign currency liabilities and loans prevents testing for currency mismatches—a key omission given their historical role in crises under fixed exchange rate regimes. - Warning sign: loan-to-deposit ratio increases are strongest in the upper tail (banks that increase loans the most and deposits the least), which could signal tail risks requiring surveillance. ### Policy implications and recommendations - Need for a sound macroprudential framework when capital accounts are liberalized, given that implied credit fluctuations associated with capital flows are larger than those from domestic real activity. - Loan expansions funded by deposit increases may reflect healthy financial deepening; the key question is whether expansion exceeds fundamentals. - Policy responses must balance promoting financial deepening and addressing financial vulnerabilities; careful judgment required to assess whether expansion is excessive relative to economic fundamentals. - Given data gaps (notably currency composition of assets/liabilities), further micro- and macro-prudential surveillance and continued research are policy priorities. - Areas for active policy debate and further guidance include: - When capital controls are justifiable. - Monetary policy autonomy with open capital accounts. - Use and implications of macroprudential policies. - Foreign exchange interventions and their domestic financial sector implications. - The IMF’s “Integrated Policy Framework” work is relevant for guidance under these policy interactions. ### Selected exact statistics and model diagnostics (preserved verbatim) - Sample: 32 countries and 584 banks (Table 1 note). - Sample periods vary across tables (examples): - Table 1 sample period 1989-2019 (columns (1)–(3)) and 1991-2019 (column (4)). - Table 3 sample periods include 1992-2019, 1989-2019, 1993-2018, etc. - Table 1 key coefficients and significance: - To t a l gross ca pital inflows (log): 0.12*** (0.03) in column (1). - Gross loans (log), lag: 0.68*** (0.02) in column (1). - Real GDP (log), lag: 0.37*** (0.10) in column (3). - US inflation, lag: 6.11*** (2.10) in column (4). - Table 3 highlights: - Wholesale funding lag coefficient: 0.38*** (0.05) (column (1)). - Loan-to-deposit ratio lag (winsorized): 0.48*** (0.05) (column (3)). - Regulatory capital ratio lag (winsorized): 0.44*** (0.05) (column (4)). - Table 4 highlights: - Total gross capital inflows (log) effects: Liquid asset to short-term funding ratio -0.79** (0.35); Number of employees 0.02*** (0.01). - Instrument diagnostics (Table 7): - First-stage F: 11.80 (with year fixed effects) and F: 61.64 (without year fixed effects when using global control variables). - 2SLS estimates: Gross loans 0.02 (column (1) 2SLS) and 0.14** (column (4) 2SLS with global controls). - AIPW per-unit ATEs (Table 8) relative to medium level per one ppt of capital inflows: - Gross loans low levels 0.03 (0.05) in [15, 85]; high levels 0.03** (0.01) in [15, 85]. - Deposits high levels 0.04*** (0.01) in [15, 85]. *Italic attribution: Source: wpiea2021237-print-pdf (canonical URL provided with the content unit).* ### REFERENCES ### REFERENCES ### Academic articles, books, and working papers - Abidi, N., B. Hacibedel, and N. Mwanza. 2016. “Changing Times for Frontier Markets: A Perspective from Portfolio Investment Flows and Financial Integration.” IMF Working Paper, 16/177. - Agarwal, Isha, Rupa Duttagupta, and Andrea F. Presbitero. 2019. “International Commodity Prices and Domestic Bank Lending in Developing Countries.” Economic Inquiry, 58 (2): 953-979. - Akinci, Ozge, and Jane Olmstead-Rumsey. 2018. “How Effective are Macroprudential Policies? An Empirical Investigation.” Journal of Financial Intermediation, 33: 33–57. - Alam, Zohair, Adrian Alter, Jesse Eiseman, Gaston Gelos, Heedon Kang, Machiko Narita, Erlend Nier, and Naixi Wang. 2019. “Digging Deeper—Evidence on the Effects of Macroprudential Policies from a New Database.” IMF Working Paper, 19/66. - Anderson, T. W., and H. Rubin. 1949. “Estimation of the Parameters of Single Equation in a Complete System of Stochastic Equations.” Annals of Mathematical Statistics, 20: 46-63. - Angrist, Joshua D., Òscar Jordà, and Guido M. Kuersteiner. 2016. “Semiparametric Estimates of Monetary Policy Effects: String Theory Revisited.” Journal of Business and Economic Statistics, 36: 371–387. - Araujo, Juliana D., Antonio C. David, Carlos van Hombeeck, and Chris Papageorgiou. 2017a. “Non-FDI Capital Inflows in Low-Income Countries: Catching the Wave?” IMF Economic Review, 70: 157–82. - Araujo, Juliana D., Antonio C. David, Carlos van Hombeeck, and Chris Papageorgiou. 2017b. “Joining the Club? Procyclicality of Private Capital Inflows in Lower Income Developing Economies.” Journal of International Money and Finance, 70: 157–82. - Araujo, Juliana D., Povilas Lastauskas, and Chris Papageorgiou. 2017c. “Evolution of Bilateral Capital Flows to Developing Countries at Intensive and Extensive Margins.” Journal of Money, Credit and Banking, 49(7): 1517-1554. - Arce, Fernando, Julien Bengui, Javier Bianchi. 2019. “A Macroprudential Theory of Foreign Reserve Accumulation.” NBER Working Paper, 26236. - Arellano, Manuel, and Stephen Bond. 1991. “Some Tests of Specification for Panel Data: Monte Carlo Evidence and An Application to Emplyment Equations.” Review of Economic Studies, 58(2): 277-297. - Arellano, Manuel, and Olympia Bover. 1995. “Another Look at the Instrumental Variable Estimation of Error-components Models.” Journal of Econometrics, 68(1): 29-51. - Asonuma, Tamon, Marcos Chamon, and Akira Sasahara. 2016. “Trade Costs of Sovereign Debt Restructurings: Does a Market-Friendly Approach Improve the Outcome?” IMF Working Paper, 16/222. - Bal Gündüz, Yasemin. 2016. “The Economic Impact of Short-term IMF Engagement in Low-Income Countries.” World Development, 87: 30-49. - Banerjee, Abhijit V., and Esther Duflo. 2014. “Do Firms Want to Borrow More? Testing Credit Constraints using a Directed Lending Program.” Review of Economic Studies, 81(2): 572-607. - Barajas, A., T. Beck, E. Dabla-Norris, and R. Yousefi. 2013. “Too Cold, Too Hot, or Just Right? Assessing Financial Sector Development Across the Globe.” IMF Working Paper, 13/81. - Baskaya, Yusuf Soner, Julian di Giovanni, Şebnem Kalemli-Özcan, José Luis Peydro, and Mehmet Fatih Ulu. 2017a. “Capital Flows and the International Credit Channel.” Journal of International Economics, 108(1): 15–22. - Baskaya, Yusuf Soner, Julian di Giovanni, Şebnem Kalemli-Özcan, and Mehmet Fatih Ulu. 2017b. “International Spillover and Local Credit Cycles.” NBER Working Paper, 23149. - Blanchard, Olivier, Gustavo Adler, Irineu de Carvalho Filho. 2015. “Can foreign exchange intervention stem exchange rate pressures from global capital flow shocks?” NBER Working Paper, 21427. - Blanchard, Olivier, Jonathan D. Ostry, Atish R. Ghosh, and Marcos Chamon. 2016. “Capital Flows: Expansionary or Contractionary?” American Economic Review, 106(5): 65-69. - Blundell, Richard, and Stephen Bond. 1998. “Initial Conditions and Moment Restrictions in Dynamic Panel Data Models.” Journal of Econometrics, 87(1): 115-143. - Bruno, Valentina, and Hyun Song Shin. 2015a. “Cross-Border Banking and Global Liquidity.” Review of Economic Studies, 82(2): 535–64. - Bruno, Valentina, and Hyun Song Shin. 2015b. “Capital Flows and the Risk-Taking Channel of Monetary Policy.” Journal of Monetary Economics, 71(C): 119-32. - Caballero, Julián A. 2016. “Do Surges in International Capital Inflows Influence the Likelihood of Banking Crises?” Economic Journal, 126: 281-316. - Calderón, César, and Megumi Kubota. 2012. “Gross Inflows Gone Wild: Gross Capital Inflows, Credit Booms and Crises.” World Bank Policy Research Working Paper, 6270. - Calvo, Guillermo, Leonardo Leiderman, and Carmen M. Reinhart. 1993. “Capital Inflows and Real Exchange Rate Appreciation in Latin America: The Role of External Factors.” IMF Staff Papers, 40(1): 108-151. - Calvo, Guillermo, Leonardo Leiderman, and Carmen M. Reinhart. 1996. “Inflows of Capital to Developing Countries in the 1990s.” Journal of Economic Perspecives, 10(2): 123-139. - Cameron, A. Colin, Jonah B. Gelbach, and Douglas L. Miller. 2008. “Bootstrap-based Improvements for Inference with Clustered Errors.” Review of Economics and Statistics, 90: 414–427. - Cerdeiro, Diego A. and Andras Komaromi. 2019. “Financial Openness and Capital Inflows to Emerging Markets: In Search of Robust Evidence.” IMF Working Paper, 19/194. - Cecchetti, Stephen, Tommaso Mancini-Griffoli, Machiko Narita, and Ratna Sahay. 2020. “US or Domestic Monetary Policy: Which Matters More for Financial Stability?” IMF Economic Review, 68: 35-65. https://doi.org/10.1057/s41308-020-00108-2 - Cecchetti, Stephen, Machiko Narita, Umang Rawat, and Ratna Sahay. 2021. “Addressing Spillovers from Prolonged U.S. Monetary Policy Easing.” IMF Working Paper, 21/182. - Cerutti, Eugenio, Catherine Koch, and Swapan-Kumar Pradhan. 2018. “The Growing Footprint of EME Banks in the International Banking System.” BIS Quarterly Review, December 2018. - Chudik, Alexander, M. Hashem Pesaran, and Jui-Chung Yang. 2018. “Half-Panel Jackknife Fixed-Effects Estimation of Linear Panels with Weakly Exogenous Regressors.” Journal of Applied Econometrics, 33(6): 816-836. https://doi.org/10.1002/jae.2623 - Correia, Sergio. 2014. “REGHDFE: Stata Module to Perform Linear or Instrumental-Variable Regression Absorbing Any Number of High-Dimensional Fixed Effects.” Statistical Software Components S457874, Boston College Department of Economics, revised 18 Nov 2019. - Correia, Sergio. 2017. “Linear Models with High-Dimensional Fixed Effect s: An Efficient and Feasible Estimator.” Mimeo. https://github.com/sergiocorreia/reghdfe/ - Dahlberg, Stefan, Aksel Sundström, Sören Holmberg, Bo Rothstein, Natalia Alvarado Pachon, and Cem Mert Dalli. 2021. The Quality of Government Basic Dataset, version Jan21. University of Gothenburg: The Quality of Government Institute. - Dhaene, Geert, and Koen Jochmans. 2015. “Split-Panel Jackknife Estimation of Fixed-Effect Models.” The Review of Economic Studies, 82(3): 991-1030. - Driscoll, John C., and Aart C. Kraay. 1998. “Consistent Covariance Matrix Estimation with Spatially Dependent Panel Data.” Review of Economics and Statistics, 80(4): 549-560. - Eberhardt, Markus, and Andrea Presbitero. 2021. “Commodity prices and banking crises.” Journal of International Economics, 131: 1034-1074. - Forbes, Kristin J., and Francis E. Warnock. 2012. “Capital Flow Waves: Surges, Stops, Flight, and Retrenchment.” Journal of International Economics, 88(2): 235-51. - Ghosh, Atish R., Jonathan D. Ostry, and Mahvash S. Qureshi. 2016. “When Do Capital Inflow Surges End in Tears?” American Economic Review, 106(5): 581-585. - Ghosh, Atish R., Jonathan D. Ostry, and Mahvash S. Qureshi. 2017. “Managing the Tide: How Do Emerging Markets Respond to Capital Flows?” IMF Working Paper, 17/69. - Glynn, Adam N., and Kevin M. Quinn. 2010. “An Introduction to the Augmented Inverse Propensity Weighted Estimator.” Political Analysis, 18: 36-56. - Gori, Filippo, Bin Grace Li, and Andrea F. Presbitero. 2015. “Capital Flows and Financial Deepening in Developing Economies.” Chapter 1, Box 1.3 in October 2015 World Economic Outlook, Washington: International Monetary Fund. - Gruss, Bertrand, and Suhaib Kebhaj. 2019. “Commodity Terms of Trade: A New Database.” IMF Working Paper, 19/21. - Gruss, Bertrand, Malhar Nabar, and Marcos Poplawski-Ribeiro. 2017. “Roads Less Traveled: Growth in Emerging Market and Developing Economies in a Complicated External Environment.” Chapter 2 in April 2017 World Economic Outlook, Washington: International Monetary Fund. - Harrison, Ann, Inessa Love, and Margaret McMillan. 2004. “Global capital flows and financing constraints.” Journal of Development Economics, 74(1): 269-301. - Hayakawa, Kazuhiko. 2009. “A Simple Efficient Instrumental Variable Estimator for Panel AR(p) Models When Both N and T are Large.” Econometric Theory, 25: 873-90. - Horn, Sebastian, Carmen M. Reinhart, and Christoph Trebesch. 2019. “China's Overseas Lending.” NBER Working Paper, 26050. - Jahan, Sarwat, and Daili Wang. 2016. “Capital Account Openness in Low-Income Developing Countries: Evidence from a New Database.” IMF Working Paper, 16/252. - Jordà, Òscar, Moritz Schularick, and Alan M. Taylor. 2011. “Financial Crises, Credit Booms, and External Imbalances: 140 Years of Lessons.” IMF Economic Review, 59 (2): 340-378. - Jordà, Òscar, and Alan M. Taylor. 2016. “The Time for Austerity: Estimating the Average Treatment Effect of Fiscal Policy.” The Economic Journal, 126(590): 219-255. - Kang, D. Y. Joseph, and Joseph L. Schafer. 2007. “Demystifying Double Robustness: A Comparison of Alternative Strategies for Estimating a Population Mean from Incomplete Data.” Statistical Science, 22(4): 523-539. - Khan, Shakeeb, and Elie Tamer. 2010. “Irregular Identification, Support Conditions, and Inverse Weight Estimation.” Econometrica, 78(6): 2021-2042. - Kim, Kyoo il. 2019. “Efficiency of Average Treatment Effect Estimation When the True Propensity is Parametric.” Econometrics, 7, 25. https://doi.org/10.3390/econometrics7020025. - Korinek, Anton. 2018. “Regulating Capital Flows to Emerging Markets: An Externality View.” Journal of International Economics, 111: 61–80. - Kuvshinov, Dmitry, and Kaspar Zimmermann. 2019. “Sovereigns going bust: Estimating the cost of default,” European Economic Review, 119: 1–21. - Laeven, Luc, and Fabian Valencia. 2020. “Systemic Banking Crises Database II.” IMF Economic Review, 68: 307-361. https://doi.org/10.1057/s41308-020-00107-3 - Lane, Philip R. 2015. “International Fiancial Flows in Low-Income Countries.” Pacific Economic Review, 20(1): 49-72. - Lane, Philip R., and Peter Mcquade. 2014. “Domestic Credit Growth and International Capital Flows.” Scandinavian Journal of Economics, 116(1): 218-252. - Machado, José A.F., and Joao M.C. Santos Silva. 2019. “Quantiles via Moments.” Journal of Econometrics, 213(1): 145-173. - MacKinnon, James G., and H. White, 1985. “Some Heteroskedasticity Consistent Covariance Matrix Estimators with Improved Finite Sample Properties.” Journal of Econometrics, 29: 305-325. - MacKinnon, James G., Matthew D. Webb. 2017. “Wild Bootstrap Inference for Wildly Different Cluster Sizes.” Journal of Applied Econometrics, 32: 233-254. - Magud, Nicolas, Carmen M. Reinhart, and Esteban R. Vesperoni. 2014. “Capital Inflows, Exchange Rate Flexibility, and Credit Booms.” Review of Development Economics, 18(3): 415-430. - Mendoza, Enrique G., and Marco E. Terrones. 2008. “An Anatomy of Credit Booms: Evidence from Macro Aggregates and Micro Data.” NBER Working Paper, 14049. - Meng, Channarith, and Roberto Leon Gonzalez. 2017. “Credit Booms in Developing Countries: Are They Different from Those in Advanced and Emerging Market Countries?” Open Economies Review, 28: 547–579. - Miranda-Agrippino, Silvia, and Hélène Rey. 2020. “US Monetary Policy and the Global Financial Cycle.” Review of Economic Studies, 87(6): 2754-2776. - Montiel Olea, José Luis, and Carolin Pflueger. 2013. “A Robust Test for Weak Instruments.” Journal of Business and Economic Statistics, 31: 358-369. - Morais, Bernardo, José-Luis Peydró, Jessica Roldán-Pena, and Claudia Ruiz. 2019. “The International Bank Channel of Monetary Policy Rates and QE: Credit Supply, Reach-for-Yield, and Real Effects.” Journal of Finance, 74(1): 55-90. - Naimi, Ashley I., Erica E.M. Moodie, Nathalie Auger, and Jay S. Kaufman. 2014. “Constructing Inverse Probability Weights for Continuous Exposures: A Comparison of Methods.” Epidemiology, 25(2): 292-299. - Nickell, Stephen. 1981. “Biases in Dynamic Models with Fixed Effects.” Econometrica, 49(6): 1417-1426. - Nier, Erlend W., T. Tjoervi Olafsson, and Yan G. Rollinson. 2020. “Exchange Rates, Domestic Credit, and Macroprudential Policy.” IMF Working Paper, 20/187. - Puy, Damien. 2016. “Mutual Funds Flows and the Geography of Contagion.” Journal of International Money and Finance, 60: 73-93. - Pflueger, Carolin E. and Su Wang. 2015. “A Robust Test for Weak Instruments in Stata.” The Stata Journal, 15(1): 216-225. - Presbitero, Andrea, Dhaneshwar Ghura, Olumuyiwa S. Adedeji, and Lamin Njie. 2016. “Sovereign Bonds in Developing Countries: Drivers of Issuance and Spreads.” Review of Development Finance, 6(1), 1-15. - Reinhart, Carmen M., and Vincent R. Reinhart. 2008. “Capital Flow Bonanzas: An Encompassing View of the Past and Present.” NBER International Seminar on Macroeconomics, 5(1): 9-62. - Reinhart, Carmen M., Vincent Reinhart, and Christoph Trebesch. 2016. “Global Cycles: Capital Flows, Commodities, and Sovereign Defaults, 1815-2015.” American Economic Review, 106(5): 574-580. - Rey, Hélène. 2016. “International Channels of Transmission of Monetary Policy and the Mundellian Trilemma.” IMF Economic Review, 64(1): 6-35. - Richter, Bjorn, Moritz Schularick, and Ilhyock Shim. 2019. “The Costs of Macroprudential Policy.” Journal of International Economics, 118: 263-282. - Robins, M. James, Andrea Rotnitzky, and Lue Ping Zhao. 1994. “Estimation of Regression Coefficients When Some Regressors are not Always Observed.” Journal of the American Statistical Association, 89(427): 846-866. - Roodman, D. 2009. “How to Do xtabond2: An Introduction to "Difference" and "System" GMM in Stata.” Stata Journal, 9(1): 86-136. - Schlicht, Ekkehart. 2008. “Trend Extraction from Time Series with Structural Breaks and Missing Observations.” Journal of the Japan Statistical Society, 38(2): 285-292. - Schularick, Moritz, and Alan M. Taylor. 2012. “Credit Booms Gone Bust: Monetary Policy, Leverage Cycles, and Financial Crises, 1870-2008.” American Economic Review, 102(2): 1029-1061. - Stock, J. H. and Motohiro Yogo. 2005. “Testing for Weak Instruments in Linear IV Regression.” Chapter 5, 80-108, in Identification and Inference for Econometric Models: Essays in Honor of Thomas Rothenberg, edited by Ronald W. K. Andrews and James H. Stock. Cambridge: Cambridge University Press. - Sun, Yutao, and Geert Dhaene. 2019. “XTSPJ: A Command for Split-Panel Jackknife Estimation.” Stata Journal, 19(2): 335-374. https://doi.org/10.1177/1536867X19854016 - Taylor, John B. 2019. “Capital Flows, the IMF’s Institutional View, and an Alternative.” Chapter Five, 105-116, in Currencies, Capital, and Central Bank Balances, edited by Cochrane, Palermo, and Taylor. Stanford: Hoover Institution Press. - Qin, Jing, Biao Zhang, and Denis H. Y. Leung. 2017. “Efficient Augmented Inverse Probability Weighted Estimation in Missing Data Problems,” Journal of Business and Economic Statistics, 35(1): 86-97. https://doi.org/10.1080/07350015.2015.1058266 - Wooldridge, Jeffrey M. 2010. Econometric Analysis of Cross Section and Panel Data, Second Edition. Cambridge: MIT Press. - Yamada, Hiroshi. 2021. “Trend Extraction from Economic Time Series with Missing Observations by Generalized Hodrick-Prescott Filters.” Econometric Theory, 1-35. https://www.doi.org/10.1017/S0266466621000189 - Young, Alwyn. 2019. “Consistency without Inference: Instrumental Variables in Practical Application.” Mimeo. - Zhao, Shandong, David A. van Dyk, and Kousuke Imai. 2020. “Propensity score-based methods for causal inference in observational studies with non-binary treatments.” Statistical Methods in Medical Research, 29(3): 709-727. ### Datasets, indexes, and institutional publications - Baum, Christopher, Mark E. Schaffer, and Steven Stillman. 2010. “IVREG2: Stata Module for Extended Instrumental Variables/2SLS, GMM and AC/HAC, LIML and k-class Regression.” https://ideas.repec.org/c/boc/bocode/s425401.html - Chicago Board Options Exchange. 2021. Cboe Volatility Index: VIX [VIXCLS], retrieved from FRED, Federal Reserve Bank of St. Louis; https://fred.stlouisfed.org/series/VIXCLS, accessed in 2021. - Fitch Solutions. 2020. Fitch Fundamental Financial Data, accessed in February 2020. - International Monetary Fund. 2014. “Macroeconomic Developments in Low-Income Developing Countries: 2014 Report.” Policy Paper, Washington: International Monetary Fund. - International Monetary Fund. 2015. “Macroeconomic Developments in Low-Income Developing Countries.” Policy Paper, Washington: International Monetary Fund. - International Monetary Fund. 2018. “Macroeconomic Developments in Low-Income Developing Countries—2018.” Policy Paper, March, Washington: International Monetary Fund. - International Monetary Fund. 2019a. Annual Report on Exchange Arrangements and Exchange Restrictions. Washington: International Monetary Fund. - International Monetary Fund. 2019b. “A Strategy For IMF Engagement on Social Spending.” Policy Paper, Washington: International Monetary Fund. - International Monetary Fund. 2020. “Toward an Integrated Policy Framework.” Policy Paper, Washington: International Monetary Fund. - International Monetary Fund. 2021a. Financial Flows Analytics, confidential and internally accessed in June 2021, Washington: International Monetary Fund. - International Monetary Fund. 2021b. International Financial Statistics, accessed in June 2021, Washington: International Monetary Fund. - International Monetary Fund. 2021c. Statistical Appendix of the April 2021 World Economic Outlook, Washington: International Monetary Fund. - PRS Group. 2021. International Country Risk Guide (ICRG) Dataset. New York: The PRS Group. ### Software, commands, and methodological references - Baum, Christopher, Mark E. Schaffer, and Steven Stillman. 2010. “IVREG2: Stata Module for Extended Instrumental Variables/2SLS, GMM and AC/HAC, LIML and k-class Regression.” https://ideas.repec.org/c/boc/bocode/s425401.html - Correia, Sergio. 2014. “REGHDFE: Stata Module to Perform Linear or Instrumental-Variable Regression Absorbing Any Number of High-Dimensional Fixed Effects.” Statistical Software Components S457874, Boston College Department of Economics, revised 18 Nov 2019. - Correia, Sergio. 2017. “Linear Models with High-Dimensional Fixed Effects: An Efficient and Feasible Estimator.” Mimeo. https://github.com/sergiocorreia/reghdfe/ - Montiel Olea, José Luis, and Carolin Pflueger. 2013. “A Robust Test for Weak Instruments.” Journal of Business and Economic Statistics, 31: 358-369. - Pflueger, Carolin E. and Su Wang. 2015. “A Robust Test for Weak Instruments in Stata.” The Stata Journal, 15(1): 216-225. - Roodman, D. 2009. “How to Do xtabond2: An Introduction to "Difference" and "System" GMM in Stata.” Stata Journal, 9(1): 86-136. - Sun, Yutao, and Geert Dhaene. 2019. “XTSPJ: A Command for Split-Panel Jackknife Estimation.” Stata Journal, 19(2): 335-374. https://doi.org/10.1177/1536867X19854016 *Reference list as provided in the source PDF wpiea2021237-print-pdf - REFERENCES.* ### APPENDIX I. DATA DESCRIPTIONS ### APPENDIX I. DATA DESCRIPTIONS ### A. Country groups and variable definitions - Groupings of economies: - Low-income developing countries (LIDCs; 59): Afghanistan, Bangladesh, Benin, Bhutan, Burkina Faso, Burundi, Cambodia, Cameroon, Central African Republic, Chad, Comoros, Democratic Republic of the Congo, Republic of the Congo, Côte d'Ivoire, Djibouti, Eritrea, Ethiopia, The Gambia, Ghana, Guinea, Guinea-Bissau, Haiti, Honduras, Kenya, Kiribati, Kyrgyz Republic, Lao P.D.R., Lesotho, Liberia, Madagascar, Malawi, Mali, Mauritania, Moldova, Mozambique, Myanmar, Nepal, Nicaragua, Niger, Nigeria, Papua New Guinea, Rwanda, São Tomé and Príncipe, Senegal, Sierra Leone, Solomon Islands, Somalia, South Sudan, Sudan, Tajikistan, Tanzania, Timor-Leste, Togo, Uganda, Uzbekistan, Vietnam, Yemen, Zambia, Zimbabwe. - Emerging market economies (EMEs; 95): Albania, Algeria, Angola, Antigua and Barbuda, Argentina, Armenia, Azerbaijan, The Bahamas, Bahrain, Barbados, Belarus, Belize, Bolivia, Bosnia and Herzegovina, Botswana, Brazil, Brunei Darussalam, Bulgaria, Cabo Verde, Chile, China, Colombia, Costa Rica, Croatia, Dominica, Dominican Republic, Ecuador, Egypt, El Salvador, Equatorial Guinea, Eswatini, Fiji, Gabon, Georgia, Grenada, Guatemala, Guyana, Hungary, India, Indonesia, Iran, Iraq, Jamaica, Jordan, Kazakhstan, Kosovo, Kuwait, Lebanon, Libya, Malaysia, Maldives, Marshall Islands, Mauritius, Mexico, Micronesia, Mongolia, Montenegro, Morocco, Namibia, Nauru, North Macedonia, Oman, Pakistan, Palau, Panama, Paraguay, Peru, Philippines, Poland, Qatar, Romania, Russia, Samoa, Saudi Arabia, Serbia, Seychelles, South Africa, Sri Lanka, St. Kitts and Nevis, St. Lucia, St. Vincent and the Grenadines, Suriname, Syria, Thailand, Tonga, Trinidad and Tobago, Tunisia, Turkey, Turkmenistan, Tuvalu, Ukraine, United Arab Emirates, Uruguay, Vanuatu, Venezuela. - Advanced economies (AEs; 39): Australia, Austria, Belgium, Canada, Cyprus, Czech Republic, Denmark, Estonia, Finland, France, Germany, Greece, Hong Kong S.A.R. of China, Iceland, Ireland, Israel, Italy, Japan, Korea, Latvia, Lithuania, Luxembourg, Macao S.A.R. of China, Malta, Netherlands, New Zealand, Norway, Portugal, Puerto Rico, San Marino, Singapore, Slovak Republic, Slovenia, Spain, Sweden, Switzerland, Taiwan Province of China, United Kingdom, United States. - Note on EMEs: "EMEs are defined as the residual group of economies that are not included in AEs nor LIDCs." - Bank sample coverage (Panel B — countries with number of banks in our sample; 32 LIDCs; Total 584 banks): - Examples of country-level bank counts (Country: Number of banks): Bangladesh 55; Burkina Faso 9; Cameroon 12; Democratic Republic of the Congo 16; Republic of the Congo 3; Côte d'Ivoire 14; Ethiopia 2; The Gambia 5; Ghana 50; Guinea 3; Guinea-Bissau 1; Honduras 31; Kenya 63; Liberia 5; Madagascar 8; Malawi 12; Mali 10; Moldova 17; Mozambique 16; Myanmar 1; Nicaragua 16; Niger 6; Nigeria 41; Papua New Guinea 1; Senegal 16; Sierra Leone 9; Tanzania 39; Togo 10; Uganda 29; Vietnam 54; Yemen 8; Zambia 22. Total 584. - Variable definitions and data sources (selected highlights from Table A2): - Loans: Transformation = Natural logarithm; Series code = Gross Loans; Database = Fitch Fundamental Financial Data. - Wholesale funding: Natural logarithm; Series code = Wholesale Funding; Database = Fitch Fundamental Financial Data. - Deposits: Natural logarithm; Series code = Total Deposits; Database = Fitch Fundamental Financial Data. - Loan-to-deposit ratio: Winsorized at 5 percent; Series = Net Loans / Total Deposits; Database = Fitch Fundamental Financial Data. - Regulatory capital ratio: Winsorized at 5 percent; Series = Total Regulatory Capital Ratio; Database = Fitch Fundamental Financial Data. - Total asset to equity ratio: Winsorized at 5 percent; Series = 100 / (Equity / Total Assets); Database = Fitch Fundamental Financial Data. - Nonperforming loans: Winsorized at 5 percent; Series = Impaired Loans (NPLs) / Gross Loans; Database = Fitch Fundamental Financial Data. - Return on assets: Winsorized at 5 percent; Series = Operating ROAA; Database = Fitch Fundamental Financial Data. - Net interest margin: Winsorized at 5 percent; Series = Net Interest Margin; Database = Fitch Fundamental Financial Data. - Liquid asset to short-term funding ratio: Winsorized at 5 percent; Series = Liquid Assets / Deposits and ST Funding; Database = Fitch Fundamental Financial Data. - Personnel expenses per overhead: Natural logarithm; Series = Personnel Expenses / Overheads; Database = Fitch Fundamental Financial Data. - Number of employees: Natural logarithm; Series = Number of Employees; Database = Fitch Fundamental Financial Data. - Total capital inflows: Natural logarithm; Series code = ICAPFL; Database = FFA. - Private capital inflows: Natural logarithm; Series code = ICAPFLP; Database = FFA. - Foreign direct investment (FDI) inflows: Natural logarithm; Series code = IFDI; Database = FFA. - Total inflows to the region: Natural logarithm; Series = The sum of ICAPFL for the countries in the same regions, excluding the country itself; Database = FFA. - Real GDP, level: Natural logarithm; Series code = NGDP_R; Database = WEO. - Domestic interest rate: None; Combined sequence of five interest rates from IFS when not available: FIMM_PA, FID_PA, FPOLM_PA, FIDR_PA, FITB_PA. - Banking crisis dummy: None; Source = Banking Laeven and Valencia (2020). - Currency crisis dummy: None; Source = Currency Laeven and Valencia (2020). - Real effective exchange rate (REER): Percent change; Series code = EREER_IX; Database = IFS. - Wang-Jahan capital account openness index: None; Series code = WJ_ka_new; Source = Jahan and Wang (2016), updated by the authors. - ICRG, quality of government (QoG) index: None; Series code = icrg_qog; Source = Dahlberg and others (2021), PRS Group, Inc., International Country Risk Guide (ICRG). - Trading partners’ growth: Percent change; Series code = NGDP_R_WX001; Database = WEO (GEE). - Export commodity prices: Natural logarithm; Series code = x_gdp; Source = Gruss and Kebhaj (2019). - Import commodity prices: Natural logarithm; Series code = m_gdp; Source = Gruss and Kebhaj (2019). - Volatility Index (VIX): Natural logarithm; Series code = The Cboe Volatility Index; Source = Chicago Board Options Exchange (Cboe, 2021). - U.S. policy rate: None; Series code = 111FPOLM_PA; Database = IFS. - U.S. deflator: Percent change; Series code = 111NGDP_D; Database = WEO. - Notes on transformations and databases: - "Note. Percent changes are calculated as a difference in natural logarithm of the level." - The FFA database is compiled from IMF’s Balance of Payments Statistics, IFS, and WEO databases, World Bank’s WDI database, Haver Analytics, CEIC Asia database, and CEIC China database. - Abbreviations: GEE: Global Economic Environment; LIDCs: low-income developing countries. ### Summary statistics for LIDCs (Table A3 — selected statistics) - Sample period: 1989-2019. - Bank-level variables (percentiles, mean, S.D., observations): - Loan: 25th = 3.78; 50th = 4.89; 75th = 6.01; Mean = 4.91; S.D. = 1.69; Within S.D. = 0.79; Observations = 5,279; # of Banks = 584; Average T = 9. - Wholesale funding: 25th = 1.71; 50th = 3.09; 75th = 4.41; Mean = 2.99; S.D. = 2.25; Within S.D. = 1.29; Observations = 4,293; # of Banks = 559; Average T = 8. - Deposits: 25th = 4.19; 50th = 5.27; 75th = 6.38; Mean = 5.26; S.D. = 1.68; Within S.D. = 0.76; Observations = 5,279; # of Banks = 584; Average T = 9. - Loan-to-deposit ratio: 25th = 53.21; 50th = 68.41; 75th = 82.71; Mean = 72.52; S.D. = 35.17; Within S.D. = 17.15; Observations = 5,279; # of Banks = 584; Average T = 9. - Regulatory capital ratio: 25th = 13.00; 50th = 18.24; 75th = 26.00; Mean = 22.11; S.D. = 12.97; Within S.D. = 6.61; Observations = 2,541; # of Banks = 364; Average T = 7. - Return on assets: 25th = 0.89; 50th = 2.22; 75th = 3.94; Mean = 2.46; S.D. = 2.62; Within S.D. = 1.66; Observations = 5,171; # of Banks = 583; Average T = 9. - Liquid asset to short-term funding ratio: 25th = 18.58; 50th = 30.07; 75th = 47.31; Mean = 36.13; S.D. = 23.12; Within S.D. = 14.47; Observations = 5,278; # of Banks = 584; Average T = 9. - Personnel expenses per overhead (log): 25th = 3.63; 50th = 3.81; 75th = 3.96; Mean = 3.77; S.D. = 0.33; Within S.D. = 0.21; Observations = 4,690; # of Banks = 552; Average T = 8. - Country-level variables (percentiles, mean, S.D., observations): - Total capital inflows: 25th = -1.24; 50th = -0.32; 75th = 0.60; Mean = -0.32; S.D. = 1.49; Within S.D. = 0.98; Observations = 481; # of Countries = 32; Average T = 15. - Private capital inflows: 25th = -1.19; 50th = -0.35; 75th = 0.59; Mean = -0.31; S.D. = 1.47; Within S.D. = 0.97; Observations = 474; # of Countries = 32; Average T = 15. - FDI inflows: 25th = -2.24; 50th = -1.02; 75th = 0.05; Mean = -1.17; S.D. = 1.80; Within S.D. = 1.28; Observations = 458; # of Countries = 32; Average T = 14. - Total capital inflows to the region: 25th = 2.93; 50th = 4.20; 75th = 4.70; Mean = 3.96; S.D. = 1.16; Within S.D. = 0.86; Observations = 477; # of Countries = 32; Average T = 15. - Real GDP, level: 25th = 5.29; 50th = 8.27; 75th = 9.66; Mean = 7.84; S.D. = 2.96; Within S.D. = 0.33; Observations = 481; # of Countries = 32; Average T = 15. - Real GDP, growth: 25th = 0.03; 50th = 0.05; 75th = 0.06; Mean = 0.04; S.D. = 0.04; Within S.D. = 0.04; Observations = 481; # of Countries = 32; Average T = 15. - Domestic interest rate: 25th = 0.05; 50th = 0.09; 75th = 0.14; Mean = 0.11; S.D. = 0.09; Within S.D. = 0.06; Observations = 474; # of Countries = 32; Average T = 15. - Banking crisis dummy: 25th = 0.00; 50th = 0.00; 75th = 0.00; Mean = 0.01; S.D. = 0.10; Within S.D. = 0.10; Observations = 481; # of Countries = 32; Average T = 15. - Currency crisis dummy: 25th = 0.00; 50th = 0.00; 75th = 0.00; Mean = 0.04; S.D. = 0.19; Within S.D. = 0.18; Observations = 481; # of Countries = 32; Average T = 15. - REER: 25th = -0.04; 50th = 0.01; 75th = 0.04; Mean = -0.00; S.D. = 0.12; Within S.D. = 0.12; Observations = 481; # of Countries = 32; Average T = 15. - Wang-Jahan capital account openness index: 25th = 0.13; 50th = 0.33; 75th = 0.69; Mean = 0.41; S.D. = 0.33; Within S.D. = 0.10; Observations = 337; # of Countries = 29; Average T = 12. - ICRG, quality of government index: 25th = 0.33; 50th = 0.40; 75th = 0.47; Mean = 0.40; S.D. = 0.10; Within S.D. = 0.05; Observations = 481; # of Countries = 32; Average T = 15. - VIX: 25th = 2.66; 50th = 2.86; 75th = 3.19; Mean = 2.93; S.D. = 0.31; Within S.D. = 0.30; Observations = 480; # of Countries = 32; Average T = 15. - U.S. policy rate: 25th = 0.13; 50th = 1.25; 75th = 4.25; Mean = 1.94; S.D. = 2.09; Within S.D. = 1.95; Observations = 481; # of Countries = 32; Average T = 15. - U.S. deflator, growth: 25th = 0.01; 50th = 0.02; 75th = 0.02; Mean = 0.02; S.D. = 0.01; Within S.D. = 0.01; Observations = 481; # of Countries = 32; Average T = 15. - Sources for Table A3: Cboe (2021); Financial Flows Analytics (FFA, IMF, 2021a); Fitch Fundamental Financial Data (Fitch Solutions, 2020); Gruss and Kebhaj (2019); International Financial Statistics (IFS, IMF, 2021b); Jahan and Wang (2016); Laeven and Valencia (2020); Quality of Government Basic Dataset (Dahlberg and others, 2021; PRS Group, Inc., 2021); World Economic Outlook (WEO, IMF, 2021c). ### B. Update of the Wang-Jahan index - Concept and construction: - The Wang-Jahan index of capital account openness uses AREAER coding across 12 subcategories; for each subcategory a dummy equals 0 or 1 indicating presence of restrictions. - The main indicator is the simple average across 12 subcategories, ranging from 0 (fully closed) to 1 (unrestricted in all 12 subcategories). - The 12 subcategories: transactions in shares or other securities; bonds and other debt; money market instruments; collective investment securities; financial credits; direct investment and its liquidation; derivatives; real estate; financial guarantees; commercial credit. - Adjacent text information, if available, is used to verify and can overrule the dummy in rare contradictions. Intensity of control is not captured. - Update performed: - Methodology follows Jahan and Wang (2016). - Original dataset covered 1996–2013; coverage extended to 2017. ### C. Bank-level data construction - Data source and selection: - Bank-level balance sheet data extracted from financial statements in Fitch Fundamental Financial Data. - When multiple financial statements exist for an institution-year, the preferred record is chosen in this order (if applicable): 1. End-year report instead of quarterly/interim reports. 2. The one that covers 12 months. 3. The one that is not consolidated. 4. The one in international accounting standards. 5. The one that is audited. 6. The one reported as of end-December. - Identification and filtering: - Financial institutions selected whose market sector level 3 description is "Banks" or whose issuer name includes "bank" or its variants using a regular expression search: "[B|b]ank|BANK|[B|b]anco|BANCO|[B|b]anque|BANQUE|[B|b]anca|BANCA.” (to capture language differences). - Central banks and bank holding companies are excluded. - Sample further restricted to banks in LIDCs with positive entries for gross loans and deposits. *APPENDIX I. DATA DESCRIPTIONS (wpiea2021237-print-pdf)* ### APPENDIX II. ROBUSTNESS CHECK ### APPENDIX II. ROBUSTNESS CHECK ### A. Nickell bias - Issue: Inclusion of the lagged dependent variable with bank-level fixed effects generates Nickell bias, which biases the coefficient on the lagged dependent variable toward zero; bias on other coefficients depends on their correlation with the lagged dependent variable. - Time-dimension mitigation: Nickell bias is proportional to the inverse of the length of the time dimension; sample has average length of 9 years and minimum of 4 years, implying the bias would not be too large. - Panel GMM approach: - Method: Generalized method of moments (GMM) using lagged variables as instruments, following Arellano and Bond (1991), Arellano and Bover (1995), and Blundell and Bond (1998). - Specification choice: System GMM rejected due to AR(1) test invalidating lagged instruments for the level equation; focus on one-equation specification where AR(2) test supports validity of instruments for all cases. - Result summary: Results broadly similar to baseline OLS, with weaker evidence on deposits. On loans, coefficient on capital flows becomes slightly larger than OLS, indicating some correction of Nickell bias toward zero; significance lost for wholesale funding and deposits. Loan-to-deposit ratio shows weak evidence of increase when capital flows come in. Results are not very robust to alternative panel GMM settings. - Split-panel jackknife bias correction: - Methods applied: Half-panel jackknifed likelihood estimator (Dhaene and Jockmans, 2015) implemented per Sun and Dhaene (2019) xtspj with model(regress) method(likelihood) and handling of unbalanced panels per Chudik, Pesaran, and Yang (2018); Driscoll-Kraay standard errors with bandwidth of 4 used. - Practical shortcuts: For unbalanced panels with odd number of observations, choose the division that gives the former half sample one more observation (example: 9 observations split into 2001-2005 and 2006-2009). For standard errors, ignore cross-equation correlations. - Key quantitative findings: - Baseline result (Table 1, column 3) of 6.7 basis points only slightly changes to 7.5 basis points under the split-panel jackknife likelihood estimator, but the point estimate is no longer statistically significant. - Estimated size of the Nickell bias is 12 percent of the size of the coefficient; estimation precision declines when correcting for Nickell bias. - Table A5 summary (coefficients of total gross capital flows; significance indicated): - Baseline (Table 1, col. 3, Table 3): Gross loans = 0.055***; Wholesale funding = 0.073**; Deposits = 0.047***; Loan-to-deposit ratio = 0.739**. - Half-panel jackknifed likelihood: Gross loans = 0.057; Wholesale funding = 0.087; Deposits = 0.054; Loan-to-deposit ratio = 0.543. - Half-panel jackknifed likelihood, excluding observations with any data gaps: Gross loans = 0.061; Wholesale funding = 0.083; Deposits = 0.078; Loan-to-deposit ratio = 0.709. - Half-panel parameter jackknife: Gross loans = 0.061*; Wholesale funding = 0.121*; Deposits = 0.058**; Loan-to-deposit ratio = 0.322*. - Robustness to sample length: Restricting sample to banks with at least four (or eight) observations reduces sample to 5,020 (or 4,125) observations with 480 (or 313) banks in 29 (or 26) countries for main regression with gross loans; results with both 4-year and 8-year restrictions are broadly similar to baseline. - Overall conclusion: Robustness checks indicate a non-negligible influence of Nickell bias; primary challenge is limited availability of long bank-level time series in LIDCs. ### B. Small number of clusters - Concern: Cluster-robust standard errors, including Driscoll and Kraay (1998) standard errors used here, rely on asymptotics where number of clusters goes to infinity. - Clustering in this study: Driscoll-Kraay standard error clusters by year; number of clusters equals number of sample periods (years), which is 31 in the unbalanced panel — possibly insufficient. - Rule-of-thumb context: MacKinnon and Webb (2017) note that even rule-of-thumb of 42 clusters may be insufficient for unbalanced panels. - Consequences of few clusters: - Small number of clusters can lead to over-rejection and spurious significance due to highly leveraged observations producing mechanically small standard errors (Young, 2019). - At first glance, few clusters might suggest large standard errors, but leverage effects can instead produce too-small standard errors and over-rejection. - The root cause relates to bias in heteroskedasticity-robust standard errors depending on observation leverages (MacKinnon and White, 1985). - Bootstrap status: While coefficient-based bootstrap methods have been advocated for cluster-robust standard errors (Young, 2019), consensus has not formed for Driscoll-Kraay standard errors and simulation evidence for bootstrap performance in this context remains limited. ### C. Capital flow surges and non-average responses - Motivation: Examine effects of capital-flow “bonanzas” or “surges” to test nonlinear effects or sharpen identification. - Surge identification: Follow Caballero (2016) method to identify 10 capital-inflow surge episodes in estimation sample that includes 481 country-year observations. A surge is identified if the HP-filtered gap of capital inflows is higher than two standard errors calculated country by country. Use HP filter allowing for missing observations (Schlicht, 2008; Yamada, 2021). - Main findings: - Capital-flow surges are associated with results similar to those based on log levels of capital inflows; baseline surge specification produces broadly similar results (Tables A6, A7). - Results are not robust across different methods to identify surges (Table A8). - Selected quantitative highlights from Tables: - Table A6 (Gross loans and capital-flow bonanzas; sample 1989-2019, Observations = 5,279): - Total inflows bonanza – two S.D.: coefficients reported across specifications include 0.04, 0.08*, 0.07**, 0.12*** with standard errors in parentheses. - Lagged gross loans (log), lag: 0.69***, 0.57***, 0.58*** (standard errors reported as 0.02, 0.03, 0.03). - Sample period 1989-2019 (one specification 1991-2019); estimation sample includes 32 countries and 584 banks. - Table A7 (Banks’ soundness and business dynamics upon bonanzas/surges): - Total inflows bonanza – two S.D. coefficients across dependent variables: Non-performing loans = 0.37; Return on assets = -0.14; Net interest margin = 0.04*; Liquid asset to short-term funding ratio = 2.51***; Personnel expenses per overhead = -0.04; Number of employees = 0.09** (standard errors in parentheses). - Observations and samples vary by outcome; number of banks ranges from 345 to 584; sample periods vary 1989-2019 to 1994-2019. - Table A8 (Comparison across definitions of bonanzas/surges; sample period 1989-2019, Observations = 5,279): - Alternative definitions produce varying coefficients, e.g., Total inflows bonanza – two S.D. = 0.07** (0.03) in one specification; Total inflows surge – top 30 percent = 0.04* (0.02) in another; several lagged and multi-period surge definitions produce coefficients near zero. - Overall: Bonanza/surge results are directionally similar to log-level results but are sensitive to the surge-identification method. *Source: APPENDIX II. ROBUSTNESS CHECK (wpiea2021237-print-pdf).* ### APPENDIX III. METHODOLOGIES TO ADRESS REVERSE CAUSALITY ### APPENDIX III. METHODOLOGIES TO ADRESS REVERSE CAUSALITY ### A. Consequences of reverse causality - Two-way causality is modeled as a simultaneous system linking bank credit C_{j,i,t} and country capital inflows K_{i,t} with country-level controls X_{i,t}, bank-level controls B_{j,i,t}, bank fixed effects α_j, country fixed effects μ_i, and innovations ε_{j,i,t} and η_{i,t}. Variable Z_{i,t} is key for identification and is assumed correlated with K_{i,t} but not with bank-level shocks ε_{j,i,t} conditional on controls. - In a simplified single-bank, single-country model without controls, the system reduces to: - C_t = β K_t + ε_t - K_t = γ C_t + η_t - Identification requires the additional assumption 1−βγ > 0 because both β and γ are positive (domestic credit growth invites capital inflows and capital inflows lead to domestic credit growth). Without this assumption the two equations are not separately identified. - Assuming ε_t and η_t are i.i.d. with means zero and variances σ_ε^2 and σ_η^2 respectively, the OLS estimator β̂ converges to a mixture of β and γ^{−1}. Formally: - plim_{N→∞} β̂ = γ^{−1} [ (γ^2 σ_ε^2) / (γ^2 σ_ε^2 + σ_η^2) ] + β [ σ_η^2 / (γ^2 σ_ε^2 + σ_η^2) ]. - The OLS bias is positive in this context: - plim_{N→∞} β̂ − β = (γ^{−1} − β) [ γ^2 σ_ε^2 / (γ^2 σ_ε^2 + σ_η^2) ]. - Because γ^{−1} > β under the maintained identification assumption, OLS will overstate the magnitude of β. ### B. Instrumental variable based on regional synchronicity - Identification strategy: use an instrument Z_{i,t} based on regional synchronicity of capital flows following Blanchard, Adler, and de Carvalho Filho (2015). Instrument exogeneity requires E(Z_{i,t} ε_{j,i,t} | X_{i,t}, B_{j,i,t}, α_j) = 0 and strong correlation with K_{i,t} to avoid weak-instrument bias (Stock and Yogo, 2005). - Construction of regional capital inflows instrument K^*_{i,t}: - K^*_{i,t} = Σ_{j ∈ Θ_i, j ≠ i} K_{j,t}, where Θ_i is the set of countries in the same region as country i, excluding reserve currency issuers (euro area, Japan, Switzerland, United Kingdom, United States). - Use U.S. dollar levels and take the natural logarithm in empirical analysis (Blanchard et al. used shares to previous year GDP; this work uses log USD levels to avoid GDP influence). - Region classification follows the World Bank’s regions: 1. East Asia and Pacific 2. Europe and Central Asia 3. Latin America and Caribbean 4. Middle East and North Africa 5. North America 6. South Asia 7. sub-Saharan Africa - Requirements for instrument exogeneity: - Fully control for other channels linking regional capital flows and domestic credit (e.g., foreign demand for domestically produced goods and services, changes in market prices of the country’s import or export goods). - Domestic economic conditions in any given country should not affect capital flows to the region as a whole (satisfied if foreign investors set total regional investment first, then allocate across countries). - Controls included to satisfy exogeneity: country-specific measures of foreign demand (change in real GDP of main trading partners weighted by export shares) and foreign prices (change in prices of commodity exports and imports weighted by export and import volumes). - Empirical strength: the instrument is significantly positively correlated with gross capital inflows (both in logs) for a majority of LIDCs — 35 out of 43 countries in the dataset — at the 1 percent significance level, with correlations ranging from 46 percent to 90 percent. - Validation conditions (sufficient, digestible form): - E(ε_{j1,i1,t} ε_{j2,i2,t}) = 0 (no cross-country correlation in bank-credit innovations after controls). - E(η_{i1,t} ε_{j2,i2,t}) = 0 (no correlation between capital-flow and credit innovations across countries). - E(η_{i1,t} η_{i2,t}) ≠ 0 for i1 ≠ i2 (non-zero cross-country correlation in capital-flow innovations, i.e., regional synchronicity). - Practical challenge: controls must capture all cross-country common factors for C_{j,i,t} while retaining common factors for K_{i,t}. Including year fixed effects risks absorbing cross-country common factors for K_{i,t}, producing a weak instrument and little identifying power. - Caveat on alternative instruments: constructing multiple instruments from different country groups tended to produce highly correlated instruments in this dataset, rendering one instrument insignificant in the first stage and making overidentifying tests unreliable. ### C. Methodological note on the AIPW estimator - Estimator choice: augmented inverse-probability weighting (AIPW) from the treatment-effect literature (Glynn and Quinn, 2010), adopted in macroeconomic contexts by Jordà and Taylor (2016), Alam and others (2019), and others. - AIPW combines: - Inverse-probability weighting (IPW): weighted average of outcomes using inverse estimated propensity scores to emphasize observations less likely to suffer reverse causality. - Regression adjustment (RA): outcome models estimated separately for each treatment level to impute counterfactual outcomes. - Advantages: - Greater parametric flexibility than OLS (OLS is a special case of RA with parameter restrictions). - Double robustness: consistent if either the treatment model or the outcome model is correctly specified. - Semiparametric efficiency among doubly robust estimators when both specifications are correct (asymptotic variance reaches theoretical lower bound). - Estimation of continuous treatment effects: - Use quantile binning to create a categorical treatment variable from continuous capital inflows (country-level distribution). - Separate locations at the 0.15 and 0.85 quantiles, yielding three treatment levels. Choice aims to distinguish very high and very low capital inflows while retaining sufficient observations per category. - Treatment model: ordered logit at the country level. - Outcome models: linear regression for each treatment level. - Covariates for both models: same as baseline OLS regressions, excluding current capital inflows but retaining lag of capital flows. - Standard errors: - Calculated in the final step using the formula valid when point estimates of both treatment and outcome models are consistent. - Although AIPW proceeds in three steps, the one-step joint estimation standard-error formula reduces to the last-step formula if both treatment and outcome models are consistent (robust to weak dependence and heteroscedasticity). - If one model is misspecified, one-step standard errors may be larger or smaller than last-step-only standard errors depending on misspecification degree; RA tends to have larger one-step standard errors, IPW tends to have smaller one-step standard errors. The AIPW combines both properties, and estimation errors from earlier steps vanish if point estimates are consistent. - Implementation choices and robustness: - Quantile binning approach is nonparametric and found robust in simulation studies (Naimi et al., 2014) across a variety of data-generating processes, including non-normal distributions and heteroscedasticity. - The choice of quantiles is critical for identification because it determines comparison groups; results depend on this choice. *Source: APPENDIX III. METHODOLOGIES TO ADRESS REVERSE CAUSALITY, wpiea2021237-print-pdf* --- _Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021237-print-pdf.pdf_