## 3.1 Baseline Setup and Methodology ## Source details **Canonical URL:** [3.1 Baseline Setup and Methodology](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020187-print-pdf.pdf) ## Other formats - [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020187-print-pdf.pdf.md) - [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020187-print-pdf.pdf.json) --- ### Research scope and objectives - Study link between exchange rate movements and domestic credit in a panel of 62 countries over the period 2000 to 2016. - Assess extent to which macroprudential policy can attenuate effects of currency movements on domestic credit cycles. - Evaluate complementary role of targeted controls on inflows when strong credit developments lead to increases in cross-border borrowing. ### Key contributions - Use the credit-to-GDP gap as a continuous indicator of the build-up of systemic risk. - Examine interaction effect of macroprudential policy (MaPP) in mitigating the impact of the exchange rate on domestic credit, improving identification of causal effects of MaPP measures. - Examine complementary role of targeted capital controls when domestic credit pulls in cross-border funding and MaPP faces leakages. ### Data and sample - Panel: 62 economies (35 AEs plus 27 EMEs). - Sample period: 2000: Q1–2016: Q4. - Macroprudential policy database: IMF iMaPP database covering 17 instruments across 138 countries (1999–2016, monthly). - Winsorization: top and bottom 1 percent of each variable winsorized, except the dependent variable and the ordinal MaPP variable. ### Conceptual channels and mechanisms - Currency appreciation can raise collateral values and net worth, increasing borrowers’ capacity and easing lenders’ constraints. - Appreciation can lower perceived risk and boost borrowers’ sense of prosperity (e.g., cheaper imports), increasing demand and supply of credit. - Increased credit can affect local asset prices and be funded by cross-border borrowing, reinforcing effects of appreciation and contributing to systemic risk. ### Baseline empirical approach (overview) - Main dependent variable: credit-to-GDP gap (Y) — credit-to-GDP ratio relative to its long-run trend. - Main explanatory variable: ∆4RER — year-over-year log change of real exchange rate vis-à-vis the U.S. dollar (lagged). - Include MaPP and interaction terms MaPP_{i,t−1} × ∆4RER_{i,t−1} to evaluate mitigation effects. - Explore feedback from domestic credit to “other investment flows” (mainly cross-border loans and deposits received by financial institutions and the nonfinancial corporate sector). ### Baseline model specification - Baseline dynamic panel equation (notation preserved from source): - Y_{i,t} = ρ Y_{i,t−1} + β1 ∆4RER_{i,t−1} + β2 MaPP_{i,t−1} + β3 MaPP_{i,t−1} × ∆4RER_{i,t−1} + θ Z_{i,t−1} + μ_i + ν_{i,t} - E[μ_i] = E[ν_{i,t}] = E[μ_i ν_{i,t}] = 0 - Control vector Z_{i,t−1} = [MPS_{i,t−1}, ∆4F_RGDP_{i,t−1}]. - Variable conventions: - ∆4RER_{i,t−1}: year-over-year log change of real exchange rate, lagged one quarter. By convention, a negative change in ∆4RER represents a real exchange rate appreciation. - MaPP_{i,t−1}: lagged ordinal indicator of MaPP actions (tightening actions net of loosening actions); examples: +3, −3, 0. - MaPP subgroups: borrower-based tools (MaPP_Br) and financial institutions-based tools (MaPP_FI). - MPS: monetary policy stance (lagged central bank policy rates or shadow policy rates where applicable). - ∆4F_RGDP: forecasted year-over-year real GDP growth (from Consensus Forecast weighted average). ### Estimation strategy and econometric choices - Estimator: Arellano-Bond difference GMM (Arellano and Bond 1991) to address endogeneity and Nickell bias. - Instrumentation: - First-differenced lagged dependent variable instrumented with its 1–3 lags of its level. - All independent and control variables treated as predetermined. - Transformations and standard errors: - Forward orthogonal deviation transformation (Arellano and Bover 1995). - Two-step covariance estimates with Windmeijer (2005) correction for robust standard errors. - Lagging strategy: MaPP and controls lagged by one quarter. - Time-fixed effects: not included due to collinearity concerns and violation of AR(1) assumption when included. ### Endogeneity concerns and identification strategy - Primary concern: reverse causality where the credit gap drives MaPP (attenuation bias). - Four mitigation strategies: - Lag MaPP and controls by one quarter and include lagged dependent variable. - Use Arellano-Bond difference GMM for predetermined endogeneity. - Focus on interaction MaPP_{i,t−1} × ∆4RER_{i,t−1} argued to be less endogenous because exchange rate changes are unlikely considered when setting MaPP. - Construct MaPP shocks (actual MaPP minus expected value conditional on prior credit gap and exchange rate) to obtain shocks orthogonal to credit gap and ∆4RER (detailed in Section V). ### Main empirical findings (baseline) - Exchange rate effect on credit: - "An appreciation of the local exchange rate vis-à-vis the U.S. dollar is followed by an increase in the credit-to-GDP gap in the next quarter." - Quantified: A 10 percent real exchange rate appreciation is associated with a subsequent increase in the credit gap of 0.5 percentage points of GDP. - Effect size robust across specifications: between 0.5–0.6 percentage points; typically statistically significant at the 1 per cent level. - Median credit gap in sample: 0.33 percent of GDP. - Direct effect of macroprudential policy: - Net tightening of MaPP estimated to decrease the credit gap by 0.875 percentage points of GDP in the next quarter. - Tightening actions separately: 1.168 percentage points of GDP (strong and significant). - Loosening actions: estimated effect statistically insignificant when considered separately (in baseline aggregate). - Effect stronger for borrower-based tools than for financial institutions-based tools. - Interaction effect (MaPP mitigates exchange rate impact): - MaPP_{i,t−1} × ∆4RER_{i,t−1} coefficient strongly and statistically significant: MaPP weakens the impact of exchange rate movements on the credit gap. - Magnitudes: - One standard deviation increase in iMaPP reduces sensitivity of credit gap to RER movements by 0.82 percentage points of GDP. - Borrower-based tools: one standard deviation increase reduces sensitivity by 1.4 percentage points of GDP. - Financial institutions-based tools: one standard deviation increase reduces sensitivity by 0.9 percentage points of GDP. - Loosening and interaction asymmetry: - Prior relaxation (loosening) of MaPP can significantly increase the extent to which an exchange rate shock affects the credit gap, driven by loosening of financial institutions-based tools. - Control variables: - Monetary policy: one percentage point tightening reduces the credit gap by about 0.24–0.29 percentage points of GDP. - Forecasted real GDP growth: one percent improvement in annual forecasted real GDP growth increases the credit gap by about 0.45–0.5 percentage points of GDP. ### Dynamic properties, diagnostics, and bias considerations - Lagged dependent variable coefficients close to unity; Fisher-type panel unit root test confirms credit gap is stationary (in at least one panel). - AR(1) p-values about 2 percent; AR(2) p-values range from 23–37 percent supporting instrument lag choice. - Nickell bias: sample period T = 68; theoretical semi-asymptotic bias (assuming N → ∞) is −0.0391 in baseline for autoregressive coefficient 0.98; actual bias expected slightly larger for N = 62. ### Robustness and identification strategies (summary) - Distilling exchange rate shocks: - Two-step purification: regress ∆4RER on ∆4inflation, ∆4F_RGDP, ∆4CA_Deficit, and country fixed effects; save residuals as purged shocks (sample reduced from 62 to 60 when CA included). - First-stage R-square low (~0.04); GDP forecasts and inflation individually significant. - Using purged shocks yields very slight drop in exchange rate coefficient; core results hold. - Instrumental variable attempts suffered from weak instruments. - Conclusion: controlling for growth expectations (∆4F_RGDP) is a powerful approach; explicit exchange rate shocks useful but do not change main conclusions materially. - Using macroprudential policy shocks: - Three-step method: ordered probit of MaPP conditional on observables; compute expected MaPP; shocks = actual − expected. - Replacing MaPP with MaPP_Shock yields results roughly consistent with baseline; base effects of MaPP larger and more significant (reduced attenuation bias), especially for financial-institutions-based indicators. - First-stage: change in credit gap significant in overall and FI regressions; change in exchange rate not significant — consistent with MaPP not reacting to exchange rate movements. - Alternative credit measure: - Four-quarter change of credit-to-GDP produces results aligned with credit gap findings. - Interaction effects similar or larger in magnitude (example: MaPP_{i,t−1} × ∆4RER_{i,t−1} = 0.446 versus 0.256 when using credit gap in one comparison). - Direct MaPP effects somewhat weaker overall but remain highly significant for borrower-based tools. - Caveat: number of actions for specific tools is generally small—interpret with caution. ### Extension: feedback from credit to capital inflows and policy leakages - Question: Do strong domestic credit developments increase “gross other investment inflows” and how do policies affect this? - Estimation: regress gross other inflows on lagged credit gap, include quarter time-fixed effects μ_t, domestic pull factors (MPS, ∆4F_RGDP), policy levers (MaPP, MPS, FARI), and interactions; use country-fixed effects and time-fixed effects. - Main findings (Walls — long-standing FARI): - A 10 percentage points of GDP increase in the domestic credit gap is associated with an increase of cross-border capital inflows in the subsequent quarter ranging from 0.9 to 1.3 percentage points of GDP. - Monetary policy: a 1 percentage point increase in the policy rate is associated with an increase of about 1.7 percentage points in the ratio of other investment inflows to GDP. - Macroprudential tightening associated with increases in other investment flows at the margin (significant for borrower-based tools and when using MaPP shocks); an average tightening action of borrower-based tools associated with an increase in cross-border borrowing on the order of 1.3–1.5 percentage points of GDP. - Financial institution-based tools show insignificant effects on other investment flows. - Long-standing capital controls (FARI level) have a negative and significant direct impact and significantly mitigate the impact of a high credit gap on cross-border funding (significant at 10 and sometimes 5 per cent levels): act as “walls.” - Episodic controls (∆FARI — “gates”): - Direct and indirect coefficients have expected negative signs but are not always statistically significant. - Using shocks to capital controls improves statistical significance in some regressions. - Interaction effects between episodic controls and the credit gap remain insignificant; standardized coefficients much smaller than for long-standing controls. - Interpretation: incremental changes in capital controls have weaker effects relative to long-standing targeted controls. - Implication: MaPP and capital controls are complementary — long-standing targeted capital controls help when strong domestic credit pulls in capital and MaPP causes leakages. ### Policy implications and conclusions - Core empirical conclusions: - Exchange rate appreciation is associated with subsequent increases in the credit-to-GDP gap. - Tighter macroprudential policies can mitigate the effect of exchange rate appreciation on domestic credit cycles. - Strong domestic credit can feed back into increases in gross other investment inflows (cross-border borrowing). - Macroprudential tightening, especially borrower-based tools, can lead to cross-border leakages increasing other investment inflows. - Long-standing, targeted capital inflow controls can dampen feedback from credit to cross-border borrowing and are complementary to MaPP. - Policy recommendations: - Consider complementary use of macroprudential policies and capital inflow controls when a strong feedback effect from credit to direct cross-border borrowing is expected. - Deploy targeted, long-standing capital controls as a precautionary measure to prevent credit-driven surges in specific types of flows. *Source: wpiea2020187-print-pdf — 3.1 Baseline Setup and Methodology (excerpt).* ### 3.1 Baseline Setup and Methodology ......................................................... 10 ### 3.1 Baseline Setup and Methodology ### Research scope and objectives - Study link between exchange rate movements and domestic credit in a panel of 62 countries over the period 2000 to 2016. - Ask to what extent macroprudential policy can attenuate the effects of currency movements on domestic credit cycles. - Evaluate complementary role of targeted controls on inflows when strong credit developments lead to increases in cross-border borrowing. ### Key contributions - Use the credit-to-GDP gap as a continuous indicator of the build-up of systemic risk, contributing to literature on currency appreciation and domestic credit developments. - Examine interaction effect of macroprudential policy in mitigating the impact of the exchange rate on domestic credit, improving identification of causal effects of macroprudential measures. - Examine complementary role of targeted capital controls when domestic credit pulls in cross-border funding and macroprudential policy faces leakages. ### Data and sample - Panel: 62 advanced and emerging market economies (AEs and EMEs). - Sample period: 2000: Q1–2016: Q4. - Macroprudential policy database: iMaPP database compiled by IMF staff, covering 17 instruments for a total of 138 countries since 1999. ### Conceptual channels and mechanisms - Currency appreciation can raise collateral values and net worth, increasing borrowers’ capacity and easing lenders’ constraints (Krugman, 1999; Céspedes et al., 2004; Bruno and Shin, 2015b). - Appreciation can lower perceived risk and boost borrowers’ sense of prosperity (e.g., cheaper imports), increasing demand and supply of credit. - Increased credit can affect local asset prices and be funded by cross-border borrowing, reinforcing effects of appreciation and contributing to systemic risk (Gertler et al. 2007; Borio, 2014; Bruno and Shin 2015a,b; IMF 2017; Baskaya et al. 2017). ### Baseline empirical approach (overview) - Main dependent variable: credit-to-GDP gap as an indicator of systemic risk. - Main explanatory variable: changes in the real exchange rate vis-à-vis the U.S. dollar. - Include macroprudential policy stance (MaPP) and interaction terms between exchange rate changes and MaPP to evaluate mitigation effects. - Explore feedback from domestic credit to “other investment flows” (mainly cross-border loans and deposits received by financial institutions and the nonfinancial corporate sector). ### Main empirical findings (baseline) - Exchange rate movements are associated with subsequent changes in domestic credit relative to GDP: - An appreciation of the local exchange rate vis-à-vis the U.S. dollar is followed by an increase in the credit-to-GDP gap in the next quarter. - Macroprudential policy has a direct effect on domestic credit developments: - Where macroprudential policy is tightened, it leads to a reduction in the credit-to-GDP gap in the next quarter. - Interaction effect: - Macroprudential policy weakens the extent to which exchange rate movements affect credit developments. For a given appreciation of the real exchange rate, the subsequent increase in the credit-to-GDP gap is weaker where macroprudential policies had been tightened in the previous quarter. ### Robustness and identification strategies - Address simultaneity between fundamentals, exchange rates, and credit: - Two-step procedure that “purges” the impact of domestic fundamentals on the real exchange rate and distills more “exogenous” exchange rate shocks. - Findings continue to hold using purged exchange rate shocks. - Address endogeneity of macroprudential policy: - Construct macroprudential policy shocks as the difference between the actual reading on the macroprudential indicator variable and its expected value based on prior developments in the credit gap and the exchange rate. - Results persist when using these macroprudential policy shocks. - Alternative credit measures: - Results hold when using an alternative and simpler measure of domestic credit developments instead of the credit gap. ### Extension: feedback from credit to capital inflows and policy leakages - Controlling for global push factors and domestic pull factors, increases in domestic credit are associated with increases in “other investment flows.” - Evidence of leakages: - Where macroprudential policy is tightened this leads to further increases in cross-border flows, interpreted as corporate/financial sector substitution toward direct borrowing from abroad. - Effectiveness of targeted capital controls: - Targeted capital controls that aim to limit “other investment flows” appear to be effective. - Where these controls are in place, the effect of credit developments in stimulating other investment flows is reduced, thereby limiting further build-up of systemic risk from direct cross-border borrowing. ### Organization of the empirical analysis (paper structure) - Section II: Related literature. - Section III: Empirical framework and details of variables and data. - Section IV: Baseline empirical findings. - Section V: Further analyses and robustness checks. - 5.1 Using Actual Rather than Forecasted GDP Growth - 5.2 Distilling Exchange Rate Shocks - 5.3 Using Macroprudential Policy Shocks - 5.4 Effects of Macroprudential Policies by Country Characteristics - 5.5 Alternative Measure of Domestic Credit Developments - Section VI: Extension on feedback from credit to capital inflows. - Section VII: Conclusion and policy implications. *Source: wpiea2020187-print-pdf — 3.1 Baseline Setup and Methodology (excerpt).* ### 3.1 Baseline Setup and Methodology ### 3.1 Baseline Setup and Methodology ### Baseline model specification - Dependent variable Y: credit gap (credit-to-GDP ratio relative to its long-run trend). - Baseline dynamic panel equation (notation preserved from source): - Y_{i,t} = ρ Y_{i,t−1} + β1 ∆4RER_{i,t−1} + β2 MaPP_{i,t−1} + β3 MaPP_{i,t−1} × ∆4RER_{i,t−1} + θ Z_{i,t−1} + μ_i + ν_{i,t} - E[μ_i] = E[ν_{i,t}] = E[μ_i ν_{i,t}] = 0 - Control vector Z_{i,t−1} = [MPS_{i,t−1}, ∆4F_RGDP_{i,t−1}]. - Subscripts: i = country, t = time (quarter). μ_i = country fixed effect; ν_{i,t} = error term. - Variable definitions and conventions: - ∆4RER_{i,t−1}: year-over-year log change of real exchange rate, lagged one quarter. By convention, a negative change in ∆4RER represents a real exchange rate appreciation. - MaPP_{i,t−1}: lagged ordinal indicator of macroprudential policy actions (tightening actions net of loosening actions) during period t−1. Example: +3 (three tightenings), −3 (three loosenings), 0 (no action). - MaPP subgroups: borrower-based tools (MaPP_Br) and financial institutions-based tools (MaPP_FI). - MPS: monetary policy stance (lagged central bank policy rates or shadow policy rates where applicable). - ∆4F_RGDP: forecasted year-over-year real GDP growth (constructed from Consensus Forecast weighted average). ### Estimation strategy and econometric choices - Estimator: Arellano-Bond difference GMM (Arellano and Bond 1991) to address endogeneity and Nickell bias arising from the lagged dependent variable. - Instrumentation: - First-differenced lagged dependent variable instrumented with its 1–3 lags of its level. - All independent and control variables treated as predetermined rather than strictly exogenous. - Transformations and standard errors: - Forward orthogonal deviation transformation (Arellano and Bover 1995) used to mitigate data-gap issues in unbalanced panels. - Two-step covariance estimates and Windmeijer (2005) correction applied to obtain robust standard errors and correct downward bias. - Lagging strategy: - MaPP and controls lagged by one quarter; all independent variables lagged to mitigate endogeneity. - Justification against including time-fixed effects: - Quarterly time-fixed effect not included due to potential collinearity (making domestic driver coefficients statistically insignificant) and violation of first-order serial correlation assumption (AR(1) p-values range from 0.35 to 0.9 when included). ### Data, sample, and variable construction - Sample coverage: - 62 economies (35 AEs plus 27 EMEs). - Sample period: 2000: Q1 to 2016: Q4. - Winsorization: top and bottom 1 percent of each variable winsorized, except the dependent variable and the ordinal MaPP variable. - Credit gap construction: - Calculated using a one-sided HP filter with smoothing parameter λ = 400,000 (credit defined as total claims on private non-financial sector from banks and non-bank financial institutions; financial corporations’ domestic claims on private sector from IMF IFS used where available; otherwise depository corporations‘ domestic claims). - Robustness check: simpler 4-quarter change of credit-to-GDP ratio considered. - Forecasted YoY real GDP growth (∆4F_RGDP): - Constructed from Consensus Forecasts as a weighted average across quarters (weights described in source). - For five countries without consensus forecasts (Iceland, Lebanon, Luxembourg, Malta, Mongolia), realized forward growth rates used as a “perfect foresight” measure. - Real exchange rate (∆4RER): - Lagged year-over-year log change of weighted average of bilateral nominal exchange rate (denominated in national currency relative to USD) deflated by U.S. CPI relative to domestic CPI. - Weighted average calculation: X_WA = 0.4*X + 0.3*L1.X + 0.2*L2.X + 0.1*L4.X (alternative specifications tested with consistent results). - Macroprudential policy data: - Source: IMF iMaPP database (Alam et al., 2019), covering 17 instruments across 138 countries (1999–2016, monthly). - Aggregate MaPP index and two subgroups (MaPP_Br and MaPP_FI). Distinction between tightening (T_MaPP) and loosening (L_MaPP) actions possible. ### Endogeneity concerns and identification strategy - Primary endogeneity issue: potential reverse causality where credit gap drives macroprudential action (attenuation bias if high credit gaps induce tightening). - Four strategies to mitigate endogeneity: - Lag MaPP and control variables by one quarter and include lagged dependent variable in baseline. - Use Arellano-Bond difference GMM suitable for predetermined/relevant endogeneity structures. - Focus on the interaction MaPP_{i,t−1} × ∆4RER_{i,t−1}, which is argued to be less endogenous because changes in exchange rates are unlikely to be considered when setting MaPP; the exchange rate functions as an exogenous shifter of the effect of prior MaPP. - Construct MaPP shocks in further analysis (section V) by taking the difference between actual MaPP and its expectation conditional on developments in the credit gap and exchange rate, producing shocks orthogonal to both the credit gap and ∆4RER. ### Key baseline empirical findings (summarized) - Exchange rate effect on credit: - A 10 percent real exchange rate appreciation is associated with a subsequent increase in the credit gap of 0.5 percentage points of GDP. - Effect size robust across specifications: between 0.5–0.6 percentage points; typically statistically significant at the 1 per cent level. - Median credit gap in sample: 0.33 percent of GDP. - Direct effect of macroprudential policy: - Net tightening of MaPP estimated to decrease the credit gap by 0.875 percentage points of GDP in the next quarter. - This effect exceeds the median and is roughly 0.08 of the standard deviation of the credit gap. - Tightening actions separately: 1.168 percentage points of GDP (strong and significant). - Loosening actions: estimated effect statistically insignificant when considered separately (in baseline aggregate). - Effect stronger for borrower-based tools than for financial institutions-based tools (columns 5 and 9 in baseline regressions). - Interaction effect (MaPP mitigates exchange rate impact): - Interaction coefficient MaPP_{i,t−1} × ∆4RER_{i,t−1} strongly and statistically significantly indicates MaPP weakens the impact of exchange rate movements on the credit gap. - Magnitudes: - One standard deviation increase in iMaPP reduces the sensitivity of the credit gap to real exchange rate movements by 0.82 percentage points of GDP. - Borrower-based tools: one standard deviation increase reduces sensitivity by 1.4 percentage points of GDP. - Financial institutions-based tools: one standard deviation increase reduces sensitivity by 0.9 percentage points of GDP. - Loosening and interaction asymmetry: - Prior relaxation (loosening) of MaPP can significantly increase the extent to which an exchange rate shock affects the credit gap. - This effect appears driven by loosening of financial institutions-based tools rather than borrower-based ones (columns 8 and 12). - Control variables (signs and magnitudes): - Monetary policy: one percentage point tightening of monetary policy estimated to reduce the credit gap by about 0.24–0.29 percentage points of GDP. - Forecasted real GDP growth: one percent improvement in annual forecasted real GDP growth leads to an increase of the credit gap by about 0.45–0.5 percentage points of GDP. - Dynamic properties and diagnostics: - Lagged dependent variable coefficients close to unity; Fisher-type panel unit root test confirms credit gap is stationary (in at least one panel). - Large autoregressive coefficient partially attributed to forward orthogonal deviation transformation in GMM estimation. - Instrument validity and autocorrelation diagnostics: - AR(1) p-values about 2 percent (reject null of no first-order autocorrelation in first differences as expected). - AR(2) p-values range from 23–37 percent (do not reject null of no second-order autocorrelation in levels), supporting instrument lag choice. - Nickell bias consideration: - Sample period T = 68; theoretical semi-asymptotic bias (assuming N → ∞) is −0.0391 in the baseline for an autoregressive coefficient of 0.98; actual bias expected slightly larger for N = 62. *Source: 3.1 Baseline Setup and Methodology (wpiea2020187-print-pdf).* ### 5.1 Using Actual Rather than Forecasted GDP Growth ### 5.1 Using Actual Rather than Forecasted GDP Growth ### Comparison of actual (∆4RGDP) versus forecasted GDP growth - Baseline replaces forecasted real GDP growth with realized growth: the (lagged) year-over-year log change of quarterly real GDP from the IMF WEO database (∆4RGDP). - Literature note: actual GDP growth rates are more commonly used (Fendoğlu 2017; Cerutti, Claessens, and Laeven 2017) but may be inferior here as a control for simultaneity when credit and exchange rate changes are jointly driven by expectations. ### Main findings - Results in Table 5 are consistent with Table 4 but indicate GDP forecasts are a much stronger control than past realizations. - Coefficients: - A one percent increase in actual real GDP growth (∆4RGDP) leads to an increase of the credit gap by about 0.24-0.29 percentage points of GDP. - A one percent increase in forecasted real GDP growth leads to an increase of the credit gap by about 0.45-0.5 percentage points of GDP. - Interpretation: evolution of credit gaps is more closely related to expectation of future growth than to past growth. - Real exchange rate movement coefficient is somewhat larger when using actual GDP than when using forecast GDP, consistent with capture of “good” news effects by the larger coefficient when actual GDP is used. - Interaction term MAPP_{i,t-1} × ∆4RER_{i,t-1} remains roughly the same in magnitude and significance. --- ### 5.2 Distilling Exchange Rate Shocks ### Rationale - Concern: inflation, current account developments, and growth expectations may simultaneously drive real exchange rate and credit developments, introducing simultaneity bias that overstates the causal effect of exchange rate changes. - Goal: distill more “exogenous” exchange rate shocks for baseline regressions. ### Two-stage purification procedure - First stage: fixed-effect regression of quarterly real exchange rate change (∆4RER_{i,t}) on domestic fundamentals: - ∆4inflation (year-over-year change of CPI) - ∆4F_RGDP (forecasted real GDP growth) - ∆4CA_Deficit (year-over-year change in the current account deficit; positive values entail a greater deficit) - Country fixed effects (η_i) - Residuals e_{i,t} saved as purged shocks - Data note: sample reduced from 62 to 60 economies when current account included. - Second stage: use lagged residuals from first stage to replace ∆4RER_{i,t-1} in baseline dynamic panel regression. ### First-stage diagnostics and findings (Table 6) - Overall explanatory power low: overall R-square about 0.04 (first-stage regressions have very low explanatory power, overall R-square less than 0.05 noted). - Individual fundamentals: - Growth forecasts and inflation developments individually significant in explaining exchange rate movements. - Effect of the GDP forecast is overall the strongest. ### Second-stage results (Table 7) and robustness - Using purged exchange rate shocks: - Very slight drop in the size of the exchange rate coefficient relative to baseline, but core results continue to hold strongly. - Macroprudential policy continues to affect the credit gap directly and indirectly by reducing the impact of exchange rate shocks on credit. - Including inflation and changes in the current account in main regressions: these variables are not significant in affecting the credit gap; other results mainly unchanged (results not shown). - Instrumental variable attempts: - Instruments tried: time fixed effects, average changes in real exchange rates against the dollar in countries within the same region, changes in commodity prices from Gruss and Kebhaj (2019). - Problem encountered: “weak instruments” — low correlation of predicted values with actual change in real exchange rates reduces correlation with credit developments in second stage. - Conclusion: - Important to control for potential simultaneity; including GDP forecasts is already a powerful way. - Constructing explicit exchange rate shocks is useful but does not lead to major changes compared to a baseline that controls for growth expectations. - Idiosyncratic variation in exchange rate movements accounts for an important part of the co-movement between exchange rates and the credit gap. --- ### 5.3 Using Macroprudential Policy Shocks ### Motivation - Concern: reverse causality where macroprudential policy reacts to credit developments creates attenuation bias (bias toward zero) when estimating policy effects. - Baseline uses lag of macroprudential indicator; here identification strengthens by constructing macroprudential policy “shocks.” ### Three-step identification method (following Brandao-Marques and others (2020)) - Step 1: Estimate an ordered probit model of the macroprudential policy indicator conditional on observables. Independent variables: - Year-over-year change of the quarterly credit-to-GDP gap - Change of the real exchange rate (∆4RER as in baseline) - Quarterly change of net capital inflows (percent of GDP) - Indicator of lagged policy actions (sum of lags one to four of quarterly macroprudential policy indicator) - Country indicator (cross-sectional heterogeneity) - Ordered-probit results shown in Table 8. - Step 2: Compute the “expected” macroprudential policy stance using probabilities from the ordered probit conditional on independent variables. - Step 3: Compute macroprudential policy shocks as actual macroprudential indicators minus expected values; positive = tightening shocks, negative = loosening shocks. - Constructional property: shocks are orthogonal to past credit developments and to exchange rate changes, thereby reducing endogeneity and mitigating concerns of policy responding to exchange rate movements. ### Implementation details - Macroprudential indicator recategorized into five outcomes: +2 (tightening by two or more actions), +1 (tightening by one action), 0 (no change), -1 (loosening by one action), -2 (loosening by two or more actions). Ordered probit used to model these five buckets. ### Results (Table 9 and Table 8 first-step diagnostics) - Replacing MAPP_{i,t-1} with MAPP_Shock_{i,t-1} yields results roughly consistent with baseline; interaction terms remain essentially the same. - Marginal differences: - Base effects of macroprudential action measured larger and more significant compared to baseline, notably for aggregate indicator and financial institutions-based indicator — suggesting reduced attenuation bias when using shocks. - First-step (ordered probit) findings: - Change in the credit gap is sizable and statistically significant at the five and one per cent levels in regressions explaining overall and financial-institutions-based indicators, respectively. - Credit gap not significant in regressions explaining borrower-based tools — implying borrower-based tools respond less to credit developments. - Change in the exchange rate not statistically significant in any first-stage regressions — consistent with prior that macroprudential policy does not tend to react to exchange rate movements. - Overall conclusion: using macroprudential policy shocks reduces attenuation bias from reverse causality, mainly affecting base effects (especially financial-institutions-based tools), while interactions of all types of tools with exchange rate changes are not substantially affected. --- ### 5.4 Effects of Macroprudential Policies by Country Characteristics ### Sample splits and specification - Initial split: G7 countries (source countries of global capital flows) versus non-G7 group (tend to receive flows). - Specifications use both macroprudential policy shocks and exchange rate shocks (as introduced above) for the full sample. ### Key findings (Table 10 and referenced columns) - Full-sample columns 1–4: - Macroprudential policy shocks significantly reduce credit. - Interaction: a macroprudential tightening reduces the expansionary effect of an appreciation on the credit gap. - G7 versus non-G7 (columns 5–12): - Effects strong for non-G7 group, not for G7. - Base effect of exchange rate on credit is significant and strong for non-G7; insignificant for G7 (as indicated by ∆4RER_{i,t-1} coefficients in columns 9–12). - Tightening macroprudential policy weakens the currency–credit interaction strongly and significantly for non-G7 but is insignificant for G7 countries. - Further split of non-G7 by financial openness (columns 13–21): - Interaction effects between macroprudential policy and exchange rate shocks remain strong and significant in the more financially open group (sample size reduced to 27 countries). - Both base and interaction effects are statistically insignificant for relatively more financially closed economies. - Overall interpretation: effects are strongest for small open economies, and less relevant statistically for advanced economies (G7) or relatively closed EMDEs. *Source: IMF working paper chapter 5 content provided.* ### 5.5 Alternative Measure of Domestic Credit Developments ### 5.5 Alternative Measure of Domestic Credit Developments ### Alternative outcome and specification - Alternative measure: four-quarter change of the credit-to-GDP ratio (avoids filtering methods used for the credit gap). - Specification retains: exchange rate shocks, macroprudential shocks, monetary policy, and GDP forecasts as controls. - Key comparison: results using the four-quarter change of credit-to-GDP align with those using the credit gap, without relying on statistical filtering. ### Main empirical findings (alternative measure) - Currency appreciations are associated with increases in credit using the alternative measure; statistical significance of increases in the ratio of credit to GDP is comparable to results for the credit gap. - Direct effects of macroprudential policy on the alternative measure: - Somewhat weaker overall, but effects remain statistically highly significant for borrower-based tools. - Interaction effects (macroprudential policy × exchange rate) are similar or larger in economic magnitude versus the credit gap specification: - Example: coefficient of MaPP_{i,t−1} × ∆4 RER_{i,t−1} = 0.446 (bottom panel, column 2) versus 0.256 when using the credit gap (upper panel). - Relaxations of macroprudential policy strengthen the effect of appreciation on credit for both borrower-based and financial institution-based tools. - Headline conclusion: the link between exchange rates and credit and the mitigating interaction effects of macroprudential policy carry over to the alternative measure. ### Caveats - Results for individual macroprudential tools should be interpreted with caution because the number of actions recorded for specific tools is in general quite small. --- ### VI. Extension: Feedback Effect from Credit to Capital Inflows ### Research question and approach - Question: Does strong domestic credit (measured by lagged credit-to-GDP gap) lead to increases in “gross other investment inflows” (cross-border borrowing by domestic financial institutions and non-financial corporates), and how do policies affect this relationship? - Identification strategy: - Regress gross other inflows on lagged credit gap, include quarter time-fixed effects (μ_t) to account for global push factors. - Include domestic pull factors: monetary policy and forecasted real GDP growth. - Include policy levers: macroprudential policy (MaPP), monetary policy stance (MPS), and capital controls (FARI). - Analyze two forms of capital controls: - “Walls”: level of Financial Accounts Restrictiveness Index (FARI) on other investment inflows. - “Gates”: 1-quarter changes (∆FARI), coded as +1 for net tightening and -1 for net loosening. - Interaction terms included to identify how policies alter the sensitivity of other inflows to the credit gap. - Estimation uses country-fixed effects and time-fixed effects; dynamic panel (GMM) not used here because AR(1) tests range from 0.62 to 0.8 (evidence against first order serial correlation in residuals in differences). ### Main empirical findings (Walls, Tables 13) - Feedback effect from credit to other investment inflows: - A 10 percentage points of GDP increase in the domestic credit gap is associated with an increase of cross-border capital inflows in the subsequent quarter, ranging from 0.9 to 1.3 percentage points of GDP. - Monetary policy: - A 1 percentage point increase in the policy rate is associated with an increase of about 1.7 percentage points in the ratio of other investment inflows to GDP. - Interaction terms between monetary policy and the credit gap are not significantly different from zero. - Macroprudential policy: - Macroprudential tightening is associated with increases in other investment flows at the margin, significant for borrower-based tools and particularly when using macroprudential policy shocks. - An average tightening action of borrower-based tools is associated with an increase in cross-border borrowing on the order of 1.3–1.5 percentage points of GDP. - Financial institution-based tools show insignificant effects on other investment flows. - Capital controls (long-standing “walls”): - Direct impact of long-standing controls is negative and significant (at the 10 percent significance level). - Long-standing controls significantly mitigate the impact of a high credit gap in increasing cross-border funding (significant at the 10 and sometimes 5 per cent level). - Interpretation: targeted long-standing controls act as a “wall” when the credit tide is up and pulls in additional capital from abroad. ### Episodic controls (“gates”, Table 14) - Direct and indirect coefficients have expected negative signs but are not always statistically significant. - Using shocks to capital controls improves statistical significance in some regressions, indicating potential reverse causality bias if unaddressed. - Interaction effects between episodic controls and the credit gap remain insignificant throughout. - Standardized coefficients for episodic controls are much smaller than for long-standing controls. - Interpretation: incremental changes in capital controls have weaker effects versus long-standing targeted controls. ### Implications from extension - Macroprudential policy and capital controls are complementary: - Long-standing, targeted capital controls are useful when strong domestic credit pulls in additional capital inflows and macroprudential measures are insufficient or cause cross-border leakages. - Targeted controls may be most effective when deployed precautionarily, ahead of a surge in specific flows. --- ### VII. Conclusion & Policy Implications - Core findings: - Exchange rate appreciation is associated with subsequent increases in the credit-to-GDP gap. - Tighter macroprudential policies can mitigate the effect of exchange rate appreciation on domestic credit cycles. - Strong domestic credit can feed back into increases in gross other investment inflows (cross-border borrowing). - Macroprudential tightening, especially of borrower-based tools, can lead to cross-border leakages, increasing other investment inflows. - Long-standing, targeted capital inflow controls can dampen the feedback from credit to cross-border borrowing and are complementary to macroprudential measures. - Policy recommendations: - Consider complementary use of macroprudential policies and capital inflow controls when a strong feedback effect from credit to direct cross-border borrowing is expected. - Deploy targeted, long-standing capital controls as a precautionary measure to prevent credit-driven surges in specific types of flows, taking into account alternatives and tradeoffs. *Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020187-print-pdf.pdf* ### REFERENCES ### REFERENCES ### Macroprudential policy, capital flows, and financial cycles - Ahnert, T., K. Forbes, C. Friedrich, and D. Reinhardt. 2018. “Macroprudential FX Regulations: Shifting the Snowbanks of FX Vulnerability?” Bank of England Working Paper No. 758. - Akinci, O., and J. Omstead-Rumsey. 2018. “How Effective Are Macroprudential Policies? An Empirical Investigation.” Journal of Financial Intermediation 33: 33–57. https://doi.org/10.1016/j.jfi.2017.04.001. - Alam, Z., A. Alter, J. Eiseman, G. Gelos, H. Kang, M. Narita, E. Nier, and N. Wang. 2019. “Digging Deeper—Evidence on the Effects of Macroprudential Policies from a New Database.” IMF Working Paper WP/19/66. - Brandao-Marques, L., R. G. Gelos, M. Narita, and E. 2020. “Leaning Against the Wind: A Cost-Benefit Analysis for an Integrated Policy Framework.” IMF Working Papers 20/123, International Monetary Fund. - Bruno, V., I. Shim, and H. S. Shin. 2017. “Comparative Assessment of Macroprudential Policies.” Journal of Financial Stability 28: 183–202. https://doi.org/10.1016/j.jfs.2016.04.001. - Cerutti, E., S. Claessens, and L. Laeven. 2017. “The Use and Effectiveness of Macroprudential Policies: New Evidence.” Journal of Financial Stability 28: 203–24. https://doi.org/10.1016/j.jfs.2015.10.004. - IMF. 2017. “Increasing Resilience to Large and Volatile Capital Flows : The Role of Macroprudential.” IMF Policy Paper 17/260. Washington, D.C. - IMF-FSB-BIS. 2016. “Elements of Effective Macroprudential Policies,” no. August: 1–22. https://doi.org/10.1177/0018726713492349. - Lim, C. H., F. Columba, A. Costa, P. Kongsamut, A. Otani, M. Saiyid, T. Wezel, and X. Wu. 2011. “Macroprudential Policy : What Instruments and How to Use Them ? Lessons from Country Experiences.” IMF Working Paper Series WP/11/238: 1–85. https://doi.org/10.1080/1350293X.2011.623511. - Cerutti, E. and H. Zhou, 2018. “Cross-border Banking and the Circumvention of Macroprudential and Capital Control Measures.” IMF Working Papers 18/217, International Monetary Fund. - Claessens, S., S. R. Ghosh, and R. Mihet. 2013. “Macro-Prudential Policies to Mitigate Financial System Vulnerabilities.” Journal of International Money and Finance 39: 153–85. https://doi.org/10.1016/j.jimonfin.2013.06.023. ### Capital flows, exchange rates, and external financial linkages - Avdjiev, S, R McCauley, and P McGuire. 2012. “Rapid Credit Growth and International Credit: Challenges for Asia.” BIS Working Papers, No 377, April. - Baskaya, Y.S., J. Di Giovanni, S. Kalemli-Özcan, J. L. Peydró, and M.F. Ulu. 2017. “Capital Flows and the International Credit Channel.” Journal of International Economics, 108, pp. S15–S22. - Blanchard, O., J. D. Ostry, A. R. Gosh, and M. Chamon. 2015. “Are Capital Inflows Expansionary or Contractionary? Theory, Policy Implications, and Some Evidence.” IMF Working Paper WP/15/226. - Caballero, J. A. 2016. “Do Surges in International Capital Inflows Influence the Likelihood of Banking Crises?” The Economic Journal, Volume 126, Issue 591, March. pp. 281–316. - Gourinchas, P. O., and M. Obstfeld. 2012. “Stories of the Twentieth Century for the Twenty-First.” American Economic Journal: Macroeconomics 4 (1): 226–65. https://doi.org/10.1257/mac.4.1.226. - Lane, P. and P. McQuade. 2014. “Domestic Credit Growth and International Capital Flows.” Scandinavian Journal of Economics, vol 116, issue 1, January, pp 218–52. - Lane, P. and G. M. Milesi-Ferretti. 2007. “The External Wealth of Nations Mark II: Revised and Extended Estimates of Foreign Assets and Liabilities,” Journal of International Economics, Vol. 73, No. 2, pp. 223–50. - Hofmann, B., I. Shim, and H. S. Shin. 2019. “Bond Risk Premia and the Exchange Rate.” BIS Working Paper No. 775 (775). - Cespedes, L. F., R. Chang, and A. Velasco. 2004. “Balance Sheets and Exchange Rate Policy.” American Economic Review 94 (4), pp. 1183–1193. ### Financial stability, bank behavior, and risk-taking - Borio, C. 2014. “The Financial Cycle and Macroeconomics: What have we Learnt?” Journal of Banking & Finance, Elsevier, Vol. 45(C), pp. 182–198. - Borio, C., and H. Zhu. 2012. “Capital Regulation, Risk-Taking and Monetary Policy: A Missing Link in the Transmission Mechanism?” Journal of Financial Stability 8 (4): 236–51. https://doi.org/10.1016/j.jfs.2011.12.003. - Hahm, J., H. S. Shin, and K. Shin. 2013. “Non-Core Bank Liabilities and Financial Vulnerability.” Journal Of Money, Credit and Banking 45 (1): 1–40. - Shin, H. S. 2018. “Bank Capital and Monetary Policy Transmission.” in Hartmann, P., Huang, H., and Schoenmaker, D. (eds) The Changing Fortunes of Central Banking. Cambridge: Cambridge University Press, pp. 80–100. - Gertler, M., S. Gilchrist and F. M. Natalucci. 2007. “External Constraints on Monetary Policy and the Financial Accelerator.” Journal of Money, Credit and Banking, Blackwell Publishing, Vol. 39(2-3), pp. 295–330, March. - Caballero, R., and A. Krisnamurthy. 2004. "Exchange Rate Volatility and the Credit Channel in Emerging Markets: A Vertical Perpective." NBER Working Paper No. 10517. ### Empirical methods, panel econometrics, and inference - Arellano, M. and O. Bover. 1995. “Another Look at the Instrumental Variable Estimation of Error-Components Models.” Journal of Econometrics Vol. 68 (1): pp. 29–51. - Arellano, M., and S. Bond. 1991. “Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations.” Review of Economic Studies 58: pp. 277–97. - Kiviet, J.F., 1995. “On Bias, Inconsistency, and Efficiency of Various Estimators in Dynamic Panel Data Models.” Journal of Econometrics, 68(1): 53-78. - Windmeijer, F. 2005. "A Finite Sample Correction for the Variance of Linear Efficient Two-step GMM Estimators." Journal of Econometrics 126(1): pp. 25-51. - Roodman, D. 2009. “How to do xtabond2: An introduction to difference and system GMM in Stata.” The Stata Journal 9(1): 86-136. - Bun, M. J.G. and T. D. Harrison. 2019. “OLS and IV Estimation of Regression Models Including Endogenous Interaction Terms." Econometric Reviews 38:7, 814-827. - Davidson, R. and J. G. MacKinnon. 1993. “Estimation and Inference in Econometrics.” OUP Catalogue, Oxford University Press. - Phillips, P. C. B., and H. R. Moon. 2000. “Nonstationary Panel Data Analysis : An Overview of Some Recent Developments” Econometric Review 19 (3): 263–286. ### Databases, indices, and data sources used or cited - IMF IFS; BIS; central banks (credit and related series). - IMF iMaPP Database (aggregated macroprudential policy measures). - IMF WEO; Consensus Forecast (forecasted real GDP growth). - IMF FFA (capital inflows). - IMF AREAER (financial account restriction index). - Basel Committee on Banking Supervision (BCBS). 2010. Guidance for National Authorities Operating the Countercyclical Capital Buffer. - Bank for International Settlements. 2019. BIS Annual Economic Report 2019. June 2019. ### Tables, figures, and empirical components referenced in this section - Table 1. Description of Variables - Dependent variable: Credit-to-GDP gap Y — Calculated by one-sided HP filter using smoothing parameter=400,000. - Independent variables: Macroprudential policy stance MaPP, TM aPP, L_MaPP; +1 tightening, -1 loosening, 0 no action; IMF iMaPP Database. - YoY change of real exchange rate ∆4 RER: Year-over-year log change of weighted average of quarterly bilateral nominal exchange rate in NC per USD, deflated by U.S. CPI against domestic CPI. Positive value = depreciation against USD. IMF IFS. - Control variables: Monetary policy stance MPS (central bank policy rates; use shadow policy rates for Euro Area, U.S., U.K., and Japan); Forecasted YoY real GDP growth ∆4 F_RGDP (Consensus Forecast, IMF WEO). - Other variables: YoY real GDP growth ∆4 RGDP (IMF WEO); Capital inflows CFLOW (Gross other investment inflows within quarter t as percent of GDP in t-1) IMF FFA; Capital control FARI (Financial account restriction index on gross other investment inflows, range 0-1) IMF AREAER. - Note: (a) Initially take the first 24 quarters, then compute the trend and cyclical components recursively (add one quarter at a time). (b) We use the financial corporations’ domestic claims on private sector from IMF IFS where available, otherwise, we use depository corporations (or monetary) domestic claims on private sector from IMF IFS. The credit data of Iceland and Taiwan POC are from the central banks. (c) The weighted average of variable X is calculated as X_WA= 0.4*X + 0.3*L1.X + 0.2*L2.X + 0.1*L4.X. - Table 2. Summary of Statistics — Source: Authors’ calculations. Note: We winsorize the top and bottom 1 percent observations of each variable except the dependent variable Y and the ordinal variable MaPP. - Table 3. Definition of Macroprudential Policy Instruments — Source: Alam et al. (2019). (List of instruments 1–17 with definitions, including 1 CCBA, 2 LVRA, 3 LLP, 4 Capital, 5 LTV, 6 DSTI, 7 LFC, 8 RR, 9 Tax, 10 SIFI, 11 Liquidity, 12 LTD, 13 LFX, 14 LCG, 15 Conservation, 16 LoanR, 17 Other.) - Tables 4–14 and Figures 2–3 — Multiple regression results, robustness checks, ordered probit regression, leakages and gates/walls effects of capital controls, country-characteristic heterogeneity, and visualizations of Credit-to-GDP Gap and usage of iMaPP actions. Key recurring empirical sample values: Observations = 3,842 (Tables 4–6); # of Economies = 62; Observations = 3,699 (Table 8); Observations = 3,505 and 3,842 appear across robustness tables; sample period references include 2000: Q1–2016: Q4 and 2000: Q1–2016: Q5 in figures. - Notes across tables: Standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. AB AR(1) test and AR(2) test p values, Hansen test - p value often = 1 in reported tables. *Source: wpiea2020187-print-pdf - REFERENCES* --- _Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020187-print-pdf.pdf_