## wpiea2020067-print-pdf — References and selected sections

## Source details

**Canonical URL:** [wpiea2020067-print-pdf — References and selected sections](https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020067-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2020/english/wpiea2020067-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2020/english/wpiea2020067-print-pdf.pdf.json)

---

### I. Introduction — scope, goals, and headline findings
- Objective: construct a novel database of empirical findings on the effects of macroprudential policy (MPM) and undertake a forensic examination of the existing evidence.
- Database scope and composition:
  - 6,627 results from fifty-eight papers.
  - Results split evenly between cross-country and country-specific studies.
  - Database captures effects by: (i) outcome variable, (ii) instrument, (iii) measurement of MPM, (iv) categories of controls, (v) time horizon, (vi) sample (country and period), and (vii) methodology and unit of analysis.
  - Includes actual estimated coefficients and standardized coefficients (effects expressed in standard deviations of the outcome variable).
- Meta-analysis focus questions:
  - Average effects of MPMs on credit, household credit, and house prices.
  - Heterogeneity by instrument type, country setting, and unit of analysis (macro vs. micro).
  - Unintended effects: leakages, spillovers, and costs to economic activity.
- Headline empirical findings:
  - Tightening of MPM has statistically significant effects on credit.
  - Stronger effects found for liquidity measures.
  - Tightening LTV or DSTI produce similar average effects on reducing household credit but weaker and imprecise effects on house prices.
  - Housing and liquidity-based measures appear to have larger average effects in emerging markets, with wider confidence bands.
  - Average effects up-to-three times larger in micro-level studies than aggregate studies.
  - Statistically significant average leakages and spillover effects documented.
  - Statistically significant and negative effects on economic activity, at least in the near term.
  - Evidence suggests larger and quantitatively stronger effects for tightening than loosening actions.
  - Persistent impacts documented; a large fraction of effects observed within the first year.
- Robustness: findings robust to controlling for selection bias and journal publication status.

### II. Construction of the database — empirical framework and scope
- Empirical framework used across studies:
  - General form: y_it = ˆβ · MPM_it + ζ · X_it + u_it.
  - Unit i: aggregate (country) or micro (banks, firms, loans); period t: month, quarter, or year.
- Comparative scope: 6,627 estimates across fifty-eight papers; broader than some prior meta-analyses.
- Inclusion rule: included all estimated ˆβ coefficients within selected papers, excluding specifications with interactions (no linear-combination tests reported).

### III. Paper selection methodology
- Three-method selection:
  - Methods 1 & 2: collect empirical papers cited in recent literature reviews.
  - Method 3: Google Scholar search for "effectiveness of macroprudential policies" among first 100 hits.
  - Snowballing through references until no new empirical papers found.
- Final set: excludes 8 duplicate papers; broadly balanced between published and unpublished papers.

### IV. MPM measurement and standardization
- Measurement approaches in the literature:
  - Cross-country policy-setting dummies: (0,1) indicator of instrument in place.
  - Direction-of-change discrete variable: (-1,0,1) for loosening, neutral, tightening.
  - Intensive-margin measures: intensity/size of policy changes (levels or changes).
- Composite MPM measures categorized into: (i) housing, (ii) non-housing (broad-based, liquidity, other), and (iii) general.
- Standardization approach:
  - Normalize ˆβ and standard errors by standard deviation of the corresponding outcome variable.
  - For discrete MPMs: effect interpreted as standard deviation change in outcome in response to policy change.
  - For intensive MPMs: effect standardized by outcome SD and, when available, SD of MPM change; interpreted as standard deviation change in outcome per unit change in policy.
  - Visual evidence: unstandardized effects widely dispersed (Figure 2a); standardized distribution approximately normal with effect sizes concentrated in the 0 to -0.2 standard-deviations region (Figure 2b).
- Note on standardized standard errors: dividing standard error by outcome SD is a good approximation under certain conditions (Yuan and Chan (2011)).

### V. Descriptive statistics — database composition and raw magnitudes
- Study composition:
  - 43 percent: cross-country macro-level.
  - 10 percent: cross-country micro-level.
  - 12 percent: country studies macro-level.
  - 34 percent: country studies micro-level.
- Common controls (percent of results):
  - Other MPMs: 61 percent.
  - Interest rates: 77 percent.
  - Other macro variables: 61 percent.
  - Cross-sectional fixed effects: 80 percent.
  - Time fixed effects: 35 percent.
- Empirical approaches: panel-data techniques, lagged effects, some GMM/IV; micro-level: event-study or difference-in-difference.
- Time horizons: most studies focus on effects up to a year; some use VARs or local projections for longer horizons.
- Statistical significance:
  - About half of 6,627 coefficients pertain to household credit, house prices, total credit; heterogeneous and mixed significance across tools and outcomes.
- Magnitudes — selected standardized short-term (up to one year) statistics (Table 1 entries preserved exactly):
  - Panel: MPM in place (0,1):
    - Balance sheet fragility: Obs. 249; Mean -0.02; Std. Dev. 0.09; Min -0.94; Max 0.21
    - Capital inflow: Obs. 780; Mean 0.05; Std. Dev. 0.19; Min -0.83; Max 0.77
    - Corporate credit: Obs. 51; Mean -0.01; Std. Dev. 0.54; Min -1.69; Max 2.24
    - Credit: Obs. 190; Mean -0.38; Std. Dev. 0.71; Min -3.52; Max 1.90
    - Economic activity: Obs. 300; Mean 0.13; Std. Dev. 0.55; Min -1.67; Max 0.95
    - House price: Obs. 157; Mean -0.09; Std. Dev. 0.39; Min -2.06; Max 1.19
    - Household credit: Obs. 176; Mean -0.15; Std. Dev. 0.29; Min -1.23; Max 0.93
    - All Outcomes: Obs. 931; Mean -0.12; Std. Dev. 0.44; Min -3.52; Max 2.24
  - Panel: MPM change (-1,0,1):
    - Balance sheet fragility: Obs. 5; Mean -0.02; Std. Dev. 0.03; Min -0.05; Max 0.01
    - Bank default risk: Obs. 38; Mean -0.31; Std. Dev. 0.62; Min -3.35; Max 0.40
    - Capital inflow: Obs. 116; Mean 0.00; Std. Dev. 0.10; Min -0.37; Max 0.34
    - Corporate credit: Obs. 370; Mean 0.08; Std. Dev. 0.07; Min -0.11; Max 0.34
    - Credit: Obs. 168; Mean -0.07; Std. Dev. 0.45; Min -2.21; Max 2.40
    - Economic activity: Obs. 392; Mean -0.08; Std. Dev. 0.51; Min -5.36; Max 1.57
    - House price: Obs. 353; Mean -0.01; Std. Dev. 0.31; Min -1.44; Max 2.69
    - Household credit: Obs. 411; Mean -0.14; Std. Dev. 0.32; Min -1.59; Max 1.02
    - Non bank credit: Obs. 180; Mean 0.02; Std. Dev. 0.04; Min -0.01; Max 0.14
    - All Outcomes: Obs. 1538; Mean -0.07; Std. Dev. 0.38; Min -5.36; Max 2.68
  - Panel: MPM intensity (level or change in levels):
    - Balance sheet fragility: Obs. 14; Mean -0.25; Std. Dev. 0.23; Min -0.66; Max -0.01
    - Credit: Obs. 55; Mean -1.21; Std. Dev. 1.14; Min -5.91; Max -0.06
    - Economic activity: Obs. 1500; Mean 0.00; Std. Dev. 0.01; Min -0.04; Max 0.05
    - House price: Obs. 238; Mean -0.28; Std. Dev. 0.20; Min -0.94; Max 0.40
    - Household credit: Obs. 1680; Mean 0.01; Std. Dev. 0.04; Min -0.22; Max 0.20
    - All Outcomes: Obs. 625; Mean -0.21; Std. Dev. 0.50; Min -5.91; Max 0.40

### VI. Symmetry (tightening vs easing)
- Sample evidence on direction:
  - About sixty percent of studies discuss tightening or loosening effects.
  - More than a quarter study only tightening actions.
  - Four studies study easing actions alone.
  - About 20 percent study both tightening and loosening.
- Statistical significance comparisons:
  - Among papers looking at tightening and/or loosening: 46 percent of results find tightening coefficients statistically significant at the 10 percent level, versus 36 percent for easing actions.
  - Within 13 papers analyzing both directions in comparable regressions: tightening coefficients significant in 61 percent of cases, easing in 37 percent.
- Magnitude comparisons:
  - About 57 percent of cases: estimated tightening coefficients larger in absolute value than easing coefficients.
  - Restricting to tightening coefficients statistically significant and negative (about 40 percent), they overwhelmingly exceed effects of easing actions.
  - In about 20 percent of cases, effects of easing are stronger or too close to call.
- Conclusion: tightening shows larger quantitative impact and is more often statistically significant than easing, partly due to more observations for tightening.

### VII. Meta-regressions — methodology and baseline results
- Meta-regression model:
  - ˆβ_jk = ∑_m θ_m · MPM_mjk + γ X_jk + δ · SE^2(β_jt) + ε_jk, estimated by weighted least squares (weights inverse of standard errors).
  - SE^2(β_jt) included to correct for publication/selection bias (Stanley and Doucouliagos (2014) approach).
- Baseline results for short-term effects of tightening (MPM change -1,0,1) on credit (selected Table 2 Column (1) entries preserved exactly):
  - Broad based: -0.045 (standard error 0.002) — significant at 1% (-0.045 ∗∗∗ (0.002)).
  - Housing: -0.041 (0.007) ∗∗∗.
  - Liquidity & Other: -0.118 (0.005) ∗∗∗.
  - Interpretation: tightening liquidity & other tools associated with a 0.12 standard deviation reduction in total credit; tightening housing and broad-based tools associated with 0.04 standard deviation reduction in total credit.
- Robustness: weighting to avoid study over-representation, publication-bias correction, controls for publication status and specification completeness, journal-quality weighting — patterns stable across checks.
- Existence-based (MPM in place (0,1)) effects (text summary): 0.2, 0.1 and 0.09 standard deviation reduction in credit from broad-based, housing, and liquidity/other tools respectively (may capture cumulative effects or cross-country differences).

### VIII. Heterogeneity of average effects — instruments, country setting, unit of analysis
- Instrument-specific average standardized effects (selected Table 3 entries preserved exactly):
  - LTV:
    - Credit: -0.065 ∗∗ (0.031)
    - Household Credit: -0.061 ∗∗ (0.027)
    - House Price: -0.007 (0.015)
  - DSTI:
    - Credit: -0.063 ∗∗ (0.025)
    - Household Credit: -0.052 ∗∗ (0.018)
    - House Price: 0.001 (0.011)
  - Capital Requirements:
    - Credit: -0.065 ∗∗ (0.030)
    - Household Credit: -0.030 (0.022)
    - House Price: -0.042 ∗∗ (0.018)
  - Loan Loss Provisions:
    - Credit: -0.030 (0.030)
    - Household Credit: -0.033 (0.021)
    - House Price: 0.016 (0.010)
  - Liquidity: Credit: -0.139 ∗∗∗ (0.030)
  - Other: Credit -0.035 (0.029); Household Credit -0.034 (0.020); House Price -0.002 (0.010)
- Country setting and unit of analysis (selected Table 4 entries preserved exactly):
  - EM (Column (1), Observations 198; r2 0.660):
    - Broad based: -0.044 ∗∗ (0.018)
    - Housing: -0.153 ∗∗∗ (0.042)
    - Liquidity & Other: -0.126 ∗∗∗ (0.011)
  - Mixed (Column (2), Observations 239; r2 0.388):
    - Broad based: -0.033 ∗∗ (0.011)
    - Housing: -0.034 ∗∗∗ (0.007)
    - Liquidity & Other: -0.030 ∗∗∗ (0.008)
  - Micro (Column (3), Observations 176; r2 0.694):
    - Broad based: -0.045 ∗∗ (0.018)
    - Housing: -0.192 ∗∗∗ (0.009)
    - Liquidity & Other: -0.130 ∗∗∗ (0.007)
  - Macro (Column (4), Observations 261; r2 0.229):
    - Broad based: -0.032 ∗ (0.015)
    - Housing: -0.039 ∗∗∗ (0.009)
    - Liquidity & Other: -0.030 ∗∗∗ (0.009)
- Key heterogeneity findings:
  - Larger average effects in EMs for housing and liquidity measures, though EM results have wider confidence intervals.
  - Micro-level studies show substantially larger effects (e.g., tightening housing tools reduce credit for banks or firms by 0.2 standard deviations vs 0.04 in aggregate).
  - Explanations: micro studies’ higher identification power, focus on constrained agents, and leakages/spillovers that reduce aggregate transmission.

### IX. Leakages, spillovers, and net effects
- Meta-estimates on cross-border and non-bank lending responses (Table 6 selected preserved entries):
  - Broad based: 0.054** (0.015) ; 0.049** (0.017) ; 0.066** (0.024)
  - Housing: 0.004* (0.002) ; 0.005*** (0.001) ; 0.005*** (0.000)
  - Liquidity & Other: 0.060*** (0.000) ; 0.059*** (0.001) ; 0.077** (0.024)
  - SE sq. (Pub. Bias Correction): -0.190 (0.847)
  - Observations: 59, 59, 59 ; r2: 0.407, 0.365, 0.445
- Interpretation:
  - All measures associated with leakages/spillovers; strongest for broad-based and liquidity measures.
  - Positive sign indicates tighter MPM increases cross-border and/or nonbank lending.
- Back-of-the-envelope netting:
  - Subtracting average leakages from average micro-level effects suggests a net negative effect of tightening MPM on credit.
  - For liquidity, net effect approx. -0.05 standard deviations, roughly equivalent to average macro-effects of -0.03 standard deviations (Table 4).
  - Suggests leakages reduce but do not fully eliminate MPM impacts.

### X. Impact on economic activity and dynamics
- Meta-regression average effects on cross-sector index (Table 7 preserved exactly):
  - Credit: Cross-Sector Index = -0.011*** (0.003)
  - House Price: Cross-Sector Index = 0.002 (0.003)
  - Economic Activity: Cross-Sector Index = -0.004*** (0.000)
  - Observations: 80 (Credit), 55 (House Price), 143 (Economic Activity)
  - r2: 0.643, 0.216, 0.095
- Interpretation:
  - Tightening MPMs have a negative and statistically significant effect on economic activity at the 1 percent level.
  - Coefficient -0.004 standard deviations on economic activity is about one-third of the effect on credit in the same table.
- Selected case estimates from database:
  - Richter, Schularick, and Shim (2019): Following a 10 percentage points reduction in LTV limits, real GDP declines by 0.5 percent after 2 years and 1.1 percent after 4 years.
  - Alam and others (2019): Tightening up to 10 percentage points in LTV associated with 1.5 percent decline in real consumption growth; tightening between 10-25 percentage points associated with 1.1 percent decline.
  - Kim and Mehrotra (2018, 2019): Comparable effects on real GDP, private consumption, and investment between a one-standard-deviation shock to an index of MPMs and a one-standard-deviation monetary policy tightening.
- Dynamics and persistence:
  - Peak effects often at 1-2 years for credit, and 2-3 years for macro variables (GDP, consumption, prices).
  - First-year effect typically between 20 percent and 80 percent of the peak effect.
  - Appendix Table A10: contemporaneous effects larger than lagged in many specifications; evidence of hump-shaped responses.
  - Non-trivial fraction of total effects occurs beyond the first year.

### XI. Reconciling macro and micro estimates — power, leakages, mapping to raw magnitudes
- Statistical power:
  - Using conventional calibration (5% significance, 80% power): standard error needs to be smaller than true effect divided by 2.8.
  - Based on meta-analysis average effects:
    - Almost 80% of macro-level studies are under-powered to detect standardized effects below 0.05.
    - Only 20% of micro-level studies are under-powered.
    - Precision of effects in micro-level studies is almost five times larger than macro studies.
- Mapping standardized to raw effects (Table 5 examples preserved exactly):
  - y-o-y quarterly (Macro): Average S.D. of Outcome = 13% ; Std. Effect Size = -0.05 ; Effect Size = - 0.6 p.p.
  - q-o-q quarterly (Micro): Average S.D. of Outcome = 30% ; Std. Effect Size = -0.15 ; Effect Size = - 4.5 p.p.
- Guidance: multiply standardized effect by average SD of the outcome to obtain unstandardized magnitude.

### XII. Robustness, selection bias, and corrections
- Funnel-plot and regression evidence (Table A3 preserved exact coefficients):
  - Significant negative association between estimated effects and their standard error in multiple specifications, indicating selective reporting, particularly in macro studies.
  - Example coefficients: Standard Error coefficient = -0.737 ∗∗∗ (0.099) in Column (1); other columns show similar negative and significant coefficients.
- Robustness checks reported:
  - Excluding imputed summary statistics and specific papers: housing and liquidity effects remain significant in many exclusions (Table A6 examples).
  - Selection-correction (Andrews and Kasy (2019), Table A9) yields qualitatively similar patterns; many effects remain statistically significant (selected exact coefficients reproduced in Appendix summary).

### XIII. Conclusions and suggested future research directions
- Main empirical conclusions:
  - On average, MPM tools have statistically significant effects on credit.
  - Effects larger in micro-level studies than macro-level studies due to identification, statistical power, and leakages/spillovers.
  - Significant heterogeneity across country settings and instruments; effects stronger among EMs with larger confidence intervals.
  - Evidence of asymmetry: tightening stronger than easing in magnitude and statistical significance.
  - Tightening MPMs associated with negative near-term impacts on economic activity.
- Suggested future research:
  - Improve measurement of MPM actions to better capture the intensive margin.
  - Increase research using micro-level data, which appears more adequately powered.
  - Expand evidence on easing MPMs (leveraging COVID-19 policy relaxations) to understand effects in downturns.
  - Further explore nonlinearities and interactions with other policies.

*Source: wpiea2020067-print-pdf — References and selected sections (excerpted content provided).*

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

### References (wpiea2020067-print-pdf) — Selected content summary

### I. Introduction — scope, goals, and headline findings
- Macroprudential policy has been deployed in over one hundred countries using a wide range of instruments (Alam and others, 2019; Cerutti, Claessens, and Laeven, 2017).
- Recent actions have concentrated in relaxation measures as part of the policy response to the COVID-19 pandemic (Benediktsdóttir, Feldberg, and Liang, 2020; IMF, 2020).
- Paper objective: construct a novel database of empirical findings on the effects of macroprudential policy and undertake a forensic examination of the existing evidence.
- Database scope and composition:
  - 6,627 results from fifty-eight papers.
  - Results split evenly between cross-country and country-specific studies.
  - Database captures information on effects by: (i) outcome variable, (ii) instrument, (iii) measurement of macroprudential policy, (iv) categories of controls, (v) time horizon of effect, (vi) sample (country and period), and (vii) methodology and unit of analysis.
  - Database includes actual estimated coefficients and standardized coefficients (effects expressed in standard deviations of the outcome variable).
- Meta-analysis approach:
  - Quantitatively synthesize research on policy effects and examine how average effects vary with study or result characteristics.
  - Focus questions: average effects of macroprudential policies on credit, household credit, and house prices; heterogeneity by instrument type, country setting, and unit of analysis (macro vs. micro).
  - Also examines unintended effects (leakages, spillovers, costs to economic activity).
- Key empirical findings reported:
  - Tightening of macroprudential policy has statistically significant effects on credit.
  - Stronger effects found for liquidity measures.
  - Considerable variation across tools and outcomes; e.g., tightening LTV or DSTI produce similar average effects on reducing household credit but have weaker and imprecise effects on house prices.
  - Housing and liquidity-based measures appear to have larger average effects, but with wider confidence bands, in emerging markets.
  - Average effects are up-to-three times larger among studies using micro-level data than those using aggregate data.
  - Statistically significant average leakages and spillover effects documented.
  - Statistically significant and negative effects on economic activity, at least in the near term.
  - Evidence suggests somewhat larger and quantitatively stronger effects for tightening than loosening actions.
  - Persistent impacts documented, with a large fraction of effects observed within the first year.
- Robustness: findings robust to controlling for selection bias and journal publication status.

### II. Construction of the database — empirical framework and scope
- Empirical model framework used across studies (general form):
  - y_it = ˆβ · MPM_it + ζ · X_it + u_it
  - Unit of analysis i can be aggregate (country) or micro (banks, firms, loans); period t can be month, quarter, or year.
- Purpose: construct summaries of ˆβ using meta-analysis techniques for each outcome variable / MPM tool pair from a systematically selected set of papers.
- Comparative scale: broader than some prior meta-analyses (e.g., Gambacorta and Murcia (2019) covering 7 studies); comparable in scope to Magud, Reinhart, and Rogoff (2011) for diversity of tools/outcomes.

### III. Paper selection methodology
- Three-method selection procedure:
  - Method 1 & 2: collect all empirical papers cited in recent literature reviews (Galati and Moessner, 2018; Gambacorta and Murcia, 2019).
  - Method 3: Google Scholar search using "effectiveness of macroprudential policies" and record empirical papers among the first 100 hits.
  - Snowballing: iterate through references of identified empirical papers until no new empirical papers were found.
- Final selection notes:
  - Final set excludes 8 papers that appeared under multiple methods.
  - Final set broadly balanced between published and unpublished papers.
  - Appendix Table A12 shows the final list of papers in the database (not reproduced here).

### IV. Estimates and statistics selection
- Collection rule: included all estimated ˆβ coefficients within selected papers to reduce arbitrariness.
- One exception: exclusion of estimates from specifications with interactions between MPMs and other variables (no papers reported necessary linear-combination tests and standard errors).
- Results presented only in charts were requested from authors and included if provided.
- For each estimate the database records:
  - coefficient ˆβ and its standard error;
  - type of MPM tool and how it was measured;
  - outcome variable;
  - estimation methodology;
  - unit of analysis (macro or micro);
  - sample characteristics (e.g., EM countries or all);
  - whether the specification corresponded to the most complete one in the paper;
  - journal of publication if any.

### V. Classification of MPMs and outcome variables
- Tool taxonomy benchmark: IMF’s Integrated Macroprudential Policy Database (iMaPP).
  - iMaPP taxonomy includes 17 categories and 4 classes of subcategories.
  - Database mapping into broader groups per IMF (2014):
    - Broad-based tools: counter-cyclical capital buffers, conservation buffers, capital requirements, leverage limits, loan loss provisions, limits on credit growth, loan restrictions, limits on foreign currency loans.
    - Liquidity tools: liquidity coverage ratios, limits on loan-to-deposit ratio, limits on FX positions, reserve requirements.
    - Housing tools: housing sector specific measures including limits to loan-to-value ratios, limits to debt-service-to-income ratio, loan restrictions, and other sector-specific capital requirements, loan loss provisions, taxes and levies.
    - Other tools: measures on systemically important financial institutions, taxes and levies, and other measures.
  - If no information on whether a measure is targeted, it is classified as broad-based.
  - Note: total number of categories and subcategories used in the database is 33 while the iMaPP includes 27 (some finer distinctions captured when papers reported sub-targeting).
- Outcome variables grouped into 10 categories for parsimony:
  - balance sheet fragility, bank default risk, capital flows, corporate credit, credit, house price, household credit, economic activity, non-bank credit, and other outcomes.
- Appendix Table A1 provides mapping details of individual tools (not reproduced here).

### VI. Meta-analysis and heterogeneity considerations (methodology overview)
- Meta-analytic goal: summarize β estimates across heterogeneous specifications, instruments, outcomes, and samples.
- Attention to quality of estimates within and across papers; Section IV.A addresses concerns including selection bias and specification completeness.
- Database intentionally captures study characteristics to allow analysis of heterogeneity across:
  - instrument type (e.g., liquidity, housing, broad-based),
  - country setting (e.g., emerging markets vs. others),
  - unit of analysis (micro vs. macro),
  - direction of action (tightening vs. easing),
  - publication status (peer-reviewed vs. other).

### VII. Key quantitative magnitudes and patterns emphasized in the paper
- Database size: 6,627 results; fifty-eight papers.
- Relative effect sizes and patterns:
  - Average effects up-to-three times larger in micro-level studies compared to aggregate studies.
  - Large fraction of effects observed within the first year.
- Instrument-specific tendencies:
  - Liquidity measures: stronger effects on credit.
  - LTV and DSTI tightening: similar average effects on household credit; weaker and imprecise effects on house prices.
  - Housing and liquidity measures: larger average effects in emerging markets but wider confidence bands.
- Externalities:
  - Statistically significant average leakages and spillover effects consistent with cross-border or non-bank lending responses.
  - Statistically significant negative near-term effects on economic activity.
- Asymmetry:
  - More evidence supportive of stronger effects for tightening than for loosening actions.

*Source: wpiea2020067-print-pdf — References and selected sections (Introduction; Construction of the Database; Paper Selection; Estimates and Statistics Selection; MPM and Outcome Classification) — pages excerpted from the supplied PDF content.*

### Appendix Table A2 shows a description of the relevant outcome variables included in each of

### wpiea2020067-print-pdf - Appendix Table A2 shows a description of the relevant outcome variables included in each of

### D. MPM’s Measurement
- Three measurement approaches in the literature:
  - Cross-country policy-setting dummies: MPMs defined as dummies taking value one during years an instrument was used/in place and zero otherwise (examples: Cerutti, Claessens, and Laeven, 2017; Claessens, Ghosh, and Mihet, 2013; Lim and others, 2011).
  - Direction-of-change discrete variable: MPM takes values -1, 0, and 1 for loosening, neutral, and tightening actions respectively (example: Kuttner and Shim, 2016). Bulk of evidence comes from tightening episodes.
  - Intensive-margin measures: measure the intensity/size of policy changes (examples: Jiménez and others, 2017; Richter, Schularick, and Shim, 2019).
- Composite MPM measures:
  - Examples include composites of LTV and DSTI limits, and composites based on broad-based, liquidity, housing and other measures (examples: Kuttner and Shim, 2016; Cerutti, Claessens, and Laeven, 2017; Claessens, Ghosh, and Mihet, 2013; Akinci and Olmstead-Rumsey, 2018; Vandenbussche, Vogel, and Detragiache, 2015).
  - Composite MPM measures categorized into: (i) housing, (ii) non-housing (broad-based, liquidity and other), and (iii) general (includes both housing and non-housing).

### E. Standardization
- Heterogeneity in outcome variables and measures across studies requires standardization for comparability.
- Standardization approach:
  - Normalize selected ˆβ coefficients and their standard errors by the standard deviation of the corresponding outcome variable.
  - When MPMs are discrete (dummy or -1/0/1), effect size interpreted as a standard deviation change in outcome in response to a policy change (tightening or adoption).
  - When MPMs measure intensive margin, effect standardized by outcome standard deviation and, when available, database includes the standard deviation of the MPM change; effect interpreted as a standard deviation change in outcome in response to unit change in the policy variable (often corresponds to one percentage point change).
  - Use regression-specific summary statistics when available; requested from authors if not available; otherwise use summary statistics available in the paper.
- Visual evidence:
  - Figure 2a: range of unstandardized effects is wide and dispersed.
  - Figure 2b: once standardized, distribution is approximately normal with effect sizes concentrated in the 0 to -0.2 standard-deviations region.
- Note on standardized standard errors:
  - Yuan and Chan (2011) show dividing the standard error by the outcome’s standard deviation is a good approximation of the standardized standard error under certain conditions.

### III. Descriptive Statistics

#### A. Database Overview
- Figure 3: matrix of number of papers studying each MPM tool (rows) on each outcome variable (columns); darker shade = more papers.
- Most-studied pairs concentrate in housing sector (e.g., borrower-based housing measures impact on house prices and household credit).
- Study sample composition:
  - 43 percent of studies are cross-country using macro-level data.
  - 10 percent of studies are cross-country using micro-level data.
  - 12 percent of studies are country studies using macro-level data.
  - 34 percent of studies are country studies using micro-level data.
- Common control variables (percent of results):
  - Other MPMs: 61 percent
  - Interest rates: 77 percent
  - Other macroeconomic variables: 61 percent
  - Cross-sectional fixed effects: 80 percent
  - Time fixed effects: 35 percent
- Empirical approaches:
  - Many use panel-data techniques and lagged effects to address endogeneity; some use GMM or instrumental variables; micro-level studies often use event-study or difference-in-difference.
- Time horizons and methods:
  - Most studies focus on effects up to a year.
  - Some use VARs or local projection methods to study various horizons.
- Note: half of the studies present at least one result using OLS; one quarter present at least one result using GMM; 15 percent present at least one result using difference-in-difference; half present at least one result using other methods (e.g., Panel VAR, Quantile Regression).

#### B. Statistical Significance
- Figure 4 summarizes number of statistically significant coefficients (at least at the 10 percent level) versus total number of coefficients for selected outcomes: household credit, house prices, total credit (about half of 6627 coefficients).
- Most studied measures for household credit and house prices: loan-to-value (LTV) limits and debt-service-to-income (DSTI) limits.
- For total credit growth: loan loss provision, reserve requirements, and composite measures are widely studied with a large fraction statistically significant.
- Overall picture: heterogeneous and mixed statistical significance across tools and outcomes.

#### C. Magnitudes
- Table 1: summary statistics of standardized effects for short-term horizon (up to one year); summarizes ~50 percent of all coefficients.
- Table 1 (selected entries, all numbers preserved exactly):
  - Panel: MPM in place (0,1):
    - Balance sheet fragility: Obs. 249; Mean -0.02; Std. Dev. 0.09; Min -0.94; Max 0.21
    - Capital inflow: Obs. 780; Mean 0.05; Std. Dev. 0.19; Min -0.83; Max 0.77
    - Corporate credit: Obs. 51; Mean -0.01; Std. Dev. 0.54; Min -1.69; Max 2.24
    - Credit: Obs. 190; Mean -0.38; Std. Dev. 0.71; Min -3.52; Max 1.90
    - Economic activity: Obs. 300; Mean 0.13; Std. Dev. 0.55; Min -1.67; Max 0.95
    - House price: Obs. 157; Mean -0.09; Std. Dev. 0.39; Min -2.06; Max 1.19
    - Household credit: Obs. 176; Mean -0.15; Std. Dev. 0.29; Min -1.23; Max 0.93
    - All Outcomes: Obs. 931; Mean -0.12; Std. Dev. 0.44; Min -3.52; Max 2.24
  - Panel: MPM change (-1,0,1):
    - Balance sheet fragility: Obs. 5; Mean -0.02; Std. Dev. 0.03; Min -0.05; Max 0.01
    - Bank default risk: Obs. 38; Mean -0.31; Std. Dev. 0.62; Min -3.35; Max 0.40
    - Capital inflow: Obs. 116; Mean 0.00; Std. Dev. 0.10; Min -0.37; Max 0.34
    - Corporate credit: Obs. 370; Mean 0.08; Std. Dev. 0.07; Min -0.11; Max 0.34
    - Credit: Obs. 168; Mean -0.07; Std. Dev. 0.45; Min -2.21; Max 2.40
    - Economic activity: Obs. 392; Mean -0.08; Std. Dev. 0.51; Min -5.36; Max 1.57
    - House price: Obs. 353; Mean -0.01; Std. Dev. 0.31; Min -1.44; Max 2.69
    - Household credit: Obs. 411; Mean -0.14; Std. Dev. 0.32; Min -1.59; Max 1.02
    - Non bank credit: Obs. 180; Mean 0.02; Std. Dev. 0.04; Min -0.01; Max 0.14
    - All Outcomes: Obs. 1538; Mean -0.07; Std. Dev. 0.38; Min -5.36; Max 2.68
  - Panel: MPM intensity (level or change in levels):
    - Balance sheet fragility: Obs. 14; Mean -0.25; Std. Dev. 0.23; Min -0.66; Max -0.01
    - Credit: Obs. 55; Mean -1.21; Std. Dev. 1.14; Min -5.91; Max -0.06
    - Economic activity: Obs. 1500; Mean 0.00; Std. Dev. 0.01; Min -0.04; Max 0.05
    - House price: Obs. 238; Mean -0.28; Std. Dev. 0.20; Min -0.94; Max 0.40
    - Household credit: Obs. 1680; Mean 0.01; Std. Dev. 0.04; Min -0.22; Max 0.20
    - All Outcomes: Obs. 625; Mean -0.21; Std. Dev. 0.50; Min -5.91; Max 0.40
- Summary interpretation:
  - First panel effects represent standard deviation change in outcome from having the MPM in place (comprises 30 percent of short-term results).
  - Second panel effects represent standard deviation change in outcome from changing MPM in the direction of tightening (comprises 50 percent of short-term results).
  - Third panel effects represent standard deviation change in outcome from a unit change in intensity of MPM use.
  - Effects encompass both positive and negative values; results concentrated around credit and house prices; high variability in several outcome variables.

#### D. Symmetry
- Asymmetry rationale:
  - Tightening measures often used in expansionary phases when incentives to leverage are stronger, so measures may bind and show larger effects.
  - Loosening measures used during correction phases to release buffers; agents may not take advantage due to reduced income or increased uncertainty.
  - Leakages (incentives to circumvent regulation) can influence symmetry.
- Sample evidence on direction of policy:
  - About sixty percent of studies discuss effects of either tightening or loosening.
  - More than a quarter study only tightening actions.
  - Four studies study easing actions alone.
  - About 20 percent study both tightening and loosening.
  - More evidence exists on tightening than loosening, largely due to developments in the past decade.
- Statistical significance comparisons:
  - Among papers looking at tightening and/or loosening, 46 percent of results find tightening coefficients statistically significant at the 10 percent level, versus 36 percent for easing actions.
  - Within 13 papers analyzing both tightening and loosening in comparable regressions, tightening coefficients are significant in 61 percent of cases, easing coefficients in 37 percent.
- Magnitude comparisons:
  - Among the 13 papers studying both directions, Figure 6 suggests more mass above the 45-degree line: tightening coefficients generally larger in absolute value.
  - About 57 percent of cases: estimated tightening coefficients larger in absolute value than easing coefficients.
  - Restricting to tightening coefficients which are statistically significant and negative (about 40 percent), they overwhelmingly exceed effects of easing actions.
  - In about 20 percent of cases, effects of easing are stronger than tightening or too close to call.
- Conclusion on asymmetry:
  - Effect of tightening has larger quantitative impact and is more often statistically significant than easing.
  - Partly due to larger number of observations for tightening actions (greater statistical power).

### IV. Meta-Regressions

#### A. Methodology
- Meta-analysis framework to synthesize estimated effects and examine heterogeneity.
- Model setup:
  - Assume standardized true average effect of macroprudential policy on credit denoted by β.
  - Database includes estimated effect ˆβ_jk ∼ N(β, σ^2_jk), where σ^2_jk is variance (square of standard error).
  - Precision-weighted average of ˆβ_jk recovers β; weights inversely proportional to standard error.
- Meta-regression specification:
  - ˆβ_jk = ∑_m θ_m · MPM_mjk + γ X_jk + δ · SE^2(β_jt) + ε_jk
  - Where MPM_mjk are dummies for MPM measures; θ_m denote average effect of MPM m controlling for X_jk attributes.
  - X_jk includes controls such as dummy for peer-reviewed publication and dummy identifying most complete specification in a study.
  - SE^2(β_jt) included as publication-bias selection correction (Stanley and Doucouliagos (2014) approach).
- Publication bias:
  - Addressed by including SE^2(β_jt) as control; Appendix shows bias significant among published papers but smaller and statistically insignificant in micro-based studies.
- Sample and estimation details:
  - Sample: short-term effects (up to one year) and excludes results that could not be standardized and outliers exceeding -3 or +3 standard deviations (for MPM existence or changes).
  - Error term ε_jk ∼ N(0, σ^2_jk) is heteroskedastic; estimation via weighted least squares with inverse of standard errors as weights.

#### B. Baseline Results
- Table 2: baseline meta-regression results for short-term effects of tightening MPP changes on credit (including household credit).
- Sample: results where MPM change measured via -1,0,1 dummies (tightening episodes); excludes loosening actions and MPM indices in baseline.
- Key baseline findings (Table 2, Column (1) simple specification):
  - Broad based: -0.045 (standard error 0.002) — significant at 1% (notation: -0.045 ∗∗∗ (0.002))
  - Housing: -0.041 (0.007) ∗∗∗
  - Liquidity & Other: -0.118 (0.005) ∗∗∗
  - Interpretation: tightening liquidity & other tools associated with a 0.12 standard deviation reduction in total credit; tightening housing and broad-based tools associated with 0.04 standard deviation reduction in total credit.
- Robustness checks and adjustments (Columns (2)-(5)):
  - Weighting to avoid study over-representation: sampling weights proportional to inverse of number of estimates per study; results largely unchanged.
  - Publication-bias correction (SE sq.) added; selection term significant in some specifications but overall pattern stable.
  - Controls added: dummy for non-published papers; dummy for incomplete specification; results robust though broad-based loses precision in some specs.
  - Journal-quality weighting introduced: results robust and stable; broad-based measure regains precision.
  - Results robust to excluding observations where standardized effects calculated using imputed summary statistics and to sequentially excluding specific studies.
- Results from studies measuring MPM in place (0,1) (Appendix Table A4 referenced):
  - Effects of having MPM in place: 0.2, 0.1 and 0.09 standard deviation reduction in credit from broad-based, housing sector, and liquidity/other tools respectively (explanatory notes: may capture cumulative effects or cross-country differences).

#### C. Heterogeneity of Average Effects
- Table 3: average effects by specific instruments (standardized effects):
  - LTV:
    - Credit: -0.065 ∗∗ (0.031)
    - Household Credit: -0.061 ∗∗ (0.027)
    - House Price: -0.007 (0.015)
  - DSTI:
    - Credit: -0.063 ∗∗ (0.025)
    - Household Credit: -0.052 ∗∗ (0.018)
    - House Price: 0.001 (0.011)
  - Capital Requirements:
    - Credit: -0.065 ∗∗ (0.030)
    - Household Credit: -0.030 (0.022)
    - House Price: -0.042 ∗∗ (0.018)
  - Loan Loss Provisions:
    - Credit: -0.030 (0.030)
    - Household Credit: -0.033 (0.021)
    - House Price: 0.016 (0.010)
  - Liquidity: Credit: -0.139 ∗∗∗ (0.030)
  - Other: Credit -0.035 (0.029); Household Credit -0.034 (0.020); House Price -0.002 (0.010)
  - SE sq. (Pub. Bias Correction) included; coefficients reported with standard errors.
- By country setting and unit of analysis (Table 4 and discussion):
  - Table 4 reports average effects on credit for EM (emerging markets), Mixed (advanced, emerging, low-income or exclusively advanced), Micro, and Macro samples.
  - Selected entries (Table 4):
    - EM (Column (1)):
      - Broad based: -0.044 ∗∗ (0.018)
      - Housing: -0.153 ∗∗∗ (0.042)
      - Liquidity & Other: -0.126 ∗∗∗ (0.011)
      - Observations: 198; r2 0.660
    - Mixed (Column (2)):
      - Broad based: -0.033 ∗∗ (0.011)
      - Housing: -0.034 ∗∗∗ (0.007)
      - Liquidity & Other: -0.030 ∗∗∗ (0.008)
      - Observations: 239; r2 0.388
    - Micro (Column (3)):
      - Broad based: -0.045 ∗∗ (0.018)
      - Housing: -0.192 ∗∗∗ (0.009)
      - Liquidity & Other: -0.130 ∗∗∗ (0.007)
      - Observations: 176; r2 0.694
    - Macro (Column (4)):
      - Broad based: -0.032 ∗ (0.015)
      - Housing: -0.039 ∗∗∗ (0.009)
      - Liquidity & Other: -0.030 ∗∗∗ (0.009)
      - Observations: 261; r2 0.229
  - Findings:
    - Evidence of relatively larger average effects in EMs for housing and liquidity (and other) measures, though EM results have wider confidence intervals.
    - Micro-level studies show significantly larger effects of tightening MPMs than macro-level studies.
      - Example: tightening housing tools reduce credit for banks or firms on average by 0.2 standard deviations versus 0.04 reduction in aggregate credit.
    - Explanations for micro vs macro differences:
      - Higher statistical and identification power in micro studies.
      - Micro studies focus on entities around binding constraints or constrained agents.
      - Leakages/spillovers may reduce transmission from micro effects to aggregate.

*Italicized source: IMF Working Paper wpiea2020067 (excerpted content provided)*

### Appendix Table A8 shows that relatively larger effects found for EM (relative to the mixed samples) and

### Appendix results on macroprudential policy effects (excerpts)

### Mapping standardized to raw effects (Table 5)
- Credit Growth mappings reported:
  - y-o-y quarterly (Macro): Average S.D. of Outcome = 13% ; Std. Effect Size = -0.05 ; Effect Size = - 0.6 p.p.
  - q-o-q quarterly (Micro): Average S.D. of Outcome = 30% ; Std. Effect Size = -0.15 ; Effect Size = - 4.5 p.p.
- Interpretation guidance:
  - The average macro-level standardized effect of -0.05 can correspond to a -0.6 percentage point reduction in year-on-year credit growth using a typical cross-country study S.D. of 13 percent.
  - The average micro-level standardized effect of -0.15 can correspond to a -4.5 percentage point reduction in quarter-on-quarter credit growth for banks or firms using a typical micro study S.D. of 30 percent.
- General note: All standardized effects in the paper can be converted to unstandardized magnitudes by multiplying by the average standard deviation of the outcome variable.

### Reconciling macro and micro estimates: statistical power, leakages, and spillovers
- Statistical power and precision:
  - Conventional calibration used: 5% level of statistical significance and 80% power.
  - For adequate power, a result’s standard error needs to be smaller than the absolute value of the ‘true’ effect divided by 2.8.
  - Based on the meta-analysis average effects as a lower-bound for the ‘true effect’:
    - Almost 80% of the macro-level studies are under-powered to detect standardized effects below 0.05 standard deviations.
    - Only 20% of the micro-level studies are under-powered.
    - Precision of effects in micro-level studies reported in regressions is almost five times larger than that of macro studies.
    - Consequently, precision-weighted average effects are lower for macro studies than micro studies.
- Leakages and spillovers (Table 6 average standardized effects on cross-border and non-bank lending):
  - Interpretation: A positive sign means that tighter macroprudential policy increases cross-border and/or nonbank lending.
  - Reported coefficients (standard errors in parentheses; significance: * 10%; ** 5%; *** 1%):
    - Broad based: 0.054** (0.015) ; 0.049** (0.017) ; 0.066** (0.024)
    - Housing: 0.004* (0.002) ; 0.005*** (0.001) ; 0.005*** (0.000)
    - Liquidity & Other: 0.060*** (0.000) ; 0.059*** (0.001) ; 0.077** (0.024)
    - SE sq. (Pub. Bias Correction): -0.190 (0.847)
    - Specification with incomplete controls: 0.030* (0.014)
    - Non-Published Papers: -0.034 (0.025)
    - Observations: 59, 59, 59 ; r2: 0.407, 0.365, 0.445
  - Findings:
    - All measures are associated with leakages and/or spillover effects, strongest for broad-based and liquidity measures.
    - Domestic constraints on lending can create spillovers internationally by shifting lending portfolios to other countries.
  - Back-of-the-envelope netting:
    - Subtracting the average effect of leakages/spillovers from the average micro-level effects suggests a net negative effect of tightening macroprudential policy on credit.
    - For liquidity, this net effect is approximately -0.05 standard deviations, roughly equivalent to the average macro-effects of -0.03 standard deviations (reported in Table 4).
    - This crude calculation aligns with studies suggesting leakages reduce but do not fully eliminate the impact of macroprudential policy.

### Impact on economic activity (Table 7 and illustrative studies)
- Meta-regression average effects (cross-sector index):
  - Dependent variables — standardized effects:
    - Credit: Cross-Sector Index = -0.011*** (0.003)
    - House Price: Cross-Sector Index = 0.002 (0.003)
    - Economic Activity: Cross-Sector Index = -0.004*** (0.000)
  - Additional reported items:
    - SE sq. (Pub. Bias Correction): -0.0021 (0.070) ; 0.932 (19.355) ; -0.130 (0.099)
    - Specification with incomplete controls: 0.001 (0.003) ; -0.002 (0.004) ; 0.017* (0.006)
    - Observations: 80 (Credit), 55 (House Price), 143 (Economic Activity)
    - r2: 0.643, 0.216, 0.095
- Interpretation:
  - Macroprudential policy tightening has a negative and statistically significant effect on economic activity indicators at the 1 percent level.
  - The coefficient of -0.004 standard deviations on economic activity is about a third of the effect on credit reported in the same table.
- Selected case-study estimates (examples from database):
  - Richter, Schularick, and Shim (2019): Following a 10 percentage points reduction in LTV limits, real GDP declines by 0.5 percent after 2 years and 1.1 percent after 4 years.
  - Alam and others (2019): A tightening of up to 10 percentage points in LTV limits is associated with a 1.5 percent decline in real consumption growth; a tightening between 10-25 percentage points is associated with a 1.1 percent decline.
  - Kim and Mehrotra (2018, 2019): Comparable effects on real GDP, private consumption, and investment between a one-standard-deviation shock to an index of macroprudential measures and a one-standard-deviation monetary policy tightening.
- Overall implication:
  - By reducing credit flows, macroprudential policy may also reduce economic activity in the near term; this may be consistent with reducing boom-bust cycles and lowering output volatility in the long term.

### Dynamics of effects
- Timing and persistence:
  - Full effects can take more than one year to materialize; some effects materialize with delay, while circumventing behavior may weaken effects over time.
  - Appendix Table A10 finds smaller lagged than contemporaneous effects across tool buckets, suggesting a hump-shaped response of credit to macroprudential adjustments.
  - Studies using VAR/VECM or local projections (significant at 10% level) indicate:
    - Peak effects often at 1-2 years for credit, and 2-3 years for macroeconomic variables (GDP, consumption, prices).
    - First-year effect is typically between 20 percent and 80 percent of the peak effect.
  - Dynamic panel analyses corroborate persistent but declining effects over time; about 30 percent of studies in the database use dynamic panel methods.
- Practical implication:
  - A non-trivial fraction of the total effects of macroprudential policy occurs beyond the first year and should be considered when evaluating policy efficacy.

### Conclusions and suggested future research directions
- Main findings summarized:
  - On average, macroprudential policy tools have statistically significant effects on credit.
  - Effects are relatively larger in micro-level studies than macro-level studies, partly due to stronger identification and statistical power in micro studies, presence of leakages and spillovers, and macro studies averaging across constrained and unconstrained agents.
  - Significant heterogeneity across country settings and instruments; effects appear stronger among emerging markets but with larger confidence intervals.
  - Suggestive evidence of stronger effects from tightening than loosening actions.
  - Evidence of negative near-term impacts on economic activity from tightening macroprudential policy.
- Recommendations for future work:
  - Improve measurement of macroprudential policy actions to better capture the intensive margin.
  - Increase research using micro-level data, which appears more adequately powered.
  - Expand knowledge on the effects of easing macroprudential policy, leveraging actions in response to COVID-19 to understand effects in downturns.
  - Further explore nonlinearities, including interactions with other policies.

*Source: Excerpts from the provided IMF working paper appendix (wpiea2020067-print-pdf).*

### REFERENCES

### wpiea2020067-print-pdf - REFERENCES

### Taxonomy and Definitions (Table A1 & Table A2)
- Macroprudential tool groups and exact labels:
  - broad_based: Countercyclical buffers (CCB)†; Conservation buffers (CON); Capital requirements (CR)*†§; Leverage limits (LVR); Loan loss provisions (LLP)*†; Limits on credit growth (LCG); Loan restrictions (LR)*†; Limits on foreign currency loans (LFC)
  - liquidity: Liquidity (LIQ)*; Limits on loan-to-deposit ratio (LTD); Limits on foreign exchange positions (LFX); Reserve Requirements (RR)*
  - housing: Limits on loan-to-value ratio (LTV)*; Limits on the debt-service-to-income ratio (DSTI)*
  - other: Systemically important financial institutions (SIFI); Tax measures (TAX)*†; Other (OTHER)*†
- Index measures:
  - i_housing: Index measure constructed with housing tools only.
  - i_non_housing: Index measure constructed without housing tools (broad_based, liquidity, other).
  - i_general: Index measure constructed with both housing and non-housing tools.
- Notes on classification:
  - * some papers also study FX targeted measures of this macroprudential tool (suffix: _FX)
  - † some papers also study household or housing sector targeted measures of this macroprudential tool (suffix: _HH)
  - § some papers also study corporate sector targeted measures of this macroprudential tool (suffix: _CORP)
- Outcome variable categories and examples (Table A2):
  - Balance sheet fragility: bank total asset; bank leverage ratio; bank liquidity ratio; bank non-core to core liability ratio; (-) bank capital
  - Bank default risk: non-performing loans; probability of defaults; expected default frequency
  - Capital flows*: cross border credit to banks or non-banks; ratio of cross border credit to total credit
  - Corporate credit*: international and domestic debt by corporate; credit to corporate
  - Credit: bank credit (committed or drawn); total credit; credit to GDP ratio; debt-to-liability ratio; bank borrowing
  - House price: house price index
  - Household credit: household credit; mortgage loans; consumption loans; auto loans
  - Economic activity: GDP; consumption; investment; inflation; firm sales
  - Non bank credit*: non bank credit; non bank credit as a share of total assets or total liabilities
  - Other: number of housing transactions; house price expectation; mortgage rate; interest rate spread; stock price
  - * some of the outcome variables in these categories are identified as cross-boarder spillovers or domestic leakages.

### Testing for Selection Bias (Figure 9 discussion and Table A3)
- Funnel-plot evidence:
  - More precise results are clustered at the top and show smaller effects.
  - Estimated values tend to be larger on average for larger values of the standard error (less precision).
- Regression evidence (Table A3: Selection Bias Test for All Effects; tested up to one year horizon and excluding loosening episodes):
  - Column (1): Standard Error coefficient = -0.737 ∗∗∗ (standard error (0.099)); Observations = 260; Studies = 30.
  - Column (2): Standard Error coefficient = -1.261 ∗∗ (standard error (0.577)); Observations = 226; Studies = 30; includes study fixed effects.
  - Column (3): Standard Error coefficient = -1.071 ∗∗∗ (standard error (0.267)); Observations = 212; Studies = 16.
  - Column (4): Standard Error coefficient = -0.997 ∗∗∗ (standard error (0.042)); Observations = 211; Studies = 14.
  - Column (5): Standard Error coefficient = -0.170 (standard error (2.518)); Observations = 492; Studies = 2; finds bias considerably small and statistically insignificant in micro-based studies but present in macro studies.
  - Significance notation: * p<0.10, ** p<0.05, *** p<0.01.
- Interpretation:
  - Statistically significant and negative association between estimated effects and their standard error in multiple specifications indicates selective reporting (publication/selection bias) in the collected estimates, particularly in macro studies.

### Additional Results on Credit, Household Credit, and House Prices (Tables A4 & A5, A11)
- Existence of macroprudential policy (Table A4: Average Effects of Macroprudential Tools in Place on Credit):
  - Broad based: -0.268 ∗ (0.125); alternative columns report -0.326 ∗∗ (0.108), -0.326 ∗∗ (0.109), -0.214 ∗ (0.110)
  - Housing: -0.106 ∗∗ (0.041); alternative columns report -0.102 ∗∗ (0.034), -0.102 ∗∗ (0.034), -0.099 (0.052)
  - Liquidity & Other: -0.109 ∗∗∗ (0.019); alternative columns report -0.119 ∗∗∗ (0.027), -0.119 ∗∗∗ (0.027), -0.091 ∗∗∗ (0.020)
  - SE sq. (Pub. Bias Correction): -0.012 (0.064) and -0.007 (0.066) in two specifications.
  - Non-Published Papers: -0.147 ∗ (0.062)
  - Observations = 207 in reported specifications; r2 reported up to 0.282 across columns.
  - Note: These existence-based average effects (0.2, 0.1 and 0.09 standard deviation reductions noted in text) are relatively larger than tightening-event estimates and may capture cumulative or cross-country differences.
- Effects on Household Credit and House Prices (Table A5):
  - Household Credit:
    - Broad based: -0.008 (0.028)
    - Housing: -0.038 ∗ (0.021)
    - Liquidity & Other: -0.012 (0.024)
  - House Price:
    - Broad based: -0.009 (0.010)
    - Housing: 0.002 (0.016)
    - Liquidity & Other: -0.008 (0.033)
  - SE sq. (Pub. Bias Correction): -2.090 (1.942) for Household Credit; -2.012 (1.509) for House Price.
  - Observations: Household Credit = 299; House Price = 250; r2 = 0.225 and 0.180 respectively.
  - Text interpretation: Imprecise effects overall; significant negative effects for housing macroprudential measures on household credit. Disaggregation shows broad-based measures such as capital requirements reduce house prices on average.

- Summary distribution of standardized macroprudential policy effects by instrument and outcome (Table A11 highlights selected entries; examples shown exactly):
  - MPP in place (0,1) — credit (all outcomes):
    - Housing: Obs = 11; Mean = -0.36; Min = -1.84; Max = 1.40
    - Broad Based: Obs = 25; Mean = -0.66; Min = -3.52; Max = 0.52
    - Liquidity: Obs = 4; Mean = -1.29; Min = -3.27; Max = -0.02
    - Index: Obs = 134; Mean = -0.28; Min = -1.57; Max = 1.04
  - MPP change (-1,0,1) — household credit:
    - Housing: Obs = 193; Mean = -0.18; Min = -1.58; Max = 1.02
    - Broad Based: Obs = 63; Mean = -0.08; Min = -1.59; Max = 0.86
    - Liquidity: Obs = 6; Mean = -0.04; Min = -0.64; Max = 0.43
    - Index: Obs = 135; Mean = -0.11; Min = -1.19; Max = 0.64
  - MPP quantitative (level or change in levels) — all outcomes:
    - Housing: Obs = 305; Mean = 0.00; Min = -0.37; Max = 0.20
    - Broad Based: Obs = 147; Mean = -0.25; Min = -0.92; Max = 0.40
    - Liquidity: Obs = 149; Mean = -0.62; Min = -5.91; Max = 0.26

### Robustness and Heterogeneity (Tables A6–A10)
- Robustness to excluding single papers and imputed summary statistics (Table A6: MPM Tightening Effects on Credit; selected entries exact):
  - Excluding Imputed Summary Stats.: Broad-based = -0.028 (0.033); Housing = -0.038 ∗∗ (0.017); Liquidity = -0.116 ∗∗∗ (0.028)
  - Excluding Paper 1 through Paper 22: sample of point estimates for Broad-based, Housing, Liquidity reported across exclusions; many entries show Housing and Liquidity effects remain statistically significant (examples: Excluding Paper 6: Broad-based = -0.066 (0.059), Housing = -0.070 (0.056), Liquidity = -0.147 ∗∗∗ (0.057)).
- Micro vs Macro (Table A7: Average Effects of Tightening Macroprudential Policy on Credit):
  - Micro sample:
    - Broad based: -0.024 (0.028)
    - Housing: -0.188 ∗∗∗ (0.010)
    - Liquidity & Other: -0.110 ∗∗∗ (0.019)
    - SE sq. (Pub. Bias Correction): -2.383 ∗∗∗ (0.315)
    - Observations = 176; r2 = 0.694
  - Macro sample:
    - Broad based: -0.025 (0.024)
    - Housing: -0.035 ∗ (0.019)
    - Liquidity & Other: -0.023 (0.021)
    - SE sq. (Pub. Bias Correction): 1.000 (1.709)
    - Observations = 267; r2 = 0.178
  - Interpretation: Larger estimated effects in micro studies relative to macro studies, especially for Housing and Liquidity.
- EM vs Mixed and Micro vs Macro with joint controls (Table A8; selected exact coefficients):
  - Column (1) EM sample:
    - Broad based = 0.038 ∗ (0.019)
    - Housing = -0.105 ∗∗∗ (0.012)
    - Liquidity & Other = -0.046 ∗∗ (0.017)
    - Observations = 204; r2 = 0.660
  - Column (2) Mixed sample:
    - Broad based = -0.039 ∗∗∗ (0.009)
    - Housing = -0.034 ∗∗∗ (0.007)
    - Liquidity & Other = -0.031 ∗∗∗ (0.009)
    - SE sq. (Pub. Bias Correction) = -3.282 ∗∗∗ (1.055)
    - Observations = 239; r2 = 0.397
  - Columns (3)-(4) show micro/macro splits with consistent patterns: Housing and Liquidity effects more negative in Micro and EM samples.
- Selection-correction robustness (Table A9: Andrews and Kasy (2019) correction; exact coefficients):
  - Broad based:
    - EM = -0.037 ∗∗∗ (0.005)
    - Mixed = -0.015 ∗∗∗ (0.004)
    - Micro = -0.028 ∗∗∗ (0.008)
    - Macro = -0.034 ∗∗∗ (0.003)
  - Housing:
    - EM = -0.091 ∗∗∗ (0.006)
    - Mixed = -0.037 ∗∗∗ (0.003)
    - Micro = -0.251 ∗∗∗ (0.043)
    - Macro = -0.040 ∗∗∗ (0.003)
  - Liquidity & Other:
    - EM = -0.105 ∗∗∗ (0.009)
    - Mixed = -0.011 (0.011)
    - Micro = -0.121 ∗∗∗ (0.009)
    - Macro = -0.014 ∗∗∗ (0.004)
  - Observations: 204 (EM), 239 (Mixed), 176 (Micro), 267 (Macro).
  - Interpretation: Selection-correction method yields broadly similar qualitative patterns; many effects remain statistically significant.
- Contemporaneous vs Lagged effects (Table A10):
  - Contemporaneous:
    - Broad based = -0.054 ∗∗∗ (0.008)
    - Housing = -0.066 ∗∗ (0.029)
    - Liquidity & Other = -0.131 ∗∗∗ (0.006)
    - Observations = 245; r2 = 0.671
  - Lagged:
    - Broad based = 0.004 (0.005)
    - Housing = -0.035 ∗∗∗ (0.003)
    - Liquidity & Other = -0.020 ∗∗∗ (0.003)
    - Observations = 198; r2 = 0.210
  - Specification with incomplete controls shows different signs across contemporaneous and lagged specifications.

### Database Coverage (Tables A12 and Papers in the Database)
- The references list and Tables A12+ enumerate the papers included in the database used for the meta-analysis; representative entries include:
  - Afanasieff and others (2015): Implementing loan-to-value ratios: The case of auto loans in Brazil (2010-11).
  - Akinci and Olmstead-Rumsey (2018): How effective are macroprudential policies? An empirical investigation.
  - Alam and others (2019): Digging Deeper – Evidence on the Effects of Macro-prudential Policies from a New Database, IMF Working Paper 19/66.
  - Cerutti, Claessens, and Laeven (2017): The use and effectiveness of macroprudential policies: New evidence.
  - Kang and others (2017): Macroprudential policy spillovers: A quantitative analysis, IMF Working Paper 17/170.
  - Jiménez and others (2017): Macroprudential policy, countercyclical bank capital buffers, and credit supply: evidence from the Spanish dynamic provisioning experiments.
  - Richter, Schularick, and Shim (2019): The costs of macroprudential policy.
- Note: For additional details on each paper, the source indicates accompanying data files list full citations.

*Italic source: wpiea2020067-print-pdf - REFERENCES (canonical PDF provided).*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2020/english/wpiea2020067-print-pdf.pdf_