## wpiea2021113-print-pdf — Sections 3–5 summary

## Source details

**Canonical URL:** [wpiea2021113-print-pdf — Sections 3–5 summary](https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021113-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2021/english/wpiea2021113-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2021/english/wpiea2021113-print-pdf.pdf.json)

---

### Section 3 — Data: overview and constructions
- Purpose: Describe data sources and the construction of the corporate loan growth indicator, macroprudential policy indexes, and corporate riskiness indicators used to identify effects of household-targeted macroprudential policy actions on banks’ corporate lending volumes.
- Frequency and sample window:
  - annual frequency from 1998 to 2018 for bank-level loan volumes;
  - quarterly frequency for corporate riskiness measures and macroprudential policy actions (mid-1990s until 2018Q4).

### Measuring corporate loan growth (CLG)
- Data source: bank-level volumes of corporate and residential mortgage loans from Fitch Connect.
- Sample coverage and constraints:
  - Time: 1998 to 2018 (annual).
  - Country sample: 13 countries — Brazil, Canada, Chile, China, Finland, Germany, Malaysia, Mexico, Poland, Russia, Spain, Sweden and Thailand.
  - Imposed conditions:
    - Minimum of 10 bank observations for each country-year.
    - Each bank must have at least 6 years of data.
    - Net loans are defined as total loans net provisions.
- Definition of variable:
  - relative corporate loan growth (CLG) = the difference in the annual growth rates in the ratio of corporate loans to total (net) bank loans among the top 30 and bottom 30 percent of banks, when ranked according to the size of their residential mortgage portfolios (relative to total net loans).
- Loan category definitions:
  - Corporate loans: loans and leases to corporate and commercial enterprises; excludes non-residential and commercial property mortgage loans.
  - Residential mortgage loans: loans secured by residential property; excludes unsecured loans or loans secured by assets other than residential property.

### Measuring corporate credit riskiness
- Conceptual basis: Follows Greenwood and Hanson (2013); measures combine corporate financial vulnerability and risk sentiment features.
- Firm-level data source: Datastream.
- Sample construction:
  - Quarterly measures for 29 economies (13 advanced + 16 emerging).
  - Minimum of 30 firm observations per quarter per country.
  - Series seasonally-adjusted; unbalanced panel (majority back to 2002); for 8 countries biannual series constructed.
- Three vulnerability indicators and corresponding riskiness measures:
  - TDTA (leverage-based): difference in ratio of total debt to total assets of top 20 percent debt issuers over current quarter and bottom 20 percent among listed firms.
  - TDtE (debt overhang): as above using total debt to EBITDA.
  - ICR (interest coverage ratio measure): as above using interest coverage ratio, defined as the ratio of interest expenses to EBITDA.
- Construction steps:
  - Each firm assigned a value 1–10 by decile of the indicator distribution in a given quarter and country.
  - Firms sorted by change in net debt over past year relative to past total assets; debt includes loans and bond financing.
  - Riskiness measure = difference in average vulnerability indicator for top debt-takers (values 9 or 10) and bottom debt-takers (values 1 or 2).
  - Interpretation: measures increase (decline) when average vulnerability indicators increase more (increase less or decline) among top debt-takers relative to bottom debt-takers.
- Correlations and examples:
  - Cross-country average correlations reported; correlation highest for ICR and TDtE. Examples: Denmark correlation of 0.12 between leverage-based and TDtE/ICR (second-lowest after Sweden); Mexico correlation of 0.81.
- Rationale and prior evidence:
  - IMF (2018) find these measures signal heightened downside risks to GDP growth and a higher probability of banking crises and banking sector stress beyond credit growth signals.

### Lending standards (robustness)
- Alternative measure: bank lending standards for corporate loans from bank loan officer surveys, disaggregated by SMEs and large corporations.
- Data sources and coverage:
  - ECB and national authorities via Haver Analytics.
  - Available for 15 countries: Austria, Canada, Germany, Italy, Japan, Korea, Netherlands, Norway, Philippines, Poland, Portugal, Russia, Spain, Thailand and Turkey.
  - Time coverage generally back to 2002–2004; shortest series: Thailand (2007Q4), Norway (2008Q1), Philippines (2009Q1), Russia (2009Q2).
- Note: reduced sample when using lending standards; three riskiness measures (TDTA, TDtE, ICR) treated as baseline.

### Macroprudential policies and indicators (MaPP)
- Primary source: IMF iMaPP database; quarterly record of MaPP actions by targeted sector (mid-1990s until 2018Q4).
- Coding: +1 tightening, −1 easing, 0 no change.
- Two constructed MaPP indicators:
  - MaPP_HH (household-targeting): baseline = sum of changes in LTV, LTI, DTI and DSTI limits. Robustness alternative MaPP_HH(Alternate) adds restrictions on household loan characteristics.
  - MaPP_BC (broad-based plus corporate-specific): includes changes in bank capital requirements (general and corporate-specific), countercyclical capital buffer, limits on credit growth (overall and corporate-specific) and restrictions on corporate loan characteristics. Robustness alternative MaPP_BC(Alternate) defined similarly.
- Policy action counts (2002Q1–2018Q4 in 29-economy riskiness sample):
  - 120 changes in baseline household-specific MaPPs.
  - 66 changes in baseline broad-based/corporate-specific MaPPs.
  - Little overlap between household-specific and broad-based/corporate-specific interventions.
- Regional use and directionality:
  - MaPPs used relatively more in emerging markets (EMs) than advanced economies (AEs).
  - For both MaPP types, actions in EMs are almost double those in AEs.
  - Total tightened vs eased during sample: 154 tightenings and 34 easings.
  - Household-specific MaPPs used more frequently than broad-based/corporate-specific measures.
  - 2018: advanced economies tightened LTV and DSTI limits.

### Macroprudential shocks (addressing endogeneity)
- Rationale: Use residuals from policy function estimations as shocks; follow two-stage literature where policy functions estimated first and residuals used as shocks.
- Expectation: endogeneity less of a concern for leakages but correlation between corporate credit riskiness and local household credit cycle necessitates shocks approach.
- Construction:
  - Shocks = residuals from ordered probit regressions of MaPP_HH and MaPP_BC on macrofinancial controls.
  - First-stage regressors: real GDP growth, one-year ahead GDP growth forecast, real credit growth, capital account openness, financial development index, exchange rate change vs U.S. dollar, and real house price growth.
  - Appendix A.2 describes computation and robustness to alternative specifications.
- Coverage:
  - Shocks available for 24 of 29 countries; no shocks for Argentina, Bulgaria, Chile, Peru and Vietnam due to data constraints.
  - Those countries included in robustness regressions using policy actions instead of shocks (Appendix A.5).
- Note: results robust to using policy actions instead of shocks (Appendix A.5).

### Other variables and data sources
- Macroeconomic and financial variables:
  - GDP growth rates and exchange rates: WEO database.
  - Credit to private non-financial sector / GDP, domestic corporate bond issuance, domestic bank credit to corporate sector: BIS.
  - Country-specific Financial Conditions Indices (FCI): IMF.
  - 3-month money market rates: Datastream.
  - Global variables: MSCI VIX, U.S. policy rates from Datastream and Wu and Xia (2016).
  - GDP growth forecasts, financial development index, capital account openness: Consensus forecasts, IMF, and IMF’s AREAER report.
- Appendix A.1 lists all data sources (Table A1).

---

### Section 4.1 — Empirical strategy for corporate loan growth
- Baseline panel regression (annual, bank-level relative corporate loan growth CLG):
  - CLG_s_i,t = α + β1 ̄ε_HH_i,t + β2 ̄ε_BC_i,t + β3 ∆Credit_i,t + β4 X_i,t−1 + δ_i + δ_t + ε_i,t
  - ̄ε_HH_i,t and ̄ε_BC_i,t: MaPP shocks (household-specific and broad-based/corporate-specific) in year t.
  - ∆Credit_i,t: year-on-year growth in credit-to-GDP ratio.
  - X_i,t−1: year-on-year real GDP growth and domestic currency appreciation vs U.S. dollar in previous year.
  - Country and time fixed effects; robust SEs for heteroskedasticity and autocorrelation.
  - MaPP shocks aggregated to annual frequency.
- Nonlinear specification with interaction:
  - CLG_s_i,t = α + β1 ̄ε_HH_i,t + β2 ̄ε_HH_i,t × ∆Credit_i,t + β3 ̄ε_BC_i,t + β4 ̄ε_BC_i,t × ∆Credit_i,t + β5 ∆Credit_i,t + β6 X_i,t−1 + δ_i + δ_t + ε_i,t
- Estimation: Standard OLS used (small N and T); GMM avoided due to bias in this sample.

### Robustness checks considered
- Additional controls:
  - nominal short term market interest rates (NR);
  - real growth of bank credit to private nonfinancial corporations (∆BankCredit);
  - real growth of credit to households (∆HouseholdCredit).

---

### Section 4.2 — Results: Corporate Loan Growth (Table 4 summary)
- Column 1:
  - ̄ε_HH_t coefficient: -0.14 (0.34) — negative, statistically insignificant.
  - ∆Credit_t coefficient: -10.41 (3.43) — significant (** p<0.05).
  - Interpretation: high credit growth periods decrease difference in corporate loan growth between banks most and least exposed to residential mortgages (banks with smaller residential mortgage portfolios experience higher corporate loan growth).
- Columns 2–6 (interaction terms and controls):
  - Interaction ̄ε_HH_t × ∆Credit_t:
    - Column 2: 15.19** (6.51)
    - Column 3: 15.30* (7.96)
    - Column 4: 20.51** (8.61)
    - Column 5: 22.80** (7.94)
    - Column 6: 23.56** (8.29)
  - Standalone ̄ε_HH_t becomes negative and in Columns 4–6 statistically significant:
    - Column 4: -1.22** (0.53)
    - Column 5: -1.33* (0.61)
    - Column 6: -1.47* (0.67)
  - NR_t−1 (3-month nominal domestic market interest rate) when included:
    - Column 3: -0.20** (0.08)
    - Column 4: -0.29*** (0.09)
    - Column 5: -0.27** (0.11)
  - ∆Credit_t remains negative and significant across columns (examples):
    - Column 2: -13.06** (4.96)
    - Column 4: -16.07* (8.52)
    - Column 6: -13.46** (5.48)
- Interpretation:
  - Positive and significant interaction indicates that during high credit growth banks with larger residential mortgage portfolios increase corporate loan growth more after an unexpected tightening of household MaPPs.
  - Quantitative example: when credit to GDP ratio grows at one standard deviation above sample mean, relative CLG increases by 54 percent of historical standard deviation (reported in text).
- Sample and fit:
  - Observations: 132 (Cols 1–2), 123 (Col 3), 111 (Col 4), 102 (Cols 5–6).
  - Countries: 11 to 13 across specs.
  - Adjusted R-squared: 0.07 to 0.23.

---

### Section 5.1 — Empirical strategy for corporate credit riskiness
- Baseline dynamic panel regression (quarterly):
  - CC_s_i,t = α + γ CC_s_i,t−1 + Σ_{k=1}^4 β1,k ε_HH_i,t−k + Σ_{k=1}^4 β2,k ε_BC_i,t−k + β3 X_i,t−1 + δ_i + δ_t + ε_i,t
  - ε_HH_i,t−k and ε_BC_i,t−k: MaPP shocks lagged k quarters.
  - X_i,t−1 controls: quarterly yoy growth in credit-to-GDP ratio; quarterly yoy real GDP growth; domestic currency appreciation vs U.S. dollar in previous quarter.
  - Robust SEs for heteroskedasticity and autocorrelation.
- Interaction / nonlinear specification:
  - CC_s_i,t = α + γ CC_s_i,t−1 + β1 ̄ε_HH_i,t−1 + β2 ̄ε_HH_i,t−1 × ∆Credit_i,t−1 + β3 ̄ε_BC_i,t−1 + β4 ̄ε_BC_i,t−1 × ∆Credit_i,t−1 + β5 X_i,t−1 + δ_i + δ_t + ε_i,t
  - ̄ε_HH_i,t = Σ_{k=0}^3 ε_HH_i,t−k; ̄ε_BC_i,t similar; ∆Credit_i,t = yoy credit-to-GDP growth in quarter t.

---

### Section 5.2 — Baseline results: corporate credit riskiness (Table 5)
- Persistence and controls:
  - Lagged dependent variable (t−1) significant:
    - TDTA t−1: 0.486*** (0.042)
    - ICR t−1: 0.328*** (0.040)
    - TDtE t−1: 0.385*** (0.037)
  - ∆Credit_t−1 positive and highly significant:
    - Example: 0.021*** (0.005) for TDTA (Col 1); 0.025*** (0.008) for ICR (Col 2); 0.032*** (0.007) for TDtE (Col 3).
  - ∆GDP_t−1 positive and significant (e.g., 0.044*** (0.010) for TDTA).
  - ∆^q ER_t−1 effects mixed (small negative or positive significant coefficients in some columns).
- MaPP shocks (lags) — baseline (Cols 1–3):
  - Household-specific MaPP lags on TDTA:
    - ε_HH, Lag 1: 0.004 (0.041)
    - ε_HH, Lag 2: 0.057 (0.044)
    - ε_HH, Lag 3: 0.116** (0.044)
    - ε_HH, Lag 4: 0.004 (0.045)
  - Broad/corporate MaPP lags mixed and not jointly significant.
  - For ICR and TDtE, MaPP lags generally not significant.
- Interaction specification (Cols 4–6):
  - Sum of household shocks ̄ε_HH_t−1 standalone:
    - Col 4 (TDTA): 0.020 (0.024) — insignificant.
    - Col 5 (ICR): -0.029 (0.032) — insignificant.
    - Col 6 (TDtE): -0.067* (0.038) — * p<0.1.
  - Interaction ̄ε_HH_t−1 × ∆Credit_t−1:
    - TDTA (Col 4): 0.012** (0.005)
    - ICR (Col 5): 0.014* (0.007)
    - TDtE (Col 6): 0.022** (0.008)
    - Interpretation: tightening of household-specific MaPPs increases corporate credit riskiness more during higher credit growth.
  - Interaction ̄ε_BC_t−1 × ∆Credit_t−1:
    - TDTA: -0.018* (0.010) — * p<0.1
    - ICR: -0.022 (0.019) — insignificant
    - TDtE: -0.015 (0.014) — insignificant
- Quantitative magnitudes:
  - At sample average credit-to-GDP growth, an unexpected household MaPP tightening over past year increases corporate credit riskiness by ~1 percent of its historical standard deviation on average across measures.
  - At credit-to-GDP growth one standard deviation above mean, impact ~10 percent of historical standard deviation on average (Figure 4 referenced).
  - Long-run average effect at one standard deviation above mean: ~16 percent of SD across three measures (17 percent TDTA, 16 percent ICR, 14 percent TDtE).
- Historical prevalence:
  - Sample period: 2002Q1–2018Q4.
  - 19 percent of country-quarters have credit-to-GDP growth > one standard deviation above mean (29 percent for EMs, 10 percent for AEs).
  - Pre-GFC 2005–2008: average credit-to-GDP across countries 14 percent; three riskiness measures increased ~0.8 standard deviation on average.
  - For credit growth of 14 percent, short-term spillovers from unexpected household MaPP tightening = 23 percent of historical SD across corporate credit riskiness measures.
- Inference:
  - Standalone household MaPP terms often insignificant in linear specs, but interaction with credit growth reveals robust positive effects on corporate credit riskiness.
  - Broad-based/corporate MaPP interactions negative across measures but statistically significant only for TDTA — suggests limited effectiveness in reducing corporate credit riskiness relative to household-specific tools during expansions.

---

### Section 5.3 — Robustness
- Robustness exercises:
  - Starting specification includes interaction between MaPP variables and growth rate of credit-to-GDP ratio.
  - Column 1 (Tables 6–8): control for lagged domestic FCI; FCI lag negative and significant across riskiness measures.
  - Results unchanged when using 3-month money market rates or domestic corporate spreads instead of FCI.
  - Results unchanged when including global variables (VIX and federal funds rate) instead of time fixed effects.
- Bank vs bond credit channel:
  - Riskiness measures do not distinguish bank vs bond credit; many MaPP measures affect bank lending only.
  - Hypothesis: MaPP impact on corporate credit riskiness smaller where corporate sector relies more on bond funding.
  - Extensions include lagged growth of corporate bond issuance and bonds-to-bank credit ratio and their interactions with MaPP variables.
- Key robustness findings:
  - Unexpected tightening of household-specific MaPPs increases corporate credit riskiness across robustness checks, including when controlling for corporate bond issuance growth.
  - Interaction ̄ε_HH × corporate bond issuance negative but not statistically significant.
  - ̄ε_BC interaction terms become significant and negative when corporate bond issuance included.
  - Interactions of ̄ε_BC with corporate bond issuance are significant and positive — tightening corporate/broad-based MaPPs reduces corporate credit riskiness less when recent corporate bond issuance is larger.
  - Bonds-to-bank credit ratio interactions: ̄ε_HH × credit growth positive across measures but significant only for TDtE; ̄ε_HH × bonds-to-bank credit ratio negative but never significant.
- Representative coefficients and diagnostics (selected from Tables 6–8):
  - TDTA lag coefficients range ~0.47–0.52***; ̄ε_HH_{t−1} × ∆Credit_{t−1} coefficients include 0.01* to 0.02***; FCI_{t−1} values include -0.13**, -0.15***; observations vary by column (e.g., 1,108; 924); Countries: 18–21; Adjusted R-squared ~0.37–0.42.
  - TDtE and ICR show similar persistence and interaction patterns with variations in significance and adjusted R-squared.
- Outlier and alternative tests:
  - Excluding Canada, Netherlands, South Korea (Column 5): results robust.
  - Controlling for lagged real house price growth (∆HPI_{t−1}): results remain significant.
  - Excluding five countries with high correlation between house price growth and corporate riskiness (Brazil, Finland, Italy, Poland and Spain): results unchanged.
  - Results hold with local projections (Appendix A.4), MaPP actions instead of shocks (Appendix A.5), Driscoll-Kraay SEs, and pooled mean group estimator.
- Alternative MaPP definitions (Table 9):
  - Replacing baseline MaPP shocks with alternate indicators: interaction ̄ε_HH,Alternate × ∆Credit_{t−1} always positive; sometimes loses significance for TDTA.
  - Baseline ̄ε_HH × ∆Credit_{t−1} remains positive and significant across alternate MaPP_BC indicators.
- Overall robustness conclusion:
  - Core result—that household-specific MaPP tightenings can amplify corporate credit riskiness especially when credit growth is high—persists across extensive checks, alternative definitions, and estimators. Some coefficients lose significance in specific specs/measures but the positive interaction is broadly robust.

---

### Section 5.4 — Lending standards from loan officer surveys
- Data: loan officer survey responses on changes in lending standards for corporate loans by firm size (SMEs and large corporates); positive values = tightened standards.
- Baseline regressions correspond to equation (3); Columns 1,3,5 omit domestic 3-month market rate; Columns 2,4,6 include it.
- Joint significance of four lags of household MaPP shocks (ε_HH) is statistically significant across specs; direction differs by firm size:
  - All corporates and SMEs: past unexpected MaPP tightenings have a positive (tightening) impact on lending standards.
  - Large corporates: net impact of MaPP tightening shocks over past four quarters is negative — suggesting easing of lending conditions for large corporates.
- Key coefficients (Table 10, selected):
  - Lagged dependent variable t−1 ≈ 0.675–0.716*** across columns.
  - ε_HH Lag 1: positive and significant in some columns (e.g., 1.301** (0.576); 1.897* (1.034) for SME specification).
  - ε_HH Lags 2–4: mixed signs; some significant negative coefficients for large corporates (e.g., Lag 3: -1.616* (0.869)).
  - ε_BC coefficients by lag: mixed; some negative and significant for large corporates (e.g., ε_BC Lag 1: -2.051* (1.025)).
  - ∆Credit t−1: positive and sometimes significant (e.g., 0.177* (0.095); 0.251** (0.108) for SMEs).
  - Observations: 743 (All), 644 (SMEs and Large); Countries: 15 (All), 13 (SMEs & Large); Adjusted R-squared: 0.586–0.660.
  - Joint F-test p-values for 4 lags ε_HH: 0.06–0.08 for All/SME specs; 0.03 for Large specs.
- Mechanisms and consistency:
  - Small firms often use personal real estate as collateral; asset-based loans more common for small/young firms; large corporates rely more on cash-flow lending.
  - Loan officer survey results align with firm-level corporate allocation riskiness measures.
  - Evidence suggests leakages from household-specific MaPPs more pronounced for credit to large corporates (corporate riskiness increases and lending to large corporates may ease).
- Model intuition:
  - When banks face binding funding constraints, increasing capital requirements on one portfolio (household/retail) can increase investment and riskiness in the other portfolio (corporate).
  - Time-varying risk sentiment can generate procyclicality and cross-portfolio leakage effects in good states.

### Policy implications and recommendations (from 5.4)
- Cross-sectoral leakages from household-specific macroprudential policies to corporate lending can be quantitatively important; consider these when choosing between broad-based and sector-specific MaPP tools.
- Early policy interventions recommended: implement macroprudential measures during early phase of credit cycle, when sectoral leakages are growing but likely still low.
- Timely monitoring of vulnerabilities and riskiness in both household and corporate sectors is key to avoid large undesired leakage effects.
- For corporate sector, information on distribution of firms by size in bank loan portfolios is relevant for assessing potential leakage magnitude.
- Designing macroprudential tools explicitly for corporates is an alternative, but difficult in practice due to heterogeneity of business models across firms.

*Italic — Source: wpiea2021113-print-pdf — Sections 3–5 (as provided in the content unit).*

### Section 3 describes the data sources and explains the construction of the corporate loan growth

### Section 3 — Data

### Overview
- Purpose: Describe data sources and the construction of the corporate loan growth indicator, macroprudential policy indexes, and corporate riskiness indicators used to identify effects of household-targeted macroprudential policy actions on banks’ corporate lending volumes.
- Frequency and sample window: annual frequency from 1998 to 2018 for bank-level loan volumes; quarterly frequency for corporate riskiness measures and macroprudential policy actions (mid-1990s until 2018Q4).

### Measuring corporate loan growth (CLG)
- Data source: bank-level volumes of corporate and residential mortgage loans from Fitch Connect.
- Sample coverage and constraints:
  - Time: 1998 to 2018 (annual).
  - Country sample limited to 13 countries: Brazil, Canada, Chile, China, Finland, Germany, Malaysia, Mexico, Poland, Russia, Spain, Sweden and Thailand.
  - Imposed conditions:
    - Minimum of 10 bank observations for each country-year.
    - Each bank must have at least 6 years of data.
    - Net loans are defined as total loans net provisions.
- Definition of variable of interest:
  - relative corporate loan growth (CLG) = the difference in the annual growth rates in the ratio of corporate loans to total (net) bank loans among the top 30 and bottom 30 percent of banks, when ranked according to the size of their residential mortgage portfolios (relative to total net loans).
- Corporate and residential loan category definitions:
  - Corporate loans category covers loans and leases to corporate and commercial enterprises; excludes non-residential and commercial property mortgage loans.
  - Residential mortgage loans category covers loans secured by residential property; excludes unsecured loans or loans secured by assets other than residential property.

### Measuring corporate credit riskiness
- Conceptual basis: Follows Greenwood and Hanson (2013) — time-variation in debt issuer quality signals financing conditions; measures combine corporate financial vulnerability and risk sentiment features.
- Firm-level data source: Datastream.
- Sample construction constraints:
  - Measures constructed quarterly for 29 economies (13 advanced plus 16 emerging).
  - Imposed minimum of 30 firm observations in each quarter for each country.
  - Series seasonally-adjusted to remove composition effects from lower-than-quarterly reporting frequency.
  - Sample is unbalanced; majority of economies have data back until 2002. For an additional 8 countries, biannual series were constructed.
- Three firm-level vulnerability indicators and corresponding riskiness measures:
  - TDTA (leverage-based measure): difference in the ratio of total debt to total assets of top 20 percent debt issuers over the current quarter and the bottom 20 percent debt issuers among listed firms.
  - TDtE (debt overhang measure): constructed as above but using the ratio of total debt to EBITDA instead of total debt to total assets.
  - ICR (interest coverage ratio measure): constructed as above but using the interest coverage ratio, defined as the ratio of interest expenses to EBITDA.
- Construction steps (preserving procedure precisely):
  - For each vulnerability indicator (TDTA, TDtE, ICR), each firm is assigned a value (from 1 to 10) according to the decile of the distribution of the indicator in a given quarter and the country of location.
  - Firms are sorted similarly according to the change in the net debt over the past year relative to past total assets. Debt includes both loans and bond financing.
  - Riskiness measure = difference in the average value of the vulnerability indicator for the top debt-takers (assigned values 9 or 10 for the change in net debt) and the average value of the vulnerability indicator for the bottom debt-takers (assigned values 1 and 2).
  - Interpretation: riskiness measures increase (decline) whenever the average vulnerability indicators increase more (increase less or decline) among top debt-takers relative to bottom debt-takers.
- Correlations and examples:
  - Cross-country average correlations among measures are reported (TDTA, TDtE, ICR). Noted that correlation is highest for ICR and TDtE given similar definitions.
  - Country examples: Sweden, Denmark and Mexico highlighted — Denmark has correlation of 0.12 between leverage-based and TDtE/ICR measures (second-lowest after Sweden); Mexico has correlation of 0.81.
- Rationale and prior evidence:
  - International Monetary Fund (2018) find these measures signal heightened downside risks to GDP growth and a higher probability of banking crises and banking sector stress, over and above signals from credit growth.

### Lending standards (robustness)
- Alternative robustness measure: bank lending standards for corporate loans from bank loan officer surveys, disaggregated by SMEs and large corporations.
- Data sources and coverage:
  - Data come from the ECB and national authorities, collected through Haver Analytics.
  - Available for 15 countries where macroprudential policy shocks can be constructed: Austria, Canada, Germany, Italy, Japan, Korea, Netherlands, Norway, Philippines, Poland, Portugal, Russia, Spain, Thailand and Turkey.
  - Time coverage generally goes back to 2002-2004 for most countries; shortest series: Thailand (2007Q4), Norway (2008Q1), Philippines (2009Q1), Russia (2009Q2).
- Note: Given reduction in sample size when using lending standards, the three riskiness measures (TDTA, TDtE, ICR) are treated as baseline measures.

### Macroprudential policies and indicators (MaPP)
- Primary source: IMF iMaPP database (records MaPP actions by sector targeted on a quarterly frequency from mid-1990s until 2018Q4).
- Coding convention: each policy change recorded as +1 for a tightening action, −1 for an easing action, and 0 if no policy change.
- Two constructed MaPP indicators:
  - MaPP_HH (household sector-targeting macroprudential policy actions indicator): baseline = sum of changes in LTV, LTI, DTI and DSTI limits. Robustness alternative MaPP_HH(Alternate) = sum of changes in LTV, LTI, DTI, DSTI limits and restrictions on household loan characteristics.
  - MaPP_BC (broad-based plus corporate sector-specific macroprudential policy actions indicator): includes changes in bank capital requirements (general and corporate-specific), changes in the countercyclical capital buffer, limits on credit growth (overall and corporate-specific) and restrictions on corporate loan characteristics. Robustness alternative MaPP_BC(Alternate) = captures changes in the countercyclical capital buffer, limits on credit growth (overall and corporate sector specific) and restrictions on corporate loan characteristics.
- Policy action counts in the 29-economy riskiness sample (2002Q1-2018Q4):
  - 120 changes in baseline measures of household-specific MaPPs.
  - 66 changes in baseline measures of broad-based and corporate-specific MaPPs.
  - Little overlap between household-specific and broad-based/corporate-specific interventions.
- Regional use and directionality:
  - Macroprudential policies used relatively more in emerging markets (EMs) relative to advanced economies (AEs) in the sample.
  - For both MaPP types, number of actions (tightening and easing) in EMs are almost double the number in AEs.
  - Total tightened vs eased during sample period: 154 tightenings and 34 easings.
  - Household-specific MaPPs were used more frequently than broad-based or corporate-specific measures.
  - 2018 saw a number of advanced economies tightening LTV and DSTI limits.

### Macroprudential shocks (addressing endogeneity)
- Rationale: Use shocks (residuals) from policy function estimations to reduce endogeneity concerns; follow literature applying two-stage procedures where policy functions for MaPPs are estimated first and residuals used as policy shocks.
- Expectation regarding endogeneity: endogeneity less of a concern for leakages because household-specific MaPP decisions likely taken independently of corporate sector conditions; nevertheless correlation between corporate credit riskiness and local household credit cycle necessitates shocks approach.
- Construction details:
  - Macroprudential shocks obtained as residuals from ordered probit regressions of MaPP_HH and MaPP_BC on macrofinancial control variables.
  - First-stage regressors include: real GDP growth, one-year ahead GDP growth forecast, real credit growth, capital account openness, financial development index, exchange rate change against U.S. dollar, and real house price growth.
  - Appendix A.2 (in source) describes computation methodology and shows robustness to alternative ordered probit specifications.
- Coverage:
  - Macroprudential shocks available for a subsample of 24 out of 29 countries.
  - Countries without constructed shocks due to data constraints: Argentina, Bulgaria, Chile, Peru and Vietnam.
  - Those countries are included in robustness regressions using policy actions instead of shocks (Appendix A.5).
- Note: Results are reported as robust to using policy actions instead of shocks (Appendix A.5).

### Other variables and data sources
- Macroeconomic and financial variables:
  - GDP growth rates and exchange rates: WEO database.
  - Credit to the private non-financial sector relative to GDP, domestic corporate bond issuance, domestic bank credit to the corporate sector: BIS.
  - Country-specific financial conditions indices (FCI): IMF.
  - Alternative domestic financial stance measure: 3-month money market rates from Datastream.
  - Global variables: MSCI VIX, U.S. policy rates from Datastream and Wu and Xia (2016).
  - GDP growth forecasts, financial development index, capital account openness used in first-stage regressions: Consensus forecasts, IMF, and IMF’s AREAER report, respectively.
- Appendix reference: Table A1 in Appendix A.1 lists all data sources.

*Source: wpiea2021113-print-pdf - Section 3 describes the data sources and explains the construction of the corporate loan growth*

### 4.1    Empirical Strategy

### 4.1    Empirical Strategy

### Baseline empirical specification
- Baseline panel regression for corporate loan growth (relative growth of corporate loans to net loans):
  - CLG_s_i,t = α + β1 ̄ε_HH_i,t + β2 ̄ε_BC_i,t + β3 ∆Credit_i,t + β4 X_i,t−1 + δ_i + δ_t + ε_i,t
  - ̄ε_HH_i,t and ̄ε_BC_i,t are macroprudential policy shocks (household-specific and broad-based/corporate-specific) in year t.
  - ∆Credit_i,t is year-on-year growth in the credit-to-GDP ratio.
  - X_i,t−1 includes year-on-year real GDP growth and domestic currency appreciation against U.S. dollar in the previous year.
  - Country and time fixed effects (δ_i, δ_t). Robust standard errors accounting for heteroskedasticity and autocorrelation.
  - Annual aggregation of MaPP shocks to match annual dependent variable frequency.

- Nonlinear specification with interaction with credit cycle:
  - CLG_s_i,t = α + β1 ̄ε_HH_i,t + β2 ̄ε_HH_i,t × ∆Credit_i,t + β3 ̄ε_BC_i,t + β4 ̄ε_BC_i,t × ∆Credit_i,t + β5 ∆Credit_i,t + β6 X_i,t−1 + δ_i + δ_t + ε_i,t

- Estimation approach:
  - Standard OLS used owing to relatively small sample size (both N and T small); GMM avoided because it would provide biased estimates in this sample context.

### Robustness checks (specifications considered)
- Additional controls included in robustness checks:
  - i) nominal short term market interest rates (NR) to account for domestic financial conditions;
  - ii) real growth of bank credit to private nonfinancial corporations (∆BankCredit);
  - iii) real growth of credit to households (∆HouseholdCredit).

---

### 4.2    Results and Robustness — Corporate Loan Growth

### Key regression findings (Table 4 summary)
- Column 1 (regression (1)):
  - ̄ε_HH_t coefficient: -0.14 (0.34) — negative but statistically insignificant.
  - ∆Credit_t coefficient: -10.41 (3.43) — significant (** p<0.05).
  - Interpretation: periods of high credit growth decrease the difference in corporate loan growth between banks most and least exposed to residential mortgages (banks with smaller residential mortgage portfolios experience higher corporate loan growth).

- Columns 2–6 (including interaction terms and additional controls):
  - Interaction ̄ε_HH_t × ∆Credit_t:
    - Column 2: 15.19** (6.51)
    - Column 3: 15.30* (7.96)
    - Column 4: 20.51** (8.61)
    - Column 5: 22.80** (7.94)
    - Column 6: 23.56** (8.29)
  - Standalone ̄ε_HH_t becomes negative and in Columns 4–6 statistically significant:
    - Column 4: -1.22** (0.53)
    - Column 5: -1.33* (0.61)
    - Column 6: -1.47* (0.67)
  - NR_t−1 (3-month nominal domestic market interest rate) coefficients (when included):
    - Column 3: -0.20** (0.08)
    - Column 4: -0.29*** (0.09)
    - Column 5: -0.27** (0.11)
  - ∆Credit_t remains negative and significant across columns (examples):
    - Column 2: -13.06** (4.96)
    - Column 4: -16.07* (8.52)
    - Column 6: -13.46** (5.48)

### Interpretation and quantitative magnitudes
- Non-linear relationship: the interaction term of household-specific MaPP shocks and credit growth is positive and statistically significant across Columns 2–6, indicating that during high credit growth banks with larger residential mortgage portfolios increase corporate loan growth more after an unexpected tightening of household MaPPs.
- Quantitative example:
  - When credit to GDP ratio grows at a rate of one standard deviation above the sample mean, the relative CLG increases by 54 percent of historical standard deviation (reported in text).
- Sample and fit:
  - Observations vary across columns: 132 (Cols 1–2), 123 (Col 3), 111 (Col 4), 102 (Cols 5–6).
  - Number of countries ranges from 11 to 13 across specifications.
  - Adjusted R-squared ranges from 0.07 to 0.23.

---

### 5.1    Empirical Strategy — Riskiness of Corporate Credit Allocation

### Baseline dynamic specification
- Baseline panel regression for corporate credit riskiness measures (CC_s_i,t ∈ {TDTA, TDtE, ICR}):
  - CC_s_i,t = α + γ CC_s_i,t−1 + Σ_{k=1}^4 β1,k ε_HH_i,t−k + Σ_{k=1}^4 β2,k ε_BC_i,t−k + β3 X_i,t−1 + δ_i + δ_t + ε_i,t
  - ε_HH_i,t−k and ε_BC_i,t−k are MaPP shocks in quarter t−k.
  - X_i,t−1 controls: i) quarterly yoy growth in credit-to-GDP ratio; ii) quarterly yoy real GDP growth; iii) domestic currency appreciation against U.S. dollar in previous quarter.
  - Robust standard errors accounting for heteroskedasticity and autocorrelation.

### Alternative (interaction) specification
- To capture nonlinear effects, include sum of past four quarters of MaPP shocks and interact with credit growth:
  - CC_s_i,t = α + γ CC_s_i,t−1 + β1 ̄ε_HH_i,t−1 + β2 ̄ε_HH_i,t−1 × ∆Credit_i,t−1 + β3 ̄ε_BC_i,t−1 + β4 ̄ε_BC_i,t−1 × ∆Credit_i,t−1 + β5 X_i,t−1 + δ_i + δ_t + ε_i,t
  - ̄ε_HH_i,t = Σ_{k=0}^3 ε_HH_i,t−k; ̄ε_BC_i,t similar; ∆Credit_i,t is yoy growth in credit-to-GDP in quarter t.

---

### 5.2    Baseline Results — Riskiness of Corporate Credit Allocation (Table 5)

### Persistence and control variable effects
- Lagged dependent variable (t−1) is positive and highly significant across all specifications:
  - TDTA t−1: 0.486*** (0.042)
  - ICR t−1: 0.328*** (0.040)
  - TDtE t−1: 0.385*** (0.037)
- ∆Credit_t−1 (lagged yoy growth of credit to private non-financial sector / GDP) is positive and highly significant:
  - Examples: 0.021*** (0.005) for TDTA (Col 1); 0.025*** (0.008) for ICR (Col 2); 0.032*** (0.007) for TDtE (Col 3).
- ∆GDP_t−1 positive and significant across measures (e.g., 0.044*** (0.010) for TDTA).
- ∆^q ER_t−1 (lagged qoq local currency appreciation vs USD) effects:
  - TDTA: -0.011* (0.005) in Column 1 and -0.011** (0.006) in Column 4.
  - ICR: 0.013* (0.006) in Column 2.

### MaPP shocks (lags) — baseline specification (columns 1–3)
- Household-specific MaPP lags on TDTA:
  - ε_HH, Lag 1: 0.004 (0.041)
  - ε_HH, Lag 2: 0.057 (0.044)
  - ε_HH, Lag 3: 0.116** (0.044)
  - ε_HH, Lag 4: 0.004 (0.045)
- Broad/corporate MaPP lags show mixed signs and are not jointly statistically significant.
- For ICR and TDtE, lags of MaPP shocks are generally not statistically significant.

### Interaction specification (columns 4–6) — nonlinear effects with credit growth
- Sum of household-specific MaPP shocks over past four quarters (̄ε_HH_t−1) standalone:
  - Col 4 (TDTA): 0.020 (0.024) — insignificant
  - Col 5 (ICR): -0.029 (0.032) — insignificant
  - Col 6 (TDtE): -0.067* (0.038) — significant at * p<0.1
- Interaction ̄ε_HH_t−1 × ∆Credit_t−1:
  - TDTA (Col 4): 0.012** (0.005)
  - ICR (Col 5): 0.014* (0.007)
  - TDtE (Col 6): 0.022** (0.008)
  - Interpretation: the interaction term is positive and statistically significant for all three riskiness measures, indicating that an unexpected tightening of household-specific MaPPs increases corporate credit riskiness more during periods of higher credit growth.
- Interaction ̄ε_BC_t−1 × ∆Credit_t−1:
  - TDTA: -0.018* (0.010) — statistically significant at * p<0.1
  - ICR: -0.022 (0.019) — insignificant
  - TDtE: -0.015 (0.014) — insignificant

### Quantitative magnitudes and economic significance
- Average effects:
  - When credit to GDP growth equals the sample average, an unexpected tightening of household-specific MaPPs over the past year increases corporate credit riskiness by about 1 percent of its historical standard deviation on average across measures.
  - When credit to GDP grows at one standard deviation above the sample mean, the impact on corporate credit riskiness is around 10 percent of historical standard deviation on average (Figure 4 referenced in text).
- Long-run average effect when credit relative to GDP grows at one standard deviation above mean:
  - Around 16 percent of the standard deviation across the three measures (17 percent for TDTA, 16 percent for ICR, 14 percent for TDtE).
- Historical context and sample prevalence:
  - Sample period: 2002Q1 to 2018Q4.
  - 19 percent of country-quarters have credit to GDP growth rate more than one standard deviation above sample mean (29 percent for emerging economies, 10 percent for advanced economies).
  - Pre-GFC period 2005–2008: average credit to GDP across countries stood at 14 percent; three riskiness measures increased by around 0.8 standard deviation over same period on average.
  - For credit growth of 14 percent (pre-GFC), the size of direct short-term spillovers from an unexpected household-specific MaPP tightening would be 23 percent of historical standard deviation across corporate credit riskiness measures.

### Inference
- Standalone household-specific MaPP terms are often insignificant in linear specifications but their interaction with the credit growth cycle reveals robust positive effects on corporate credit riskiness.
- Broad-based or corporate-specific MaPP interaction effects are negative across measures but statistically significant for TDTA only, suggesting limited effectiveness in reducing corporate credit riskiness relative to household-specific tools during credit expansions.

*Italic — Source: IMF working paper section 4.1–5.2 (as provided in the content unit).*

### 5.3    Robustness

### 5.3 Robustness

### Main robustness exercises and specification
- Starting specification (4) includes interaction between macroprudential variables and growth rate of credit to GDP ratio.
- Column 1 (Tables 6–8): control for domestic financial conditions index (FCI). Lag of FCI has a negative sign and is statistically significant across all three riskiness measures.
- Interaction terms of household-specific macroprudential policy shock ( ̄ε_HH ) with growth in credit to GDP remain positive and statistically significant across robustness checks.
- Results unchanged when using 3-month money market rates or domestic corporate spreads instead of FCI.
- Results unchanged when including global financial variables (VIX index and the federal funds rate) instead of time fixed effects (column 2 in Tables 6–8).

### Bank vs bond credit channel and extensions
- Measures of corporate credit riskiness do not distinguish bank versus bond credit; majority of MaPP measures affect bank lending only.
- Possible channels for household-specific MaPPs to affect non-bank corporate credit:
  - Perception of reduced risks in the economy leading agents to take on more risk (Bengui and Bianchi (2019) mechanism).
  - Banks competing more aggressively for corporate credit when retail lending opportunities shrink, potentially lowering corporate credit standards.
- Hypothesis: impact of MaPP measures on corporate credit riskiness is smaller where corporate sector relies more on bond funding versus bank lending.
- Extensions to baseline regression:
  - Include lagged year-on-year growth of quarterly corporate bond issuance (column 3, Tables 6–8).
  - Include four-quarter moving average of ratio of outstanding corporate bonds to outstanding corporate bank credit (“Bondcreditratioma”) and interactions with MaPP variables (column 4, Tables 6–8).

### Key empirical findings from robustness tables (summary)
- For all riskiness measures, an unexpected tightening of household-specific MaPPs increases corporate credit riskiness; result is robust to including growth of corporate bond issuance.
- Interaction terms of ̄ε_HH with corporate bond issuance are negative (consistent with hypothesis) but not statistically significant.
- Interaction terms of ̄ε_BC with credit to GDP growth become statistically significant and negative when corporate bond issuance included.
- Interaction terms of ̄ε_BC with corporate bond issuance are significant and positive; implying tightening of corporate and broad-based MaPP measures reduces corporate credit riskiness less when recent corporate bond issuance is larger.
- When including bonds-to-bank credit ratio:
  - Interaction between ̄ε_HH and credit to GDP growth is positive across measures but statistically significant only for TDtE.
  - Interaction of ̄ε_HH with bonds-to-bank credit ratio is negative but never statistically significant.

### Selected coefficient signs and significance (representative from Tables 6–8)
- TDTA (Table 6):
  - TDTA_{t−1}: values reported such as 0.49***, 0.50***, 0.47***, 0.47***, 0.52***, 0.48***.
  - ̄ε_HH_{t−1} × ∆Credit_{t−1}: coefficients include 0.01*, 0.01*, 0.01**, 0.01, 0.02***, 0.01*.
  - ̄ε_BC_{t−1} × ∆Credit_{t−1}: coefficients include -0.02*, -0.02, -0.03**, -0.02, -0.02*, -0.02*.
  - FCI_{t−1}: values include -0.13**, -0.15***, -0.07, -0.07, -0.14**, -0.12*.
  - Observations: 1,108; 1,108; 1,026; 1,013; 924; 1,108 (varies by column).
  - Number of Countries: 21, 21, 21, 20, 18, 21.
  - Adjusted R-squared: 0.40, 0.37, 0.39, 0.39, 0.42, 0.40.
- TDtE (Table 7):
  - TDtE_{t−1}: values include 0.36***, 0.37***, 0.36***, 0.35***, 0.37***, 0.36***.
  - ̄ε_HH_{t−1} × ∆Credit_{t−1}: coefficients include 0.02***, 0.02**, 0.02***, 0.02**, 0.02**, 0.02***.
  - FCI_{t−1}: values include -0.21**, -0.17**, -0.16*, -0.16, -0.22**, -0.21**.
  - Observations: 1,098; 1,098; 1,017; 1,003; 921; 1,098.
  - Number of Countries: 21, 21, 21, 20, 18, 21.
  - Adjusted R-squared: 0.26, 0.26, 0.26, 0.26, 0.28, 0.26.
- ICR (Table 8):
  - ICR_{t−1}: values include 0.32***, 0.33***, 0.31***, 0.30***, 0.33***, 0.32***.
  - ̄ε_HH_{t−1} × ∆Credit_{t−1}: coefficients include 0.02*, 0.01*, 0.02*, 0.01, 0.02*, 0.02*.
  - FCI_{t−1}: values include -0.20**, -0.17**, -0.15*, -0.15*, -0.21**, -0.21**.
  - Observations: 1,108; 1,108; 1,026; 1,013; 924; 1,108.
  - Number of Countries: 21, 21, 21, 20, 18, 21.
  - Adjusted R-squared: 0.19, 0.20, 0.19, 0.19, 0.20, 0.19.

(Notes: robust standard errors in parentheses; significance codes: *** p<0.01, ** p<0.05, * p<0.1.)

### Tests for outliers, house price effects, and alternative specifications
- Column 5 (Tables 6–8): exclude three countries (Canada, Netherlands, and South Korea) that predominantly used household-specific MaPPs and had few or no corporate-specific or broad-based MaPP changes. Results robust to exclusion.
- To exclude confounding from household credit market conditions (e.g., housing vulnerabilities):
  - Column 6 (Tables 6–8): control for lag of year-on-year growth in real house prices (∆HPI_{t−1}). Results remain statistically significant.
  - Cross-country average correlation between year-on-year real house price growth and the three corporate credit riskiness measures ranges between 0.08 and 0.14.
  - Re-run excluding five countries where correlation exceeded 0.4 on average (Brazil, Finland, Italy, Poland and Spain); results remain unchanged (results not shown).
- Additional robustness:
  - Results hold when using local projections akin to Jordá (2005) (Appendix A.4).
  - Results hold when using macroprudential policy actions instead of policy shocks (Appendix A.5).
  - Results robust to using Driscoll-Kraay standard errors to control for potential cross-section correlation; authors nonetheless use robust standard errors throughout because autocorrelation is a larger concern in dynamic specification.
  - Results hold when using pooled mean group estimator of Pesaran, Shin, and Smith (1997, 1999) allowing heterogeneity in error variances across countries.

### Alternative definitions of MaPP measures (Table 9)
- Replace baseline MaPP shocks with shocks constructed using alternate indicators of MaPP_HH and MaPP_BC.
- Columns 1–3 (alternate MaPP_HH indicator):
  - Interaction ̄ε_HH,Alternate × ∆Credit_{t−1} is always positive; ceases to be statistically significant for TDTA measure in some specifications.
- Columns 4–6 (alternate MaPP_BC indicator):
  - Interaction of baseline ̄ε_HH with ∆Credit_{t−1} remains positive and statistically significant for all three measures of corporate credit riskiness.
- Representative coefficients and diagnostics (Table 9):
  - Dependent variable lag coefficients: e.g., 0.489***, 0.326***, 0.381***, repeated across columns.
  - ̄ε_HH,Alternate_{t−1} × ∆Credit_{t−1}: 0.008, 0.009*, 0.014** (columns 1–3).
  - ∆Credit_{t−1}: 0.018***, 0.022**, 0.028***, etc.
  - Observations: 1,257; 1,257; 1,247; 1,257; 1,257; 1,247.
  - Number of Countries: 24 across columns.
  - Adjusted R-squared: 0.380, 0.172, 0.253, 0.380, 0.171, 0.254.

### Overall robustness conclusion
- Findings that household-specific macroprudential tightenings can amplify corporate credit riskiness (especially when credit growth is high) are robust across:
  - Controls for domestic financial conditions and global financial variables.
  - Inclusion of corporate bond issuance growth and bonds-to-bank credit ratios.
  - Exclusion of outlier countries and controls for house price growth.
  - Alternative MaPP definitions, alternative estimators (local projections, pooled mean group), and alternative standard error treatments (Driscoll-Kraay).
- While some interaction coefficients lose statistical significance in particular specifications or for particular risk measures, the core result—positive interaction of household-specific MaPP shocks with credit-to-GDP growth raising corporate credit riskiness—persists across a wide range of checks.

*Source: 5.3 Robustness, wpiea2021113-print-pdf*

### 5.4    Lending standards from loan officer surveys

### 5.4    Lending standards from loan officer surveys

### Regression results and interpretation
- Loan officer survey responses record changes in lending standards for corporate loans by firm size (SMEs and large corporates), where positive values indicate tightened standards and negative values indicate eased standards.
- Baseline regressions correspond to equation (3); Columns 1, 3, 5 report results without the domestic 3-month market rate; Columns 2, 4, 6 include the domestic 3-month market rate.
- Joint significance tests of the four lags of the household-specific macroprudential shocks (ε_HH) are statistically significant across specifications, but the direction of the impact differs by firm size:
  - For all corporates and for SMEs, past unexpected MaPP tightenings have a positive (tightening) impact on lending standards.
  - For large corporates, the net impact of MaPP tightening shocks over the past four quarters is negative, suggesting an easing of lending conditions by banks.
- Interpretation note: Columns with the domestic 3-m rate maximize sample size; positive lending-standards values = tighter lending standards.

### Table 10 — key coefficients and statistics
- Dependent variable: LS(All), LS(All) with domestic 3-m rate, LS(SMEs), LS(SMEs) with domestic 3-m rate, LS(Large), LS(Large) with domestic 3-m rate (Columns 1–6).
- Lagged dependent variable t−1:
  - 0.716*** (0.068)
  - 0.716*** (0.070)
  - 0.680*** (0.079)
  - 0.681*** (0.082)
  - 0.680*** (0.080)
  - 0.675*** (0.086)
- Household-specific MaPP shock (ε_HH) coefficients by lag:
  - Lag 1:
    - 1.301** (0.576)
    - 1.302** (0.591)
    - 1.897* (1.034)
    - 1.892* (1.045)
    - 0.0430 (0.743)
    - 0.076 (0.745)
  - Lag 2:
    - -1.216 (0.744)
    - -1.216 (0.744)
    - -1.351 (0.880)
    - -1.352 (0.880)
    - -0.991 (0.872)
    - -0.990 (0.876)
  - Lag 3:
    - -0.503 (0.737)
    - -0.503 (0.741)
    - 0.687 (1.010)
    - 0.687 (1.011)
    - -1.616* (0.869)
    - -1.615* (0.863)
  - Lag 4:
    - -1.440 (0.916)
    - -1.440 (0.915)
    - -0.037 (0.934)
    - -0.041 (0.929)
    - -0.783 (0.815)
    - -0.774 (0.830)
- Bank-condition shock (ε_BC) coefficients by lag:
  - Lag 1:
    - -1.140 (0.900)
    - -1.140 (0.979)
    - -2.258 (1.758)
    - -2.263 (1.770)
    - -2.051* (1.025)
    - -2.010* (1.125)
  - Lag 2:
    - -1.731 (1.348)
    - -1.730 (1.352)
    - -0.036 (1.425)
    - -0.041 (1.430)
    - -2.483 (1.881)
    - -2.448 (1.895)
  - Lag 3:
    - -0.680 (1.944)
    - -0.679 (1.805)
    - 0.616 (1.261)
    - 0.597 (1.220)
    - 0.570 (2.174)
    - 0.682 (1.985)
  - Lag 4:
    - -1.605 (1.498)
    - -1.604 (1.660)
    - -2.569* (1.297)
    - -2.589* (1.447)
    - -0.382 (1.930)
    - -0.261 (2.087)
- Other regressors:
  - ∆Credit t−1:
    - 0.177* (0.095)
    - 0.177* (0.100)
    - 0.251** (0.108)
    - 0.254** (0.112)
    - 0.157 (0.128)
    - 0.139 (0.137)
  - ∆GDP t−1:
    - 0.344* (0.193)
    - 0.344** (0.151)
    - 0.310 (0.242)
    - 0.302 (0.185)
    - 0.365 (0.252)
    - 0.414 (0.255)
  - ∆qER t−1:
    - 0.166 (0.132)
    - 0.166 (0.117)
    - 0.079 (0.148)
    - 0.076 (0.132)
    - 0.002 (0.123)
    - 0.023 (0.115)
  - NR t−1:
    - 0.003 (0.539)
    - -0.053 (0.560)
    - 0.335 (0.583)
- Sample and fit:
  - Observations: 743, 743, 644, 644, 644, 644
  - Number of Countries: 15, 15, 13, 13, 13, 13
  - Adjusted R-squared: 0.660, 0.659, 0.587, 0.586, 0.622, 0.622
  - Domestic 3-m rate included: No, Yes, No, Yes, No, Yes
  - Country FE: Yes (all columns)
  - Time FE: Yes (all columns)
- Joint F-test p-values for the four lags:
  - p-value (Joint F-test for 4 lags ε_HH): 0.06, 0.06, 0.08, 0.08, 0.03, 0.03
  - p-value (Joint F-test for 4 lags ε_BC): 0.28, 0.39, 0.32, 0.37, 0.16, 0.19
- Robust standard errors reported in parentheses. Significance levels: *** p<0.01, ** p<0.05, * p<0.1.
- Note: Additional variable NR is the 3-month nominal domestic market interest rate.

### Mechanisms and consistency with other evidence
- Small firms often use personal real estate as collateral for firm loans; asset-based (collateralized) loans are more common among small and young firms, while large and old corporates rely more on cash flow-based loans.
- Empirical and literature references summarized in the text:
  - The loan officer survey results align with firm-level measures of corporate allocation riskiness constructed earlier in the paper.
  - Evidence suggests leakages from household-specific MaPPs are more pronounced for credit to large corporates (i.e., corporate riskiness increases and lending to large corporates may ease).
- Model intuition (section following this empirical subsection):
  - When banks face binding funding constraints, increasing capital requirements on one loan portfolio (e.g., household/retail) can increase investment and riskiness in the other portfolio (corporate).
  - Time-varying risk sentiment can switch the economy between unconstrained and constrained funding cases, generating procyclicality and cross-portfolio leakage effects in good states of the world.

### Policy implications and recommendations
- Cross-sectoral leakages from household-specific macroprudential policies to corporate lending can be quantitatively important and should be considered when choosing between broad-based and sector-specific MaPP tools.
- Early policy interventions are supported: implement macroprudential measures during the early phase of the credit cycle, when sectoral leakages are growing but likely still low.
- Timely monitoring of vulnerabilities and riskiness in both household and corporate sectors is key to avoid large undesired leakage effects.
- For the corporate sector, information on the distribution of firms by size in bank loan portfolios is relevant for assessing potential leakage magnitude.
- Designing macroprudential tools explicitly for corporates is an alternative, but difficult in practice due to heterogeneity of business models across firms.

*Source: wpiea2021113-print-pdf - 5.4    Lending standards from loan officer surveys (https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021113-print-pdf.pdf).*

### REFERENCES

### wpiea2021113-print-pdf - REFERENCES

### References
- Acharya, Viral V., Katharina Bergant, Matteo Crosignani, Tim Eisert and Fergal McCann (2020), ‘The anatomy of the transmission of macroprudential policies’, IMF Working Paper 20/58.
- Acharya, Viral V., Tim Eisert, Christian Eufinger and Christian W. Hirsch (2019), ‘Whatever it takes: The real effects of unconventional monetary policy’, Review of Financial Studies 32(9), 3366–3411.
- Adelino, Manuel, Antoinette Schoar and Felipe Severino (2016), ‘Loan Originations and Defaults in the Mortgage Crisis: The Role of the Middle Class’, The Review of Financial Studies 29(7), 1635–1670.
- Ahnert, Toni, Kristin Forbes, Christian Friedrich and Dennis Reinhardt (2020), ‘Macroprudential fx regulations: Shifting the snowbanks of fx vulnerability?’, Journal of Financial Economics, forthcoming.
- Aiyar, Shekhar, Charles W. Calomiris and Tomasz Wieladek (2014), ‘Does macro-prudential regulation leak? evidence from a UK policy experiment’, Journal of Money, Credit and Banking 46(s1), 181–214.
- Akinci, Ozge and Jane Olmstead-Rumsey (2018), ‘How effective are macroprudential policies? an empirical investigation’, Journal of Financial Intermediation 33, 33–57.
- Alam, Zohair, Adrian Alter, Jesse Eiseman, Gaston Gelos, Heedon Kang, Machiko Narita, Erlend Nier and Naixi Wang (2019), ‘Digging deeper - evidence on the effects of macro-prudential policies from a new database’, IMF Working Paper No. 19/66.
- Araujo, Juliana D., Manasa Patnam, Adina Popescu, Fabian Valencia and Weijia Yao (2020), ‘Effects of macroprudential policy: Evidence from over 6,000 estimates’, IMF Working Paper No. 20/67.
- Auer, Raphael and Steven Ongena (2016), ‘The countercyclical capital buffer and the composition of bank lending’, BIS Working Paper No. 593.
- Ayyagari, Meghana, Thorsten Beck and Maria Soledad Martinez-Peria (2018), ‘The micro impact of macroprudential policies: Firm-level evidence’, IMF Working Paper 18/267.
- Bahaj, Saleem, Angus Foulis, Gabor Pinter and Paolo Surico (2019), ‘Employment and the collateral channel of monetary policy’, Bank of England working papers 827.
- Bengui, Julien and Javier Bianchi (2018), ‘Macroprudential policy with leakages’, Federal Reserve Bank of Minneapolis Working Paper 754.
- Brandão Marques, Luis, Gaston Gelos, Machiko Narita and Erlend Nier (2020), ‘Leaning against the wind: A cost-benefit analysis for an integrated policy framework’, IMF Working Paper No. 20/123.
- Brandão Marques, Luis, Qianying Chen, Claudio Raddatz, Jérôme Vandenbussche and Peichu Xie (2019), ‘The riskiness of credit allocation and financial stability’, IMF Working Paper No. 19/207.
- Cerutti, Eugenio, Stijn Claessens and Luc Laeven (2017), ‘The use and effectiveness of macro-prudential policies: New evidence’, Journal of Financial Stability 28, 203–224.
- Cizel, Janko, Jon Frost, Aerdt Houben and Peter Wierts (2019), ‘Effective macroprudential policy: Cross-sector substitution from price and quantity measures’, Journal of Money, Credit and Banking 51, 1209–1235.
- Danisewicz, Piotr, Dennis Reinhardt and Rhiannon Sowerbutts (2017), ‘On a tight leash: Does bank organizational structure matter for macroprudential spillovers?’, Journal of International Economics 109, 174–194.
- Forbes, Kristin and Michael Klein (2015), ‘Pick Your Poison: The Choices and Consequences of Policy Responses to Crises’, IMF Economic Review 63, 197–237.
- Gelos, Gaston and Alejandro Werner (2002), ‘Financial liberalization, credit constraints, and collateral: investment in the Mexican manufacturing sector’, Journal of Development Economics 67(1), 1–27.
- Greenwood, Robin and Samuel G. Hanson (2013), ‘Issuer Quality and Corporate Bond Returns’, The Review of Financial Studies 26(6), 1483–1525.
- Houston, Joel F., Chen Lin and Yue Ma (2012), ‘Regulatory arbitrage and international bank flows’, The Journal of Finance 67(5), 1845–1895.
- International Monetary Fund (2017), ‘Are countries loosing control of domestic financial conditions?’, Global Financial Stability Report: Chapter 3 (April).
- International Monetary Fund (2018), ‘The riskiness of credit allocation: A source of financial vulnerability?’, Global Financial Stability Report: Chapter 2 (April).
- Jordá, Òscar (2005), ‘Estimation and inference of impulse responses by local projections’, American Economic Review 95(1), 161–182.
- Kirti, Divya (2020), ‘Lending standards and output growth (July, 2018)’, ESRB: Working Paper Series No. 2018/79.
- Koop, Gary and Dimitris Korobilis (2014), ‘A new index of financial conditions’, European Economic Review 71, 101–116.
- Kuttner, Kenneth N. and Ilhyock Shim (2016), ‘Can non-interest rate policies stabilize housing markets? evidence from a panel of 57 economies’, Journal of Financial Stability 26, 31–44.
- Lian, Chen and Yueran Ma (2020), ‘Anatomy of corporate borrowing constraints’, Quarterly Journal of Economics, forthcoming.
- Ongena, Steven, Alexander Popov and Gregory F. Udell (2013), ‘When the cat’s away the mice will play: Does regulation at home affect bank risk-taking abroad?’, Journal of Financial Economics 108(3), 727–750.
- Temesvary, Judit (2018), ‘The role of regulatory arbitrage in U.S. banks’ international flows: Bank-level evidence’, Economic Inquiry 56(4), 2077–2098.
- Wu, Jing Cynthia and Fan Dora Xia (2016), ‘Measuring the macroeconomic impact of monetary policy at the zero lower bound’, Journal of Money, Credit and Banking 48(2-3), 253–291.
- Yang, Jing and Kostas Tsatsaronis (2012), ‘Bank stock returns, leverage and the business cycle’, BIS Quarterly Review March.

### Appendix A.1 — Data: sources and definitions
- Variables and data sources (selected entries):
  - total assets, total debt, EBITDA, Debt Interest Expense: balance sheet variables needed to compute the measures of corporate credit riskiness — Datastream.
  - 3-month money market rate: measure of domestic fin. conditions — Datastream.
  - MSCI VIX, U.S. policy rates: measures of global financial conditions — Datastream.
  - global shadow rate: shadow US policy rate — Wu and Xia (2016).
  - macroprudential policy indices MaPP_HH: a sum of indicators LTV, DSTI, iLoanR_HH; MaPP_BC: a sum of indicators Capital_Gen, Capital_Corp_Conservation, CCB, LCG_Gen, LCG_Corp, LoanR, LoanR_Corp — MaPP database; Alam et al. (2019).
  - real GDP: growth rates — WEO database.
  - exchange rate against USD: source not explicitly listed in table cell.
  - real house price index: growth rate — IMF.
  - Financial Conditions Index (FCI): index based on corporate, interbank, sovereign spreads, term spreads, long-term interest rates, equity return volatility, equity and house price returns, market share of the financial sector, and credit growth — IMF (2017).
  - credit to domestic nonfinancial private sector: ratios relative to GDP, and to domestic NFCs, NFC bonds outstanding — BIS.
  - corporate bond to credit ratio: ratio of NFC bonds outstanding to credit to nonfinancial corporates, four-quarter moving average — BIS.
  - Capital Account Openness: Chinn and Ito’s Index — IMF AREAER database.
  - average LTV limit: simple average of regulatory LTV limits on all existing loan categories; when a country does not have LTV limits, the value is set at 100 — MaPP database; Alam et al. (2019).
  - bank lending standards: measure of a change in lending standards from loan officer surveys; a positive value indicates a tightening — Haver Analytics.
  - residential mortgage loans, corporate and commercial loans: bank balance sheet variables needed to compute the CLG indicator — Fitch Connect.

### Appendix A.2 — Construction of macroprudential policy shocks
- Two-step procedure:
  1. Estimate ordered probit regression for categorical macroprudential indicator ̃MaPP^j_{i,t} with j∈{HH,BC}, taking values {−2,−1,0,1,2} for net actions in quarter t in country i:
     - ̃MaPP^j_{i,t} = α_i + μ_t + β_1 X_{i,t−1} + β_2 GDP^f_{i,t} + ε_{i,t}. (A.1)
     - X_{i,t−1} includes: year-on-year real GDP growth, quarter-on-quarter growth in credit to GDP ratio, financial development index (FDI), Chinn and Ito’s capital account openness index, quarter-on-quarter USD bilateral exchange rate change and quarter-on-quarter house price growth. FDI and CA openness enter with fourth lag. GDP^f_{i,t} is one-year ahead Consensus forecast of real GDP growth.
  2. Recover policy shock as residual:
     - ε̂^j_{i,t} = ̃MaPP^j_{i,t} − sum_{k=−2}^{2} p̂_k(X_{i,t−1}, GDP^f_{i,t})·k. (A.2)
     - p̂_k(·) is estimated probability that ̃MaPP^j_{i,t}=k conditional on RHS variables.
- Alternative specifications: columns (3)-(4) of Table A2 replace growth of credit to GDP ratio with bank credit growth and include bank credit growth and NFC bonds-to-GDP growth for MaPP_BC; add domestic FCI in MaPP_HH regression and 3-month money market rate in MaPP_BC regression. Preferred specification is (A.1) given sample considerations; results robust across specifications.

### Appendix A.2 — Table A2 (selected coefficients and stats)
- Table A2: Macroprudential shocks: first-stage regressions. Reported coefficients (standard errors in parentheses):
  - Columns (1) ̃MaPP_BC, (2) ̃MaPP_HH, (3) ̃MaPP_BC (alt), (4) ̃MaPP_HH (alt).
  - GDP^f: 0.142 (0.085); 0.156** (0.064); 0.250* (0.144); 0.082 (0.090).
  - FDI_{t−1}: −0.581 (1.622); −2.397** (1.104); −0.054 (2.187); −4.183*** (1.486).
  - FCI_{t−1} (in column 4): −0.353*** (0.132).
  - Observations: 2,861; 2,861; 1,718; 1,914.
  - Country FE: Yes. Time FE: Yes.
  - Significance notation: *** p<0.01, ** p<0.05, * p<0.1.

### Appendix A.3 — Macroprudential policies, household lending and house price growth
- Estimation setup:
  - Regressions of quarter-on-quarter growth in household credit-to-GDP ratio (∆_q HHCredit) and quarter-on-quarter real house price growth (∆_q HPI) on household-specific macroprudential policy shocks and their interaction with lagged household credit-to-GDP growth, controlling for real GDP growth, exchange rate appreciation against USD, past growth in household credit-to-GDP ratio and domestic 3-month money market rate.
- Main finding:
  - An unexpected household-specific MaPP tightening in the past four quarters reduces house price growth and growth of credit to the household sector when credit growth has been rapid in the recent past.
- Table A4: Impact coefficients (selected):
  - ̄ε_HH_{t−1} × ∆HHCredit_{t−1}: −0.031** (0.013) in HPI regressions (columns 1–2); −0.078*** (0.013) and −0.077*** (0.012) in HH Credit regressions (columns 3–4).
  - ∆_q HHCredit_{t−1}: 0.054*** (0.009); 0.050*** (0.009); 0.125*** (0.026); 0.116*** (0.024).
  - nominalrate_{t−1}: −0.135*** (0.046) and −0.244*** (0.068) in some specifications.
  - Observations: 1,331; 1,322; 1,331; 1,322. Number of Countries: 23. Adjusted R-squared: 0.275; 0.281; 0.370; 0.391.

### Appendix A.4 — Deriving impact on corporate credit riskiness using local projections
- Local projections setup (horizons h=1,2,..8):
  - CC^s_{i,t+h} = α^h_i + sum_{k=0}^1 β^h_{1,k} ε_HH_{i,t−k} + sum_{k=0}^1 β^h_{2,k} ε_BC_{i,t−k} + β^h_3 ε_HH_{i,t}×∆Credit_{i,t} + β^h_4 ε_BC_{i,t}×∆Credit_{i,t} + sum_{k=0}^1 β^h_{5,k} ∆Credit_{i,t−k} + sum_{k=0}^1 β^h_{6,k} ∆ER_{i,t−k} + sum_{k=0}^1 β^h_{7,k} ∆GDP_{i,t−k} + ε_{i,t+h}. (A.3)
- Result summary:
  - For all three measures of corporate credit allocation riskiness (TDTA, TDtE, ICR), the interaction term ε_HH_{i,t} × ∆Credit_{i,t} is broadly positive and at some horizons positive and statistically significant.
- Figure 6: plots coefficient β^h_3 for h=1..8 with 90 percent confidence intervals.

### Appendix A.5 — Results with MaPP actions (instead of shocks)
- Robustness test:
  - Replace MaPP shocks with MaPP_HH(BC) actions and their four-period sums MaPP_HH(BC)_sum in baseline specifications.
- Main findings from Table A5 (selected coefficients and stats):
  - Dependent variable lag t−1: 0.529*** (0.042); 0.354*** (0.035); 0.406*** (0.035) across specifications.
  - MaPP_HH, Lag 1: 0.005 (0.041); −0.064 (0.043); −0.078* (0.042).
  - MaPP_HH, Lag 2: 0.114** (0.048); 0.004 (0.042); −0.023 (0.054).
  - ∆Credit_{t−1}: 0.022*** (0.005); 0.017*** (0.005); 0.024*** (0.005) across columns.
  - MaPP_HH_sum: 0.042** (0.020); −0.021 (0.027); −0.048 (0.034).
  - MaPP_HH_sum × ∆Credit_{t−1}: 0.005 (0.005); 0.012** (0.005); 0.014*** (0.005).
  - MaPP_BC_sum × ∆Credit_{t−1}: −0.009 (0.006); −0.020* (0.010); −0.016 (0.010).
  - Observations: 1,571; 1,571; 1,559 (varies by column). Number of Countries: 27. Adjusted R-squared: 0.419; 0.156; 0.235 (varies).
- Note: MaPP_HH(BC) is the household-specific (broad-based or corporate-specific) MaPP policy action; MaPP_HH(BC)_sum is the sum of policy actions in the past four quarters.

*References and appendix content as provided in the source PDF.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2021/english/wpiea2021113-print-pdf.pdf_