## wpiea2023171-print-pdf — References

## Source details

**Canonical URL:** [wpiea2023171-print-pdf — References](https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023171-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2023/english/wpiea2023171-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2023/english/wpiea2023171-print-pdf.pdf.json)

---

### Main findings and conclusions
- Macroprudential policy is effective in helping countries experiencing credit booms avoid ending in a credit bust.
- Macroprudential tools with significant effects include:
  - capital-based tools;
  - borrower-based tools such as loan-to-value (LTV) and debt-service-to-income (DSTI) caps;
  - reserve requirements.
- When credit booms are accompanied by capital flow surges, in addition to capital-based macroprudential tools, capital controls on money market instruments, including short-term cross-border interbank lending, tend to reduce the likelihood of a bust.
- The paper focuses on whether MMs and CCs decrease the probability that a credit boom ends in a credit bust, i.e., whether these policies make the financial system more resilient by creating buffers at the level of lenders and borrowers.

### Data, sample, and empirical approach
- Dataset coverage:
  - Annual frequency from 1996-2016.
  - 53 countries: 27 emerging markets and 26 advanced economies.
- Prudential policy coding:
  - Policy loosening coded as -1;
  - Policy tightening coded as 1;
  - Unchanged policy coded as 0.
- Identification strategy requires a forward-looking window of at least three years to determine whether a credit boom ended in a bust; policy interventions analyzed only through 2016 (credit data for booms/busts extends to 2019).
- Five source databases used for prudential measures with Alam et al. (2019) as primary source, cross-checked against others.

### Construction of macroprudential measures (MPMs) and capital controls (CCs)
- MPMs:
  - Convert quarterly and monthly datasets to annual frequency; first-difference Cerutti et al. (2017a) variables to capture changes.
  - Group measures into 8 categories: RR; CG; LTV; DTI; CAP; PR; EXP; OTH.
  - Indicators take values in {-1,0,1}. Overall MPM index = sum over eight categories; index ranges from -8 to +8.
  - Limitations: intensity of instrument use not fully captured; summation across categories does not account for differing economic impacts.
- CCs (inflows):
  - Based on Fernández et al. (2015) using AREAER de jure information, annual frequency 1996–2016.
  - First-difference annual stock series to obtain flow-equivalents; recode differences into {-1,0,1}.
  - Original 10 asset categories merged into 8 instrument types; overall CC (inflows) index = sum of individual categories; ranges from -8 to +8.

### Mapping, aggregation, and sample coverage
- Merged related sub-instruments by summing values where at least one variable available; sign indicates tightening (1), loosening (-1) or no change (0).
- Final geographical coverage after harmonization:
  - Asia-Pacific: 13 countries
  - Europe: 27
  - North & Latin America: 9
  - Middle East and Africa: 4

### Identification of credit booms, busts, and capital inflow surges
- Credit boom definition (real-time):
  - Must last at least 2 years.
  - At least one year with change > 3 ppts.
  - All other years of the boom: change > 2 ppts.
  - Exclude years where increase > 2 ppts driven by GDP contraction (real GDP growth decline ≥ 1%).
  - Combine adjacent booms if gap ≤ 1 year and nominal credit growth in gap year is positive.
- Credit bust definition:
  - Must last at least 2 years.
  - At least one year with change < -3 ppts.
  - All other years of the bust: change < -1 ppts.
  - Merge two busts with gap ≤ 1 year if nominal credit growth in gap year is negative.
- Capital inflow surge definition:
  - A surge occurs when gross private capital inflow-to-GDP ratio exceeds its 5-year rolling mean by one standard deviation (Forbes and Warnock (2012) method).
  - No minimum duration restriction for surges.

### Cycle sample statistics and descriptive measures
- 72 credit boom episodes identified over 1996-2016 for 53 countries.
- 26 of the 72 booms ended in a bust within 3 years (about 36%).
- Average duration of a credit boom: 7 years (min 2 years, max 14 years).
- Average duration of a credit bust: 4 years (min 2 years, max 10 years).
- About three quarters of booms last 2–11 years; three quarters of busts last 2–5 years.
- Average absolute annual change in credit-to-GDP in booms and busts ≈ 7 percentage points; normal times annual change ≈ 1 percentage point.
- Over 170 surge episodes identified; average surge duration slightly below 2 years.
- Of 72 credit booms, 51 coincide with a surge in private capital inflows within the first three years of the boom.

### Evolution of policies over time (descriptive)
- Late 1990s: net loosening of both macroprudential and capital account restrictions.
- 2000s: capital inflow restrictions continued to be removed up to the global financial crisis; macroprudential policies tightened between 2002 and 2007.
- Global financial crisis 2008/2009: macroprudential policies loosened across sample; capital controls largely unchanged.
- Post-crisis: both policy types tightened in net terms; macroprudential steep increase driven by new frameworks and Basel III implementation; renewed use of capital inflow controls in some countries.
- Instrument use frequency (1996-2016):
  - Macroprudential: reserve requirements (RR) most actively used, followed by capital-based tools (CAP) and LTV caps; limits on credit growth (CG) least used.
  - Capital controls: overall fewer policy changes (~half the number of macroprudential policy changes). Controls on derivatives most commonly used, followed by short-term money market (mm) and equity flows.

### Policy use over cycles (descriptive findings)
- Macroprudential policy during booms:
  - Countercyclical on average: tightened during booms more often than loosened.
  - Of 72 booms: 35 had net MPM tightening, 24 had no action, 13 had net loosening.
  - Tighter macroprudential policies on average when a credit boom coincides with a capital surge.
- Capital controls during booms:
  - More gross tightening (and less gross loosening) of capital controls during booms accompanied by capital surges vs booms without surges.
  - Net effect: CCs are mostly acyclical on average; zero net tightening during booms with surges, some net loosening during booms without surges.
- Advanced vs emerging markets:
  - Much macroprudential tightening in advanced economies occurred during normal times (driven by Basel II/III and post-crisis regulation).
  - Emerging markets are on average more active in using macroprudential policies during credit booms than advanced economies.

### Econometric assessment of policy effectiveness during booms
- Approach:
  - Event-study/probit model using 72 cross-sectional boom episodes.
  - Dependent variable: dummy = 1 if boom followed by bust within 3 years, 0 otherwise.
  - Controls include average credit-to-GDP change during boom and boom duration (years).
  - Policy variables: cumulative (sum over boom) net tightening of MPMs and CCs (inflows); alternatives use mean per year and gross tightening.
- Key empirical findings:
  - Larger and longer credit booms are more likely to end in a bust.
  - MPMs reduce the probability of a bad boom; CCs not significant in baseline specifications.
  - Evaluated at sample means (average probability of a bad boom = 33%):
    - Net tightening of MPMs by 1 reduces the probability of a bad boom by about 7% on average.
- Disaggregated instrument effects (marginal effects reported):
  - Capital instruments (CAP) and DTI caps: tightening by one decreases probability of a bad boom by 22% (each).
  - LTV caps: tightening by one decreases probability by 18%.
  - Reserve requirements (RR): tightening by one decreases probability by 11%.
- Regression coefficient highlights (Table 3 baseline, cumulative measures):
  - Credit-to-GDP change (avg): coefficient ≈ 0.113*** (standard error 0.0430).
  - Boom duration: coefficient ≈ 0.0944* (standard error 0.0525).
  - MPMs (cum.): -0.184** (standard error 0.0776).
  - CCs on inflows (cum.): not statistically significant in main specifications (e.g., -0.0557 with standard error 0.0996).
  - Observations: 72 in many specifications; R2-pseudo example: 0.195.

### Robustness checks and appendix findings
- Stricter definition of booms/busts reduces identified booms to 53, of which 17 ended in a bust.
- Focused subset: 51 booms accompanied by private capital inflow surges; 18 ended in a bust.
- Main robustness results:
  - Capital-based macroprudential instruments show a consistent role in decreasing the probability of credit booms ending in busts across alternative specifications.
  - Borrower-based tools and reserve requirements also appear effective in many specifications, but significance of individual borrower-based tools (LTV, DTI) varies across specifications.
- Controls and additional checks:
  - Changes in policy rates during booms: no significant impact.
  - Country development dummy: significant and positive—booms in emerging market economies more likely to end in a bust.
  - Real GDP growth dynamics and current account balances: no significant impact.
  - Global volatility (VIX): can increase probability depending on specification, but does not qualitatively change main conclusions.
- Results for booms with capital surges:
  - Capital instruments, caps on DTI, and reserve requirements continue to decrease bust probability; exposure limits are also significant in this subset.
  - Total index of capital controls not significant, but tightening of controls on inflows via money market instruments can be statistically significant in decreasing the probability of a bad boom.
  - When focusing on debt inflow surges, results are qualitatively similar.
- Quantitative estimates for booms with surges (evaluated at means):
  - Net tightening of a single macroprudential instrument during a credit boom accompanied by a capital flow surge reduces the probability of a boom ending in a bust by around 11% from the initial 29% probability.
  - Effects of individual instruments (subset of booms with surges):
    - Capital instruments: 31%
    - Caps on DTI: 29%
    - Reserve requirements: 15%
    - Exposure limits: 35%
- Specification checks (means vs sums, net vs gross tightening) broadly confirm main roles for capital instruments and some borrower-based tools; capital controls' role is more specific to money-market instruments in surge episodes.

### Policy implications and recommendations
- Prefer domestic prudential measures (e.g., LTV, credit growth limits, reserve requirements) to restrain aggregate credit intensity; capital controls can complement by reducing other vulnerabilities (e.g., share of FX loans, reliance on external debt).
- In episodes with capital flow surges accompanying credit booms, combine capital-based macroprudential tools with capital controls on money market instruments (including short-term cross-border interbank lending) to reduce bust probability.
- Building buffers during booms—through capital-based tools, borrower-based limits (LTV, DSTI), and reserve requirements—can increase resilience even if moderation of credit growth is not fully achieved.
- Open questions for follow-up research highlighted:
  - Additional country characteristics or boom features that condition effectiveness (institutional, regulatory, governance differences; type of credit driving the boom; initial financial integration; degree of initial financial development).
  - Optimal sequencing and combinations of macroprudential instruments and capital controls over credit booms.
  - Which policies reduce the probability of particularly harmful busts (e.g., those accompanied by banking crises).
  - Effect of additional policies (fiscal or exchange rate) on reducing the probability of a credit boom ending in a credit bust.

*Source: IMF Working Paper — Macroprudential Policies and Capital Controls Over Financial Cycles (wpiea2023171-print-pdf).*

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

### wpiea2023171-print-pdf - References

### Main findings and conclusions
- Macroprudential policy is effective in helping countries experiencing credit booms avoid ending in a credit bust.
- Macroprudential tools with significant effects include:
  - capital-based tools;
  - borrower-based tools such as loan-to-value (LTV) and debt-service-to-income (DSTI) caps;
  - reserve requirements.
- When credit booms are accompanied by capital flow surges, in addition to capital-based macroprudential tools, capital controls on money market instruments, including short-term cross-border interbank lending, tend to reduce the likelihood of a bust.
- The paper focuses on whether MMs and CCs decrease the probability that a credit boom ends in a credit bust, i.e., whether these policies make the financial system more resilient by creating buffers at the level of lenders and borrowers.

### Data, sample, and empirical approach
- Dataset coverage:
  - Annual frequency from 1996-2016.
  - 53 countries: 27 emerging markets and 26 advanced economies.
- Combination of several existing datasets to construct a granular cross-country database on macroprudential measures (MMs) and capital controls (CCs).
- Identification strategy requires a forward-looking window of at least three years to determine whether a credit boom ended in a bust; policy interventions analyzed only through 2016 (credit data for booms/busts extends to 2019).
- Prudential policy coding:
  - Policy loosening coded as -1;
  - Policy tightening coded as 1;
  - Unchanged policy coded as 0.
- Five source databases used for prudential measures: Shim et al. (2013), Akinci and Olmstead-Rumsey (2018), Cerutti et al. (2017a), Cerutti et al. (2017b), and Alam et al. (2019). Alam et al. (2019) is the primary source, cross-checked against others and national sources for discrepancies.

### Relation to existing literature and contribution
- Distinction from prior studies:
  - Many previous studies analyze short-term (one to two year) effects of policy tools on credit or asset price growth.
  - This paper emphasizes the second objective of macroprudential policy: making the financial system more resilient by building buffers—an effect harder to test and the central focus here.
- Closest methodological antecedent: Dell’Ariccia et al. (2016) — similar focus on which credit booms end in busts, but differences in data sample, definitions of booms/busts, and inclusion of capital controls.
- Literature synthesis highlights robust findings across studies:
  - Caps on LTV and DSTI, ceilings on credit growth, reserve requirements, and countercyclical provisioning reduce procyclicality between credit and GDP growth (Lim et al. (2011) and others).
  - Reserve requirements and borrower-based measures repeatedly shown to be effective in taming credit growth in multiple country settings.
  - Evidence on capital controls is mixed: some studies find little direct impact on aggregate capital inflows or credit growth, while others document specific roles (e.g., reducing FX loan shares or external debt reliance).

### Policy implications and recommendations
- Prefer domestic prudential measures (e.g., LTV, credit growth limits, reserve requirements) to restrain aggregate credit intensity; capital controls can complement by reducing other vulnerabilities (e.g., share of FX loans, reliance on external debt).
- In episodes with capital flow surges accompanying credit booms, combine capital-based macroprudential tools with capital controls on money market instruments (including short-term cross-border interbank lending) to reduce bust probability.
- Building buffers during booms—through capital-based tools, borrower-based limits (LTV, DSTI), and reserve requirements—can increase resilience even if moderation of credit growth is not fully achieved.

### Empirical elements reported (selected)
- Figures and tables cataloged in the source include dynamics of macroprudential policies and capital control measures, policy tightening vs. loosening, change in credit to GDP, credit and private capital flow cycles, use of macroprudential and capital control measures during credit booms, and distinctions between good and bad credit booms.
- Tabulated metrics include use of macroprudential policy and capital inflow controls by instrument (1996-2016), net and gross policy tightening measures (sum and mean over booms), and net policy tightening during booms with surges.

*Source: IMF Working Papers — Macroprudential Policies and Capital Controls Over Financial Cycles (wpiea2023171-print-pdf).*

### introduction until they are discontinued (if applicable), but it does not reflect any changes (loosening or

### wpiea2023171-print-pdf - introduction until they are discontinued (if applicable), but it does not reflect any changes (loosening or

### Construction of macroprudential measures (MPMs)
- Data harmonization steps:
  - Convert quarterly and monthly datasets to annual frequency.
  - Redefine variables from Cerutti et al. (2017a) by first-differencing to capture policy changes rather than stock levels.
  - Group measures into 8 categories: reserve and liquidity requirements (RR); limits on credit growth (CG); loan-to-value caps (LTV); debt-to-income or debt service-to-income ratios (DTI); capital instruments (CAP); provisioning (PR); exposure and concentration limits, including on interbank exposure (EXP); and other measures, including taxes, consumer loans measures (OTH).
- Coding:
  - For each category, indicators take values in {-1,0,1}.
  - Overall MPM index = sum over the eight categories; index ranges from -8 to +8 and captures net tightening in a year.
- Limitations noted:
  - Intensity of instrument use is not fully captured.
  - Dataset controls for changes in instruments but not for the overall level of restrictions already in place.
  - Summation across subcategories does not account for differing economic impacts across categories.

### Mapping and aggregation decisions
- Where databases reported related sub-instruments (e.g., reserve vs liquidity requirements), merged by summing values when data available for at least one variable; sign of sum indicates tightening (1), loosening (-1) or no change (0).
- Final geographical coverage (after harmonization): Asia-Pacific (13 countries), Europe (27), North & Latin America (9), Middle East and Africa (4).

### Capital controls (CCs) dataset and construction
- Based on Fernández et al. (2015) dataset (annual frequency, 1996 to 2016) using de jure information from AREAER.
- Original coding: presence of a restriction = 1, absence = 0; dataset distinguishes inflows vs outflows.
- Focus of this paper: capital inflows only. For each inflow instrument, first-difference the annual stock series to obtain flow-equivalents; recode differences into {-1,0,1} to denote loosening, no change, tightening respectively.
- Asset/instrument aggregation:
  - Original 10 asset categories: eq, bo, mm, ci, de, fc, cc, gs, di, re.
  - For eq, bo, mm, ci, de: two measures (purchase locally by non-resident; sale/issue abroad by residents) averaged, then recoded to {-1,0,1}.
  - Merged 10 categories into 8 instrument types: commercial credit (cc) merged with financial credit (fc); guarantees (gs) merged with derivatives (de); others mapped to 8 instrument categories (see source mapping).
- Overall CC (inflows) index = sum of individual categories; ranges from -8 to +8.

### Macroeconomic control variables and data sources
- Credit to the private sector: IMF’s International Financial Statistics (IFS), extended with Haver Analytics when needed.
- Capital flows: IMF’s Financial Flows Analytics (FFA) database.
  - Gross private (non-official) capital inflows = total capital inflows minus other inflows to official sector (other liabilities of central bank, government, and SDRs).
  - Debt-creating inflows = portfolio debt inflows + other types of flows into non-official sector (banking flows, non-financial corporate debt).
- Real GDP growth and nominal GDP levels: IMF’s World Economic Outlook Database (WEO).

### Evolution of policies over time (descriptive)
- General patterns (1990s–post-2008):
  - Late 1990s: net loosening of both macroprudential and capital account restrictions.
  - 2000s: divergence—capital inflow restrictions continued to be removed up to the global financial crisis; macroprudential policies tightened between 2002 and 2007 in response to vulnerability accumulation.
  - Global financial crisis 2008/2009: macroprudential policies loosened across sample; capital controls largely unchanged.
  - Post-crisis: both policy types tightened in net terms; macroprudential steep increase driven by new frameworks and Basel III implementation; renewed use of capital inflow controls in some countries.
- Instrument use frequency (1996-2016):
  - Mac prudential: reserve requirements (RR) most actively used, followed by capital-based tools (CAP) and LTV caps; limits on credit growth (CG) least used.
  - Capital controls: overall fewer policy changes (~half the number of macroprudential policy changes). Controls on derivatives most commonly used, followed by short-term money market (mm) and equity flows.

### Identification of credit and capital cycles
- Credit boom definition (real-time, using changes in credit-to-GDP ratio):
  - Must last at least 2 years.
  - At least one year with change > 3 ppts.
  - All other years of the boom: change > 2 ppts.
  - Exclude years where increase > 2 ppts was driven by GDP contraction (real GDP growth decline ≥ 1%).
  - Combine adjacent booms if gap ≤ 1 year and nominal credit growth in gap year is positive.
  - Macroeconomic and credit data used from 1990 where available (policy data from 1996).
- Credit bust definition:
  - Must last at least 2 years.
  - At least one year with change < -3 ppts.
  - All other years of the bust: change < -1 ppts.
  - Merge two busts with a gap ≤ 1 year if nominal credit growth in gap year is negative.
- Threshold rationale:
  - 3 ppts threshold ≈ 70th percentile of distribution for booms.
  - -3 ppts threshold ≈ 15th percentile for busts (more extreme).
- Sample and descriptive cycle statistics:
  - 72 credit boom episodes identified over 1996-2016 for 53 countries.
  - 26 of the 72 booms ended in a bust within 3 years (about 36%).
  - Average duration of a credit boom: 7 years (min 2 years, max 14 years).
  - Average duration of a credit bust: 4 years (min 2 years, max 10 years).
  - About three quarters of booms last 2–11 years; three quarters of busts last 2–5 years.
  - Average absolute annual change in credit-to-GDP in booms and busts ≈ 7 percentage points; normal times annual change ≈ 1 percentage point.
- Capital inflow surge definition:
  - A surge occurs when gross private capital inflow-to-GDP ratio exceeds its 5-year rolling mean by one standard deviation (Forbes and Warnock (2012) method).
  - No minimum duration restriction for surges.
  - Over 170 surge episodes identified; average surge duration slightly below 2 years.
  - Of 72 credit booms, 51 coincide with a surge in private capital inflows within the first three years of the boom.

### Policy use over cycles (descriptive findings)
- Macroprudential policy during booms:
  - Countercyclical on average: tightened during booms more often than loosened.
  - Out of 72 booms: 35 had net MPM tightening, 24 had no action, 13 had net loosening (net loosening cases often involved some tightenings and some loosenings across instruments).
  - Tighter macroprudential policies on average when a credit boom coincides with a capital surge.
- Capital controls during booms:
  - More gross tightening (and less gross loosening) of capital controls during booms accompanied by capital surges vs booms without surges.
  - Net effect: CCs are mostly acyclical on average; zero net tightening during booms with surges, some net loosening during booms without surges.
- Advanced vs emerging market patterns:
  - Much macroprudential tightening in advanced economies occurred during normal times (driven by Basel II/III and post-crisis regulation).
  - Emerging markets are on average more active in using macroprudential policies during credit booms than advanced economies.

### Econometric assessment of policy effectiveness during booms
- Approach:
  - Event-study/probit model using 72 cross-sectional boom episodes.
  - Dependent variable: dummy = 1 if boom followed by bust within 3 years, 0 otherwise.
  - Controls include average credit-to-GDP change during boom and boom duration (years).
  - Policy variables: cumulative (sum over boom) net tightening of MPMs and CCs (inflows); alternative specifications use mean per year and gross tightening.
- Key empirical findings:
  - Larger and longer credit booms are more likely to end in a bust.
  - MPMs reduce the probability of a bad boom; CCs not significant in baseline specifications.
  - Evaluated at sample means (average probability of a bad boom = 33%):
    - Net tightening of MPMs by 1 (one instrument tightened net during the boom) reduces the probability of a bad boom by about 7% on average.
  - Disaggregated instrument effectiveness (economic significance order and marginal effects reported):
    - Capital instruments (CAP) and DTI caps: tightening by one decreases probability of a bad boom by 22% (each).
    - LTV caps: tightening by one decreases probability by 18%.
    - Reserve requirements (RR): tightening by one decreases probability by 11%.
  - Regression coefficient highlights (Table 3 baseline, cum. measures):
    - Credit-to-GDP change (avg): coefficient ≈ 0.113*** (standard error 0.0430) in column (1).
    - Boom duration: coefficient ≈ 0.0944* (standard error 0.0525) in column (1).
    - MPMs (cum.): -0.184** (standard error 0.0776) (significant).
    - CCs on inflows (cum.): not statistically significant in main specifications (e.g., -0.0557 with standard error 0.0996).
    - Observations: 72 in many specifications; R2-pseudo values range (e.g., 0.195 in column (1)).
- Robustness checks:
  - Using mean (per year) MPM measures instead of sum: broadly similar results, with DTI caps losing significance in some specifications.
  - Using gross tightening (ignoring loosenings) for sum and mean to compare with binary 0-1 codings; results reported in supplementary tables (Table A5, Table A6 in source).

*Source: wpiea2023171-print-pdf*

### Appendix confirm a role for capital instruments and LTV caps.

### Appendix confirm a role for capital instruments and LTV caps

### Robustness checks and sample definitions
- Stricter definition of credit booms and credit busts reduces identified booms to 53, out of which 17 ended in a bust.
- A focused subset for booms accompanied by private capital inflow surges contains 51 booms, out of which 18 ended in a bust.
- Dataset constructed covers 53 countries for two decades from 1996 to 2016.

### Main robustness results
- Capital-based macroprudential instruments show a consistent role in decreasing the probability of credit booms ending in busts across alternative specifications.
- Borrower-based tools and reserve requirements also appear effective in many specifications, but significance of individual borrower-based tools (LTV, DTI) varies across specifications:
  - In some specifications DTI caps (rather than LTV caps) turn significant; in others neither DTI nor LTV are significant.
- Additional robustness checks controlling for monetary policy, level of economic development, economic activity, external imbalances, and global volatility yield:
  - Changes in policy rates during booms: no significant impact on the probability of a boom ending in a bust.
  - Country development dummy: significant and positive, implying booms in emerging market economies are more likely to end in a bust.
  - Real GDP growth dynamics and current account balances during booms: no significant impact.
  - Global volatility (VIX): can increase probability depending on specification, but no qualitative impact on main conclusions.

### Results for booms with capital surges
- For booms accompanied by capital inflow surges, macroprudential measures such as capital instruments, caps on DTI, and reserve requirements continue to decrease the probability of a credit boom ending in a bust; exposure limits are also significant in this subset.
- Total index of capital controls is not significant, but individual tightening of controls on inflows via money market instruments can play a statistically significant role in decreasing the probability of a bad boom.
- When focusing on debt inflow surges (part of total private capital inflows), results are qualitatively similar.

### Key quantitative estimates and economic significance
- Net tightening of a single macroprudential instrument during a credit boom accompanied by a capital flow surge reduces the probability of a boom ending in a bust by around 11% from the initial 29% probability when evaluated at the means of explanatory variables.
- Effects of individual instruments (reported magnitudes for the subset of booms with surges):
  - Capital instruments: 31%
  - Caps on DTI: 29%
  - Reserve requirements: 15%
  - Exposure limits: 35%

### Specification checks and additional findings
- Expressing policy variables as means (rather than sums) over the boom confirms initial findings for capital instruments and DTI caps, and for controls on money market instruments among capital management measures.
- Using gross rather than net tightening: macroprudential tool effects hold for capital and exposure limits, and LTV caps become significant; capital controls do not appear significant in these specifications.
- Alternative capital control measures tested:
  - Adding controls on capital outflows alongside controls on inflows, or replacing inflow measure with a net overall index, yields CCs on outflows and overall index not significant and does not affect significance or magnitude of other variables.
- Robustness checks overall confirm a strong role for macroprudential instruments and money-market-oriented capital flow management tools for credit booms accompanied by capital flow surges; more detailed research needed on individual instrument channels.

### Conclusion summary (paper-level)
- Macroprudential policies are effective in reducing the probability that credit booms end in credit busts, both with and without capital surges.
- Capital controls on short-term money market instruments, including cross-border interbank lending, tended to reduce the likelihood of a bust in booms accompanied by surges.
- Capital-based macroprudential tools, borrower-based tools (caps on DTI and LTV), and reserve requirements appear to contribute to financial system resilience, though their significance varies across specifications.
- Open questions for follow-up research include:
  - Additional country characteristics or boom features that condition effectiveness (institutional, regulatory, governance differences; type of credit driving the boom; initial financial integration; degree of initial financial development).
  - Optimal sequencing and combinations of macroprudential instruments and capital controls over credit booms.
  - Which macroprudential policies or capital controls reduce the probability of particularly harmful busts (e.g., those accompanied by banking crises).
  - Effect of additional policies (fiscal or exchange rate) on reducing the probability of a credit boom ending in a credit bust.

*IMF Working Paper — Appendix excerpts from the PDF chapter provided*

### References

### wpiea2023171-print-pdf - References

### Major thematic areas in the references
- Empirical and theoretical studies on capital flows and emerging market economies: Ahmed and Zlate 2014; Binici, Hutchison, and Schindler 2010; Forbes and Warnock 2012; Martínez 2015.
- Macroprudential policy design, instruments, and effectiveness: Akinci and Olmstead-Rumsey 2018; Borio and Shim 2007; Cerutti, Claessens, and Laeven 2017; Galati and Moessner 2018; Lim et al. 2011; Poghosyan 2019.
- Macrofinancial stability, credit booms, and housing markets: Arena et al. 2015; Dell’Ariccia et al. 2012; Dell’Ariccia, Igan, Laeven, and Tong 2016; Kuttner and Shim 2013; Vandenbussche, Vogel, and Detragiache 2012.
- Capital controls, capital flow management measures, and datasets: Fernández et al. 2015 (two entries); Qureshi et al. 2011; Habermeier, Kokenyne, and Baba 2011.
- Cross-country databases and measurement work on macroprudential actions: Budnik and Kleibl 2018; Cerutti et al. 2017; Schindler 2009; Shim et al. 2013; Alam et al. 2019.
- Policy-oriented IMF and multilateral outputs on macroprudential frameworks and capital flows: IMF 2012; IMF 2013; IMF 2015; IMF 2017; IMF GFSR 2011; CGFS 2010; Das, Gopinath, and Kalemli-Özcan 2022.

### Appendix — sample composition (Table A1)
- Countries included in the sample are split into two columns:
  - Emerging markets: Argentina, Brazil, Bulgaria, Chile, China, Colombia, Czech Republic, Hungary, India, Indonesia, Latvia, Lebanon, Malaysia, Mexico, Nigeria, Peru, Philippines, Poland, Romania, Russia, Slovenia, South Africa, Thailand, Turkey, Ukraine, Uruguay, Vietnam.
  - Advanced economies: Australia, Austria, Belgium, Canada, Denmark, Finland, France, Germany, Greece, Hong Kong SAR, Iceland, Ireland, Israel, Italy, Japan, Korea, Netherlands, New Zealand, Norway, Portugal, Singapore, Spain, Sweden, Switzerland, United Kingdom, United States.

### Appendix — macroprudential measures definitions (Table A2)
- Reserve and liquidity requirements (RR): Requirements for the banks to hold at least a fraction of their liabilities with the central bank, typically differentiated by the type, maturity and currency of deposits and other funding; imposed on stocks or flows of liabilities (marginal reserve requirements). Liquidity tools are aimed at mitigating banks’ liquidity risks and are typically set as a minimum ratio of liquid assets relative to total assets or selected (less stable) liabilities. Both reserve and liquidity requirements affect the funds available for lending on to the private sector.
- Limits on credit growth (CG): Quantitative limit set directly on the growth of credit over a specified period or a maximum increase in lending over specific period of time.
- LTV caps (LTV): Caps on loan-to-value ratio limits the maximum amount that can be lent to the borrowers against their (typically real estate) collateral. An LTV ratio of 70% would imply lending to fund 70% of the purchase value (the rest being a downpayment), while 100% would allow for full lending equivalent to the value of the collateral.
- DTI or DSTI caps (DTI): Caps on debt-to-income (or debt-service-to-income) ratio limits the size of debt (or debt service payments) relative to household income.
- Capital instruments (CAP): Measures aimed at affecting the capitalization levels of banks. Countercyclical capital buffers require banks to hold more capital during upturns. Leverage ratios limits the banks from exceeding a fixed minimum leverage ratio. Capital surcharges on Systemically Important Financial Institutions force them to hold a higher capital level than other financial institutions. Sectoral capital buffers or higher risk weights on various types of exposures also belong here, as they require higher capital to be held against such exposures.
- Provisioning (PR): Adjustments (typically an increase) of specific provisions created for bad loans beyond traditional provisioning rates, such as for high LTV loans, loans in foreign currency or loans to certain sectors; adjustments to general provisioning rates, including a dynamic (countercyclical) element that requires banks to set aside reserves from profits in good times in order to cover realized losses from borrower default in bad times.
- Exposure and concentration limits (EXP): Limit the fraction of assets held by specific borrowers or sectors (concentration limit) or fraction of liabilities held by other banks (interbank exposure limit).
- Other (OTH): Residual category including various tax measures applied on financial institutions as well as selected consumer loan measures.

### Appendix — dataset mappings (Table A3 and Table A4)
- Table A3 maps defined macroprudential measures to source databases and codes. Codes column lists measure codes: RR, CG, LTV, DTI, CAP, PR, EXP, OTH, LFX.
- Table A3 metadata entries preserved exactly as in source:
  - Number of countries65605011964134
  - Time period1995-20141990-20122000-20142000-20132000-20141990-2016
  - FrequencyAMQAQM
- Table A4 maps capital control (CC) indicators and inflow restriction components, preserves variable names and descriptive items such as:
  - Overall FKRSU_i(eq+bo+mm+ci+de+ccfc+di+re)/8
  - Components listed: e q_plbn, e q_s iar, bo_plbn, bo_siar, mm_plbn, mm_s iar, ci_plbn, ci_s iar, de_plbn, de_s iar, gsi, cci, fci, didii, rere_plbn
  - Asset types listed explicitly: Collective investments (ci), Equity (eq), Bond (bo), Money market (mm), ccfc, de (Derivatives)

### Appendix — empirical tables: key dependent variable and coefficients
- Tables A5 and A6: Dependent variable defined as "probability of a credit boom ending in a bust (dummy that equals 1 when a credit boom was followed by a bust within 3 years after the boom and 0 otherwise)." Source: Authors’ estimates.
- Selected coefficient entries preserved exactly as presented (examples):
  - Table A5 (Gross policy tightening: sum over the boom):
    - Credit-to-GDP change (avg)0.106***0.108***0.0916**0.100**0.138***0.114**0.0902**0.0928**0.0766*0.0921**
    - Boom duration0.128**0.119**0.109**0.114**0.127**0.105**0.138**0.106**0.125**0.105**
    - MPMs (cum.)-0.137**-0.145**
    - CCs on inflows (cum.)0.1250.1240.09650.08410.07740.2380.09880.08550.0960
    - Observations72727272727272726572
    - R2-pseudo0.1660.1770.1360.1500.1840.1510.2180.1250.1290.123
  - Table A6 (Gross policy tightening: mean over the boom):
    - Credit-to-GDP change (avg)0.118***0.111**0.0846**0.0960**0.184***0.115**0.0937**0.0880**0.07320.0860**
    - Boom duration0.0853*0.101*0.118**0.129**0.112**0.110**0.125**0.135**0.143***0.128**
    - MPMs (avg)-0.452*-0.413
    - CCs on inflows (avg)0.925*1.008**0.942*0.8900.879*1.289**1.043**1.125*1.000**
    - Observations72727272727272726572
    - R2-pseudo0.1550.1940.1760.1910.2330.1800.2320.1680.1760.164
- Tables A7 and A8: Net policy tightening in strong booms (sum and mean over the cycle). Dependent variable same as above. Selected preserved entries:
  - Table A7 (sum over the cycle):
    - Credit-to-GDP change (avg)   0.133***0.134***0.127***0.119**0.147***0.161***0.122**0.130***0.108**0.117**
    - Boom duration0.05890.05890.07900.1050.1020.04760.1200.138*0.180*0.0820
    - MPMs (cum.)-0.153*-0.151
    - CCs on inflows (cum.)0.07820.09280.1080.1040.03640.1630.2070.06010.119
    - Observations53535353535353534853
    - R2-pseudo0.1910.1940.1770.1460.1800.2230.2020.1960.1670.163
  - Table A8 (mean over the cycle):
    - Cre dit-to-GDP change (avg)   0.132***0.133***0.129***0.122**0.134**0.158***0.134**0.127**0.107*0.120**
    - Boom duration0.06500.07410.1010.1080.09800.06080.08960.150*0.200**0.118
    - MPMs (avg)-0.256-0.232
    - CCs on inflows (avg)0.5190.6130.5180.5270.3680.8860.7300.3520.578
    - Observations53535353535353534853
    - R2-pseudo0.1600.1760.1940.1640.1660.1930.2550.1910.1960.159
- Statistical notation preserved exactly: Standard errors in parentheses; significance markers *** p<0.01, ** p<0.05, * p<0.1 where shown.

*Content unit: wpiea2023171-print-pdf - References*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2023/english/wpiea2023171-print-pdf.pdf_