## _wp15212 - References

## Source details

**Canonical URL:** [_wp15212 - References](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15212.pdf)

## Other formats

- [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15212.pdf.md)
- [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15212.pdf.json)

---

### I. Introduction
- Credit in emerging market economies (EMs) has been growing very rapidly during the past decade.
- Factors contributing to strong credit expansion include: macroeconomic stability, financial deepening, availability of new lending instruments, and economic growth.
- Country examples of high average credit growth (past decade): Brazil, Indonesia, Russia, and Turkey—average credit growth exceeded 10 percent per annum.
- Credit composition varies across EMs: corporate, consumer, and housing credit contribute differently to the expansion of the stock of credit.
  - Brazil, Indonesia, and Turkey: largest contribution of consumer credit to credit expansion; expansion of consumer and corporate credit contributed broadly equally.
  - Chile, China, and Singapore: corporate loans explain the bulk of the observed increase in credit.
- Research questions:
  - What has been the impact of credit growth on GDP growth?
  - Has the composition of credit (corporate, consumer, housing) mattered for GDP growth?
- Contribution:
  - Novel analysis of the impact of changes in credit composition on output using cross-country panel analysis.
- Core empirical findings (cross-country panel of 31 EMs, 2002–12):
  - Consumer credit: significantly positive effect on consumption but not on investment.
  - Corporate credit: impacts investment but not consumption.
  - Mortgage (housing credit): some evidence of contribution to economic growth through increasing consumption; evidence significant in two out of three empirical specifications.
- Country-specific case: Brazil—VARX model findings consistent with cross-country panel results.
- Note on risks: episodes of rapid credit growth can build vulnerabilities; historically, a minority of credit booms end in credit busts, though one-third of boom cases have ended up in financial crises.

### II. Stylized Facts
- Banking credit deepening (31 EMs sample):
  - On average, banking credit accounted for about 73 percent of GDP by end-2012 (compared to 42 percent in 2003).
- Sub-period dynamics:
  - Up to 2008: strong credit growth.
  - 2008–09 global financial crisis: credit growth receded.
  - 2010–12: credit growth picked up, though on average remained below pre-crisis rates.
- Contribution breakdown: corporate credit explains the bulk of credit growth, followed by mortgages and consumer credit.
- Regional/cross-country heterogeneity:
  - Latin America: credit more than doubled during 2003–12; current average level of financial deepening remains below average EM.
  - Mortgages share of total credit: about 60 percent in Hong Kong; less than 10 percent in Colombia, Korea, Russia, and Ukraine.
  - Corporate credit dominates in China, Philippines, Thailand, and Russia.
- Association with growth:
  - Positive association between credit growth and real GDP growth (noted for total, consumer, and corporate credit).

### III. Econometric Methodology and Data
- Objective: assess effects of corporate, consumer, and housing credit on real GDP growth and transmission channels (consumption and investment) using cross-country panel regressions.
- Dependent variables:
  - Consumption contribution to real GDP growth: (private consumption_t – private consumption_t-1)/GDP_t-1 (real terms).
  - Investment contribution to real GDP growth: analogous calculation; total investment contribution used in most EMs due to data availability.
- Credit variables: decomposed into contributions of corporate, consumer, and housing credit to total credit growth (preferred over growth rates).
- Baseline specification (notation preserved):
  - Dependent variable C_it is contribution of private consumption to real GDP growth.
  - Domestic controls X_it: domestic short-term interest rate change, real effective exchange rate, corporate issuances of bonds, equities and loans (real terms), government consumption growth.
  - Global controls Z_t: OECD real GDP growth, global interest rate (change in three-month LIBOR), and VIX.
  - Key regressors: corporate contribution, consumer contribution, housing contribution to credit growth (coefficients of interest denoted , , and  in text).
  - Δ(short-term interest rate) included to capture interest rate policy effects beyond bank lending.
  - Corporate issuances of bonds, equities and loans (deflated by CPI) control for non-bank financing.
  - Growth of government consumption captures fiscal stimulus.
  - Δ three-month LIBOR captures global liquidity and corporate funding costs.
  - VIX measures global risk aversion.
- Sample:
  - Quarterly panel of 31 EMs over 2002:Q1–2012:Q4.
  - Regional composition: 9 Asian EMs, 7 Latin American EMs, 13 European EMs, and 2 African EMs.
- Identification: decompose credit contributions and use domestic and global controls to isolate consumption and investment channels.

### IV. Data and Sample Details (Appendix I)
- Sample countries: Brazil, Bolivia, Botswana, Bulgaria, Chile, Colombia, Costa Rica, Croatia, Czech Republic, Estonia, Hong Kong, Hungary, Indonesia, Latvia, Lithuania, Macao, Malaysia, Mexico, Peru, Philippines, Poland, Romania, Russia, Singapore, Slovenia, South Africa, South Korea, Taiwan, Thailand, Turkey, and Ukraine.
- Variable definitions and sources (selected):
  - Short-term interest rate: policy rate or deposit rate; Haver analytics.
  - REER: based on CPI; IMF INS.
  - Corporate issuances: percent of GDP; Dealogic.
  - Government consumption: Haver analytics.
  - OECD real GDP, commodity prices, LIBOR: IMF IFS.
  - VIX: Bloomberg.
  - Nominal credit variables: central banks, Haver analytics, dXtime; deflated by CPI.
  - For countries with public and private lenders (e.g., Brazil), credit variables include total loans regardless of ownership.
- Estimation methods and identification:
  - Estimators: pooled OLS, OLS with cross-section fixed effects, pooled 2SLS, 2SLS with cross-section fixed effects, Arellano–Bond dynamic GMM.
  - Instruments for 2SLS/GMM: first- and second-order lags of global variables and their contemporaneous values; first- and second-order lags of credit variables; second-order lags of dependent and domestic control variables.
  - J-statistics: null that over-identifying restrictions are valid cannot be rejected in reported specifications.
  - Sample is unbalanced; time spans vary with data availability.

### V. Key Empirical Results (cross-country)
- Main channel results:
  - Consumer credit growth: significantly positive effect on real GDP growth through the consumption channel.
  - Corporate credit growth: significantly positive effect on real GDP growth through the investment channel.
  - Housing credit (mortgages): matters for economic growth through the consumption channel in two out of three specifications.
- Additional findings:
  - Δ(short-term interest rate): significantly negative impact on private consumption contribution (implies increased household saving).
  - Government consumption growth: significantly negative impact on private consumption contribution; significantly positive impact on investment contribution.
  - Capital market issuances: significant positive impact on investment contribution; not significant for private consumption.
  - ΔLog(OECD GDP): significantly positive impact on investment contribution; not significant for private consumption.
- Table 1 significance notation:
  - *** indicates statistical significance at the 5% level across all three specifications using the Arellano-Bond dynamic GMM estimator.
  - ** indicates statistical significance at the 5% level for two out of the three specifications using the Arellano-Bond dynamic GMM estimator.

### VI. Magnitudes and Interpretation
- GMM-estimated effects (average EM economy, Panels A and B, Appendix II):
  - One standard-deviation shock to consumer credit (contribution to credit growth) ≈ 4 percentage points → likely increases real GDP growth by 0.3 percentage points.
  - One standard-deviation shock to corporate credit (contribution to credit growth) ≈ 10 percentage points → can raise real GDP growth by about 0.3 percentage points.
  - Since consumption is less volatile than investment, a one-percentage-point increase in consumer credit has a larger impact on economic growth than a one-percentage-point increase in corporate credit.
- Explanations for smaller corporate-credit impact:
  - Corporates may use loan proceeds to build cash cushions when funding is cheap.
  - Borrowing may strengthen corporates’ liquidity position and resilience to financial shocks.

### VII. Brazil Case Study (VARX and time-series evidence)
- Context and credit-to-GDP:
  - Credit in Brazil reached about 50 percent of GDP by end-2012 (up from 24 percent in 2002).
- Growth averages:
  - 1996–2003: Real GDP growth averaged nearly 2 percent; credit expanded at an average of 1.5 percent in real terms.
  - 2004–08: average GDP growth rose to 4.8 percent; credit accelerated to average annual real rates of 19 percent.
  - Since 2010: real GDP growth and credit moderated to 3¾ percent and 12 percent, respectively.
- VARX specification:
  - Endogenous variables: corporate, consumer, housing contributions to credit growth; consumption and investment contributions to real GDP growth.
  - Controls: domestic and global controls as in cross-country regressions; international commodity price added.
  - Identification: Cholesky decomposition; ordering of credit variables aligned with model ordering; qualitative results robust to alternative orderings.
- Brazil VARX impulse-response magnitudes (annualized peak impact):
  - One-standard-deviation shock to (quarter-on-quarter) growth rate of Brazil’s consumer credit ≈ 2.6 percentage points → likely increases real GDP growth by about 0.2 percentage points over a year after the shock.
  - One-standard-deviation shock to (quarter-on-quarter) growth rate of corporate credit ≈ 2.9 percentage points → can raise real GDP growth by about 0.1 percentage points over a year after the shock.
  - Peak impacts: consumer credit shock on consumption contribution peaks contemporaneously; corporate credit shock on investment contribution peaks two quarters after the shock.
  - The effects of one-standard-deviation shocks to the growth rates of consumer and corporate credit are not significantly different from each other at the peak.

### VIII. Robustness and Estimation Nuances
- Multiple estimators used for robustness: pooled OLS, fixed-effects OLS, pooled 2SLS, fixed-effect 2SLS, Arellano–Bond dynamic GMM.
- OLS estimates likely biased due to endogeneity.
- Arellano-Bond dynamic GMM used to address dynamic panel bias; instruments include contemporaneous and lagged global variables and lagged credit and domestic controls.
- J-test results reported: instruments likely valid in reported specifications.
- Alternative specifications: replace Δ(interest rate) with Δ(interest rate)t-1 or include both to capture lagged monetary policy effects.

### IX. Policy Implications and Recommendations
- Credit composition matters for economic growth; different credit types operate through distinct channels:
  - Consumer credit → consumption channel.
  - Corporate credit → investment channel.
  - Housing credit → consumption channel in some specifications.
- Policy tools to influence credit composition:
  - Banking sector prudential tools (tightening/loosening loan eligibility criteria).
  - Limits on asset concentration.
  - Measures targeting credit growth for specific portfolios.
- Consideration for long-term growth: adjusting credit composition via prudential measures can help foster sustainable, balanced growth.

### X. Selected Numerical Estimates from Regression Tables (preserved exactly)
- GMM (Arellano-Bond) estimates (Panel A, selected coefficients, standard errors in parentheses):
  - Corporate credit: 0.01 (0.01), 0.01 (0.01), 0.02 (0.02) across columns.
  - Consumer credit: 0.11** (0.05), 0.03* (0.01), 0.10* (0.06) in different columns.
  - Housing credit: 0.17** (0.07), -0.03 (0.02), 0.21*** (0.07) in different columns.
  - Δ(Interest rate): -0.10** (0.05), -0.14** (0.07) reported in GMM columns.
  - ΔLog(Government consumption): -0.013*** (0.002), -0.03*** (0.01), -0.013*** (0.001) in different specifications.
  - ΔLog(OECD GDP): 0.03 (0.05), 0.11*** (0.02), 0.03 (0.05) in different columns.
  - Lagged dependent variable: 0.24*** (0.06), -0.18*** (0.05), 0.22*** (0.07) in different GMM columns.
- Panel B highlights (selected):
  - Corporate credit: 0.013** (0.006) in multiple OLS/FE pooled specifications; 0.03*** (0.01), 0.02** (0.01), 0.03*** (0.01) in select GMM columns.
  - Consumer credit: 0.04*** (0.01) in multiple OLS/FE pooled specifications; mixed and generally smaller coefficients in 2SLS/GMM columns.
  - Issuances of bonds, equities, and loans: 0.03*** (0.01) in several 2SLS/GMM columns; 0.18** (0.07), 0.36** (0.17), 0.17** (0.07) in select GMM columns.
  - ΔLog(OECD GDP): 0.18*** (0.03), 0.17*** (0.03) consistently across many specifications.
- Brazil VARX shocks (repeated):
  - Consumer credit q/q one-standard-deviation ≈ 2.6 percentage points → real GDP growth ↑ about 0.2 percentage points over a year.
  - Corporate credit q/q one-standard-deviation ≈ 2.9 percentage points → real GDP growth ↑ about 0.1 percentage points over a year.

*Source: _wp15212 - References (PDF chapter/section).*

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

### _wp15212 - References

### I. Introduction
- Credit in emerging market economies (EMs) has been growing very rapidly during the past decade.
- Factors contributing to strong credit expansion include: macroeconomic stability, financial deepening, availability of new lending instruments, and economic growth.
- Country examples of high average credit growth (past decade): Brazil, Indonesia, Russia, and Turkey—average credit growth exceeded 10 percent per annum.
- Credit composition varies across EMs: corporate, consumer, and housing credit contribute differently to the expansion of the stock of credit.
  - Brazil, Indonesia, and Turkey: largest contribution of consumer credit to credit expansion; expansion of consumer and corporate credit contributed broadly equally.
  - Chile, China, and Singapore: corporate loans explain the bulk of the observed increase in credit.
- Research questions addressed:
  - What has been the impact of credit growth on GDP growth?
  - Has the composition of credit (corporate, consumer, housing) mattered for GDP growth?
- Contribution of the paper:
  - Novel analysis of the impact of changes in credit composition on output using cross-country panel analysis.
- Summary of core empirical findings (cross-country panel of 31 EMs, 2002–12):
  - Consumer credit: significantly positive effect on consumption but not on investment.
  - Corporate credit: impacts investment but not consumption.
  - Mortgage (housing credit): some evidence of contribution to economic growth through increasing consumption; evidence significant in two out of three empirical specifications.
- Country-specific case: Brazil—findings from a VARX model are consistent with cross-country panel results.
- Note on risks: episodes of rapid credit growth can build vulnerabilities; historically, a minority of credit booms end in credit busts, though one-third of boom cases have ended up in financial crises (Dell’Ariccia et al (2012) referenced in text).

### II. Stylized Facts
- Banking credit deepening (across EMs in sample, 31 EMs):
  - On average, banking credit accounted for about 73 percent of GDP by end-2012 (compared to 42 percent in 2003).
- Sub-period dynamics:
  - Up to 2008: strong credit growth.
  - 2008–09 global financial crisis: credit growth receded.
  - 2010–12: credit growth picked up, though on average remained below pre-crisis rates.
- Contribution breakdown (across the decade): corporate credit has explained the bulk of credit growth, followed by mortgages and consumer credit.
- Regional and cross-country heterogeneity:
  - Latin America: credit more than doubled during 2003–12; current average level of financial deepening remains below average EM.
  - Chile: larger stock of credit in percent of GDP among Latin America economies; Mexico: smaller stock.
  - Mortgages share of total credit: about 60 percent in Hong Kong; less than 10 percent in Colombia, Korea, Russia, and Ukraine.
  - Corporate credit dominates in China, Philippines, Thailand, and Russia.
- Association with growth:
  - Positive association between credit growth and real GDP growth (Figure 7).
  - Correlation persists when focusing on consumer credit (Figure 8.a) and corporate credit (Figure 8.b).

### III. Literature Review
- Theoretical links between credit growth and economic activity are established (references cited in text: Levine, 2004; Harrison, Sussman, and Zeira, 1999).
- Empirical challenges: endogeneity between credit and GDP; a variety of instrument strategies used in literature.
  - Examples of empirical findings:
    - Peek et al. (2003): identified loan supply shocks effects on GDP in the U.S. using bank health measures as instruments.
    - Bassett et al. (2010): bank-level loan supply shocks in the U.S. show economically large effects on economic growth.
    - Driscoll (2004): using state-level U.S. panel with state-specific money demand shocks as instruments found no significant effects of bank loans on output.
    - Rondorf (2012): similar methodology applied to euro-area countries found significant effects of bank loans on output growth.
    - Miron et al. (1994): found no evidence for change in the lending channel of monetary policy over time.
- Limited research on credit composition effects:
  - Ludvigson (1999): U.S. aggregate time-series—positive relationship between growth of consumer credit and consumption.
  - Brigden and Mizen (1999): U.K. evidence—credit influences corporate sector investment.
  - Mian and Sufi (2014): mortgage refinancing (cash-out) in response to house price growth has large effect on household spending.
  - Beck et al. (2012): corporate credit—but not household credit—has a significantly positive impact on growth of GDP per capita; their cross-section regression averaged over the sample period does not capture dynamics.

### IV. Econometric Methodology and Data
- Objective: assess effects of corporate, consumer, and housing credit on real GDP growth and transmission channels (consumption and investment) using cross-country panel regressions.
- Dependent variables:
  - Consumption contribution to real GDP growth: (private consumption_t – private consumption_t-1)/GDP_t-1 (all in real terms).
  - Investment contribution to real GDP growth: calculated similarly; due to data availability issues, total investment contribution used rather than private investment for most EMs.
- Credit variables: decomposed into contributions of corporate, consumer, and housing credit to total credit growth—preferred over growth rates to avoid distortion (e.g., rapid mortgage growth where share is small).
- Baseline specification (notation preserved as in text):
  - Dependent variable C_it is contribution of private consumption to real GDP growth.
  - Regressors include domestic controls X_it: domestic short-term interest rate change, real effective exchange rate, corporate issuances of bonds, equities and loans (in real terms), government consumption growth.
  - Global controls Z_t: OECD real GDP growth, global interest rate (change in three-month LIBOR), and VIX.
  - Key regressors of interest: corporate contribution, consumer contribution, housing contribution to credit growth (coefficients of interest denoted , , and  in text).
  - Change in domestic short-term interest rate included to capture interest rate policy effects beyond bank lending.
  - Corporate issuances of bonds, equities and loans (deflated by CPI) included to control for non-bank financing.
  - Growth of government consumption captures fiscal stimulus.
  - Change in three-month LIBOR captures global liquidity and corporate funding costs.
  - VIX measures global risk aversion.
- Sample:
  - Quarterly panel of 31 EMs over period 2002:Q1–2012:Q4.
  - Regional composition: 9 Asian EMs, 7 Latin American EMs, 13 European EMs, and 2 African EMs.
- Identification strategy focuses on decomposing credit contributions and using domestic and global controls to isolate effects through consumption and investment channels.

### V. Key Empirical Results (summarized from text)
- Consumer credit:
  - Significantly positive effect on consumption contribution to GDP growth.
  - No significant effect on investment contribution.
- Corporate credit:
  - Significant effect on investment contribution to GDP growth.
  - No significant effect on consumption contribution.
- Housing credit (mortgages):
  - Some evidence of positive contribution to consumption and GDP growth.
  - Evidence significant in two out of three empirical specifications.
  - No robust evidence supporting impact of mortgages on investment through housing construction in this study.
- Country case: Brazil VARX results align with cross-country panel findings.

*Source: _wp15212 - References (PDF chapter/section)._

### Appendix I provides the list of the countries in the sample and a detailed description of the

### _wp15212 - Appendix I provides the list of the countries in the sample and a detailed description of the

### Data and sample
- Sample: 31 emerging markets (EMs) over the period 2002:Q1–2012:Q4, covering 9 Asian EMs, 7 Latin American EMs, 13 European EMs, and 2 African EMs.
- Countries included: Brazil, Bolivia, Botswana, Bulgaria, Chile, Colombia, Costa Rica, Croatia, Czech Republic, Estonia, Hong Kong, Hungary, Indonesia, Latvia, Lithuania, Macao, Malaysia, Mexico, Peru, Philippines, Poland, Romania, Russia, Singapore, Slovenia, South Africa, South Korea, Taiwan, Thailand, Turkey, and Ukraine.
- Time span varies depending on countries’ data availability of bank lending, particularly the composition of bank lending.

### Variable definitions and data sources
- Short-term interest rate: policy rate in most countries; where no policy rate available, deposit rates are used. Data from Haver analytics.
- Real effective exchange rate: based on consumer price index, taken from the Information Notification System (INS) of the IMF.
- Corporate issuances of bonds, equities and loans: defined in percent of GDP, taken from the Dealogic database.
- Government consumption: taken from Haver analytics.
- OECD real GDP (measure of global demand), international commodity prices, and LIBOR: obtained from the IMF’s International Financial Statistics (IFS) database.
- VIX (Chicago Board Options Exchange Market Volatility Index): taken from Bloomberg database.
- Nominal credit variables: different types of bank lending to the private sector obtained from countries’ central banks, Haver analytics, and dXtime database; nominal credit variables are deflated by CPI (from Haver analytics) to calculate real credit variables.
- For countries where bank lending is provided by both private and public financial institutions (e.g., Brazil), credit variables include total loans, regardless of lender ownership.

### Estimation methods and identification
- Cross-country dynamic panel model (model (2)) estimated with multiple methods: pooled OLS, OLS with cross-section fixed effects, pooled 2SLS, 2SLS with cross-section fixed effects, and Arellano–Bond dynamic GMM.
- 2SLS and Arellano–Bond GMM used to address endogeneity of credit variables.
- Instruments used in 2SLS and GMM: first- and second-order lags of global variables as well as their contemporaneous values; first- and second-order lags of the credit variables (consumer, corporate, housing contribution to credit growth); second-order lags of the dependent variable and domestic control variables.
- J-statistics for the J-test of over-identifying restrictions: null hypothesis that over-identifying restrictions are valid cannot be rejected (instruments likely valid).
- Alternative specifications: replace Δ(short-term interest rate) in model (2) by Δ(interest rate)t-1, or add Δ(interest rate)t-1 to model (2) to capture lagged impact of domestic monetary policy.
- Note: The sample is an unbalanced panel, and the time spans for different countries depend on their data availability. The first-order lags of the dependent variable and domestic control variables are most likely endogenous as well because the dependent variable contains a component of the lagged real GDP.

### Main empirical findings (cross-country)
- For the 31 EMs in the sample:
  - Consumer credit growth has a significantly positive effect on real GDP growth through the consumption channel.
  - Corporate credit growth has a significantly positive effect on real GDP growth through the investment channel.
  - Housing credit matters for economic growth through the consumption channel in two out of the three specifications (suggesting mortgage plays a role).
- Table 1 significance notation:
  - *** indicates statistical significance at the 5% level across all three specifications using the Arellano-Bond dynamic GMM estimator.
  - ** indicates statistical significance at the 5% level for two out of the three specifications using the Arellano-Bond dynamic GMM estimator.

### Magnitudes and interpretation
- GMM-estimated coefficients (Panels A and B, Appendix II) for the average EM economy:
  - One standard-deviation shock to consumer credit (in terms of contribution to credit growth) ≈ 4 percentage points → likely increases real GDP growth by 0.3 percentage points.
  - One standard-deviation shock to corporate credit (in terms of contribution to credit growth) ≈ 10 percentage points → can raise real GDP growth by about 0.3 percentage points.
  - Since consumption is less volatile than investment, a one-percentage-point increase in consumer credit has a larger impact on economic growth than a one-percentage-point increase in corporate credit.
- Possible explanations for smaller impact of corporate credit:
  - Corporates may use part of loan proceeds to build cash cushions when funding is cheap.
  - Borrowing may strengthen corporates’ liquidity position and resilience to financial shocks.
- Domestic short-term interest rate change:
  - Significantly negative impact on private consumption contribution to real GDP growth (implies increased household saving).
  - No significant impact on investment contribution after controlling for corporate credit and capital market issuances.
- Government consumption growth:
  - Significantly negative impact on private consumption contribution.
  - Significantly positive impact on investment contribution.
- Capital market issuances:
  - Significant positive impact on investment contribution but not on private consumption contribution.
- ΔLog(OECD GDP): significantly positive impact on investment contribution but not on private consumption contribution (consistent with export-driven investment stimulus).

### Brazil case study (VARX and time-series evidence)
- Context: Credit in Brazil expanded rapidly over the past decade, then decelerated; downturn phase started in early 2013 driven by slowdown in credit expansion by public banks while private bank credit continued moderate expansion.
- Credit-to-GDP: credit in Brazil reached about 50 percent of GDP by end-2012 (up from 24 percent in 2002).
- Growth averages:
  - 1996–2003: Real GDP growth averaged nearly 2 percent; credit expanded at an average of 1.5 percent in real terms.
  - 2004–08: average GDP growth rose to 4.8 percent; credit accelerated to average annual real rates of 19 percent.
  - Since 2010: real GDP growth and credit moderated to 3¾ percent and 12 percent, respectively.
- VARX specification:
  - Endogenous variables include corporate, consumer, housing contributions to credit growth, and consumption and investment contributions to real GDP growth.
  - Domestic and global controls as in cross-country regressions; international commodity price also included as a global variable in Brazil VARX.
  - Cholesky decomposition used to identify structural shocks; ordering of credit variables matches model ordering; qualitative results robust to different Cholesky orderings.
- Brazil VARX impulse-response magnitudes (annualized peak impact):
  - One-standard-deviation shock to (quarter-on-quarter) growth rate of Brazil’s consumer credit ≈ 2.6 percentage points → likely increases real GDP growth by about 0.2 percentage points over a year after the shock.
  - One-standard-deviation shock to (quarter-on-quarter) growth rate of corporate credit ≈ 2.9 percentage points → can raise real GDP growth by about 0.1 percentage points over a year after the shock.
  - Peak impacts: consumer credit shock on consumption contribution peaks contemporaneously; corporate credit shock on investment contribution peaks two quarters after the shock.
  - The effects of one-standard-deviation shocks to the growth rates of consumer and corporate credit are not significantly different from each other at the peak.

### Robustness and estimation nuances
- Multiple estimators used to check robustness: pooled OLS, fixed-effects OLS, pooled 2SLS, fixed-effect 2SLS, Arellano–Bond dynamic GMM.
- OLS estimates likely biased due to endogeneity of domestic explanatory variables.
- Due to dynamic panel bias, Arellano-Bond dynamic GMM estimates could be more reliable.
- Instruments (contemporaneous and lagged global variables; lagged credit and domestic controls) pass J-test for over-identifying restrictions.

### Policy implications and recommendations
- Credit growth composition matters for economic growth; shocks to different credit types operate through distinct channels (consumer → consumption; corporate → investment; housing → consumption in some specifications).
- Policy tools to achieve desired balance in credit composition:
  - Banking sector prudential tools such as tightening or loosening loan eligibility criteria.
  - Limits on asset concentration.
  - Measures targeting credit growth for certain credit portfolios.
- Consideration for long-term growth: adjusting credit composition via prudential measures can help foster sustainable, balanced growth.

### Selected numerical estimates from regression tables (examples preserved exactly as in source)
- GMM (Arellano-Bond) estimates (Panel A, selected coefficients, standard errors in parentheses):
  - Corporate credit: 0.01 (0.01), 0.01 (0.01), 0.02 (0.02) across columns reported.
  - Consumer credit: 0.11** (0.05), 0.03* (0.01), 0.10* (0.06) reported in different columns.
  - Housing credit: 0.17** (0.07), -0.03 (0.02), 0.21*** (0.07) reported in different columns.
  - Δ(Interest rate): -0.10** (0.05), -0.14** (0.07) reported in GMM columns.
  - ΔLog(Government consumption): -0.013*** (0.002), -0.03*** (0.01), -0.013*** (0.001) reported in different specifications.
  - ΔLog(OECD GDP): 0.03 (0.05), 0.11*** (0.02), 0.03 (0.05) in different columns.
  - Lagged dependent variable: 0.24*** (0.06), -0.18*** (0.05), 0.22*** (0.07) in different GMM columns.
- Brazil VARX shocks:
  - Consumer credit q/q one-standard-deviation ≈ 2.6 percentage points → real GDP growth ↑ about 0.2 percentage points over a year.
  - Corporate credit q/q one-standard-deviation ≈ 2.9 percentage points → real GDP growth ↑ about 0.1 percentage points over a year.

*Source: Appendix I–III and main text of the provided content unit.*

### Appendix II. Cross-Country Dynamic Panel Regressions Results (continued)

### _wp15212 - Appendix II. Cross-Country Dynamic Panel Regressions Results (continued)

### Panel A — Cross-country dynamic panel regressions for the contribution of consumption to real GDP growth (concluded)
- Dependent variable: Contribution of private consumption to real GDP growth.
- Estimation methods presented: OLS (Pooled), OLS (FE), 2SLS (Pooled), 2SLS (FE), GMM (Arellano-Bond).
- Cross-section fixed effect: Yes/No across columns as reported.
- Instruments reported for 2SLS/GMM columns: indicated (see Instruments1 column; some columns list "Yes").
- Prob(J-statistic) values (as reported across 2SLS/GMM specifications): 0.12 0.10 0.20 0.60 0.52 0.88 0.48 0.18 0.67
- Observations by specification (as reported): 1109 1094 1094 1109 1094 1094 1047 1047 1032 1047 1047 1032 1016 1016 1001
- Adjusted R-squared (Adj. R2) values (as reported): 0.18 0.18 0.18 0.27 0.27 0.28 -0.22 -0.15 -0.26 -0.10 -0.03 -0.22
- Note on standard errors and significance: Robust standard errors presented in parentheses. Asterisks *, **, *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.
- Definitions (as reported in note): Corporate/consumer/housing credit is the corresponding contribution to total credit growth. Government consumption is expressed in annualized quarter-on-quarter percent changes. Issuances of bonds, equities, and loans are in real terms, deflated by CPI. Interest rate and Libor are the first-order differences of short-term interest rate and the LIBOR respectively. REER is the first-order difference of the logarithms of real effective exchange rate. Lagged dependent variable is the first lag of the dependent variable.
- Instrument set description (footnote 1, as reported): The instruments for the 2SLS and GMM estimations include the first and second lags of global variables as well as their contemporaneous values, the first and second lags of credit variables, and the second lags of the dependent variable and domestic control variables.

### Panel B — Cross-country dynamic panel regressions for the contribution of investment to real GDP growth
- Dependent variable: Contribution of investment to real GDP growth.
- Estimation methods presented: OLS (Pooled), OLS (FE), 2SLS (Pooled), 2SLS (FE), GMM (Arellano-Bond).
- Key coefficient estimates (selected, reported across specifications):
  - Corporate credit:
    - 0.013** (0.006) in multiple OLS/FE pooled specifications.
    - 0.01 (0.02) in some 2SLS (Pooled) columns.
    - 0.03*** (0.01), 0.02** (0.01), 0.03*** (0.01) in select GMM columns.
  - Consumer credit:
    - 0.04*** (0.01) in multiple OLS/FE pooled specifications.
    - -0.01 (0.06), -0.01 (0.06), -0.002 (0.06) in some 2SLS (Pooled) columns.
    - 0.05 (0.04), 0.003 (0.02), 0.04 (0.04) in select GMM columns.
  - Housing credit:
    - 0.001 (0.001) to 0.002 (0.01) in pooled OLS.
    - -0.01 (0.02) to -0.01 (0.03) in some 2SLS columns.
    - -0.01 (0.05), -0.07*** (0.03), -0.02 (0.06) in select GMM columns.
  - Δ(Interest rate):
    - 0.001 (0.01); -0.01 (0.01); 0.0004 (0.009) in pooled OLS.
    - -0.06 (0.04); -0.04 (0.04) in some 2SLS/GMM columns.
  - Δ(Interest rate) t-1:
    - -0.02* (0.01) in multiple pooled OLS/FE columns.
    - Coefficients near zero (e.g., -0.005 (0.02), 0.001 (0.02)) in some 2SLS columns; 0.01 (0.01) in others.
  - ΔLog(REER):
    - 0.01 (0.01) in pooled OLS/FE.
    - -0.01 (0.06) to 0.003 (0.05) in some 2SLS/GMM columns.
    - 0.07* (0.04), 0.04 (0.04) in select GMM columns.
  - ΔLog(Government consumption):
    - 0.03*** (0.01) in pooled OLS and FE columns.
    - 0.05*** (0.02) and 0.05*** (0.01) in multiple 2SLS and GMM columns.
    - 0.04*** (0.003), 0.038*** (0.003), 0.037*** (0.003) in select GMM columns.
  - Issuances of bonds, equities, and loans:
    - -0.002 (0.003), -0.003 (0.003) in pooled OLS.
    - 0.005*** (0.002), 0.004* (0.002) in some FE OLS columns.
    - 0.03*** (0.01) in several 2SLS/GMM columns.
    - 0.18** (0.07), 0.36** (0.17), 0.17** (0.07) in select GMM columns.
  - ΔLog(OECD GDP):
    - 0.18*** (0.03), 0.17*** (0.03) consistently across many specifications.
    - 0.13*** (0.04), 0.16*** (0.03), 0.14*** (0.04) in select GMM columns.
  - Δ LIBOR:
    - Coefficients generally negative, e.g., -0.08 (0.15), -0.07 (0.15), -0.05 (0.15) in pooled OLS.
    - -0.22 (0.21), -0.21 (0.14), -0.24 (0.17) in some 2SLS columns.
    - -0.08 (0.14), -0.08 (0.15), -0.13 (0.15) in select GMM columns.
  - VIX:
    - -0.01 (0.01) across many specifications; 0.003 (0.008) in one GMM column; -0.003 (0.01) in another.
  - Lagged dependent variable:
    - -0.02 (0.07) in pooled OLS columns.
    - -0.04 (0.06) in some FE OLS columns.
    - Positive values in some 2SLS columns, e.g., 0.28 (0.24), 0.27 (0.21), 0.26 (0.21), and smaller positives 0.23 (0.25), 0.23 (0.22), 0.20 (0.22).
    - Near-zero in GMM columns: 0.02 (0.05), 0.02 (0.05), 0.01 (0.04).
- Constant terms (selected): -0.07 (0.13), -0.08 (0.13), -0.11 (0.15) in pooled OLS; -0.17 (0.21), -0.17 (0.22), -0.15 (0.22) in some FE/2SLS columns.
- Model diagnostics and samples:
  - Prob(J-statistic) values (2SLS/GMM columns reported): 0.32 0.32 0.30 0.25 0.24 0.21 0.83 0.66 0.81
  - Observations by specification (as reported): 1109 1094 1094 1109 1094 1094 1047 1047 1032 1047 1047 1032 1016 1016 1001
  - Adjusted R-squared (Adj. R2) values (as reported): 0.29 0.29 0.29 0.29 0.29 0.29 0.13 0.13 0.14 0.15 0.16 0.18
- Note on standard errors and significance: Robust standard errors presented in parentheses. Asterisks *, **, *** indicate statistical significance at the 10%, 5% and 1% levels, respectively.
- Definitions and instrument set: same definitions and instrument description as reported for Panel A (see note and footnote 1).

### Appendix III — Figure: Brazil-Specific VARX (cumulative impulse responses)
- Title: Brazil-Specific VARX: Cumulative Impulse Responses to One-Standard-Deviation Shocks.
- Impulse responses plotted for horizons 1 through 10 for:
  - Consumer credit shock → Impulse Response of Consumption Contribution (Percent).
  - Corporate credit shock → Impulse Response of Consumption Contribution (Percent).
  - Housing credit shock → Impulse Response of Consumption Contribution (Percent).
  - Consumer credit shock → Impulse Response of Investment Contribution (Percent).
  - Corporate credit shock → Impulse Response of Investment Contribution (Percent).
  - Housing credit shock → Impulse Response of Investment Contribution (Percent).
- Vertical axis scales in the figures (as printed): examples include ranges 0 to 1.2, -0.6 to 0.8, -0.4 to 0.6, -0.4 to 0.6, -0.6 to 0.6, etc., with horizons labeled 1 through 10 on the horizontal axis.
- Sources listed for the figure and underlying data: Haver Analytics; dXtime database; Dealogic database; Bloomberg database; IMF's International Financial Statistics (IFS) and Information Notice System (INS); and IMF staff calculations.

*Appendix notes, variable definitions, instrument description, and estimation details are reproduced exactly as presented in the source.*

---


_Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15212.pdf_