## wp17279 - Section 5 concludes.

## Source details

**Canonical URL:** [wp17279 - Section 5 concludes.](https://www.imf.org/-/media/files/publications/wp/2017/wp17279.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2017/wp17279.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2017/wp17279.pdf.json)

---

### Data and stylized facts
- Commodity net export price (CNEP) index
  - Country-specific CNEP index constructed from 45 commodities using the methodology of Gruss (2014) on monthly data.
  - CNEP formula: CNEP_i,t = Σ_{j=1}^J P_{j,τ} ω_{i,j,t}, where P_{j,τ} is the logarithm of the relative price of commodity j (in U.S. dollars and divided by the IMF’s unit value index for manufactured exports).
  - Weights ω_{i,j,t} are pre-determined three-year averages: ω_{i,j,t} = (1/3) Σ_{s=1}^3 [ x_{i,j,t−s} − m_{i,j,t−s} ] / [ Σ_{J} x_{i,j,t−s} + Σ_{J} m_{i,j,t−s} ].
  - Rationale: country-specific weights capture heterogeneous exposure and reflect changing export composition via pre-determined net-export shares.

- Macroeconomic and institutional data sources
  - GDP, domestic interest rates, inflation, exchange rates, credit to private sector: IMF’s World Economic Outlook and International Financial Statistics Database; World Bank’s World Development Indicators.
  - Index of capital account openness: Chinn and Ito (2006).

- Bank-level data (Bankscope sample)
  - Sample: 78 countries with 1,642 active banks for 2004-2015.
    - Out of 1,642 banks, 495 are headquartered in low-income countries and 249 banks are listed.
  - Bank-level variables: growth rate of total loans (nominal USD), equity/assets, deposits/assets, liquidity (liquid assets over deposit and short-term funding), size (log of total assets), ROA, NPLs/gross loans.
  - Stock price data for listed banks: Datastream and Bloomberg; country stock market indices: Bloomberg.
  - Summary loan-growth statistics:
    - Average loan growth: 18.6 percent.
    - Median loan growth: 12.6 percent.
    - Loan growth is about one percentage point higher for domestic versus foreign banks and for better capitalized banks.
    - Banks with low deposit funding and low NPL ratios show much higher loan growth.
  - Commodity dependence:
    - Median country share of commodity exports to GDP: around 7 percent.
    - Third quartile: above 15 percent.
    - Several countries have commodity exports > 20 percent of GDP (examples: Bolivia, Congo, Mauritania, Mongolia, Nigeria, Tajikistan, Vietnam, Yemen, Zambia).
  - Roughly 15 percent of banks in the sample are listed.

### Empirics: identification and sensitivity measures
- Bank sensitivity (SENS) to commodity prices
  - Constructed from listed-bank stock returns using a daily rolling-window regression (one-year window) per listed bank:
    - ln(P)_{bd} = α + β ln(COMMODITY_PRICE)_{d} + γ ln(MARKET_INDEX)_{d}; SENS for bank b in year t is the estimated β from the t−1 to t daily regression.
  - Distribution of β across 249 listed banks:
    - Approximately normally distributed.
    - Mean (and median) β larger in LICs than MICs: 0.284 versus 0.025.

- Imputation of SENS for unlisted banks
  - Two-step imputation:
    1. For each country-year, impute unlisted banks’ SENS as average SENS of listed banks in that country-year.
    2. For countries without listed banks, use the share of commodity exports in GDP as SENS.
  - Robustness: results robust to excluding banks headquartered in countries without listed banks.

- Outcome regression for loan growth (ΔLOANS)
  - Baseline panel specification:
    - ΔLOANS_{bct} = α1 (CNEP_{ct−1} × SENS_{bct−1}) + α2 CNEP_{ct−1} + α3 SENS_{bct−1} + COUNTRY_{ct−1} + BANK_{bct−1} + δ_b + τ_t + e_{bct}
    - COUNTRY_{ct−1}: country controls including real GDP growth and log domestic interest rates.
    - BANK_{bct−1}: lagged bank-specific controls (past loan growth, liquidity, size, capitalization, deposits-to-assets).
    - δ_b and τ_t: bank and year fixed effects; standard errors clustered at bank level.
  - Identification challenges:
    - Distinguishing credit supply versus credit demand channels; bank fixed effects and country controls mitigate but may not fully eliminate bank-specific demand heterogeneity.

### Results — main findings and magnitudes
- Baseline sample and fixed effects
  - Baseline sample: 1,642 banks in 78 countries; sequential specifications include year, country, and bank fixed effects; bank- and country-level controls added progressively.

- Commodity prices and bank lending: unconditional association
  - Positive and statistically precise relation between CNEP and bank lending in developing countries.
  - Magnitude:
    - One standard deviation in the CNEP index (broadly corresponding to the average decline between 2010 and 2013) is associated with a contraction of 1.5 percentage points in bank lending one year ahead (specification with all controls and fixed effects).

- Control variable effects (signs and significance)
  - Faster loan growth for more liquid, better capitalized, and smaller banks, and for banks with higher deposits-to-assets ratio.
  - Lagged loan growth: positive and significant (persistence).
  - Bank lending stronger in countries with faster growth; lending slows with contemporaneous interest rate increases.

- Heterogeneity by bank sensitivity (interaction CNEP × SENS)
  - Interaction term positive and significant: banks more sensitive to commodity price shocks curtail lending more when commodity prices fall.
  - Economic magnitudes:
    - One standard deviation decline of CNEP → 1.8 percentage points contraction in loan growth for banks highly exposed to the commodity sector (SENS at the 90th percentile).
    - One standard deviation decline of CNEP → 1.1 percentage points reduction for weakly exposed banks (SENS at the 10th percentile).

- Heterogeneity by country income group
  - Interaction significant and positive only for low-income countries (LICs).
  - Baseline model on LIC banks yields coefficient on interaction similar or marginally higher than whole-sample estimate.
  - Interpretation: shallower banking systems and higher commodity dependence in LICs strengthen transmission.

### 4.2 Bank heterogeneity — transmission channels and key contrasts
- Transmission channels
  - Credit demand: exporters postpone investment and reduce credit demand.
  - Credit supply: exposed banks experience liquidity shocks (retail deposits fall) or asset-quality deterioration (NPLs rise), leading to curtailed lending.
  - Supply-channel evidence: stronger transmission for banks with weak deposits-to-assets and higher NPLs.

- Heterogeneity empirical strategy
  - Banks split into "low" (lowest decile) and "high" (remaining banks) for capital, deposits, and NPLs; model interacts CNEP × SENS with bank-high / bank-low indicators.

- Key findings (LICs sample, Table 3 exact magnitudes preserved)
  - Bank capital: no significant differential role.
  - Bank deposits:
    - Low deposits-to-assets banks: one standard deviation decline in CNEP → lending contracts by 4.1 percentage points.
    - Median loan growth for low deposits-to-assets banks: 19.8 percent.
    - High-deposit bank: credit contraction of 1.1 percentage points.
  - Non-performing loans:
    - High-NPL banks: one standard deviation decline in CNEP → 1.7 percentage points lower credit growth.
    - Median growth for high-NPL banks: 15.6 percent.
  - Bank ownership:
    - No significant difference between foreign and domestic banks; point estimates extremely close.
    - Retail funding for banks located in LICs: foreign banks retail funding = 85 percent of total funding; domestic banks = 83 percent.
  - Robustness: results replicated in domestic-bank sub-sample; deposits-to-assets and NPLs remain key channels.

### Bank health and transmission channels (Table 4 exact coefficients)
- Dependent-variable regressions (LICs sample):
  - SENS_{t−1} coefficients:
    - Deposits/Assets: -0.0598.
    - NPLs: 7.9000***.
    - ROA: -0.4439.
    - Equity/Assets: 3.9417**.
  - CNEP_{t−1} × SENS_{t−1} effects:
    - Deposits/Assets: 0.0006*.
    - NPLs: -0.0787***.
    - ROA: 0.0045.
    - Equity/Assets: -0.0390**.
  - Observations and banks:
    - Deposits/Assets: Observations 2,842; Number of banks 494.
    - NPLs: Observations 2,111; Number of banks 395.
    - ROA: Observations 2,838; Number of banks 495.
    - Equity/Assets: Observations 2,848; Number of banks 495.

### 5.3 Other robustness checks — specification, sample, and measurement
- Specification and clustering
  - Static model (excluding lagged dependent variable): interaction coefficient "almost unchanged and precisely estimated".
  - Excluding bank fixed effects: results unchanged.
  - Clustering standard errors at country level: standard errors "around 10-25 percent larger" but key coefficients remain significant at least at the 5 percent level.

- Sample construction and measurement-error checks
  - Restricting to countries with at least one listed bank: sample drops "from over 495 to 239 banks"; interaction coefficient retains statistical significance.
  - Excluding countries where a single commodity has average weight > "5 percent" (Cote d’Ivoire, The Gambia, Mauritania, Niger, Tajikistan and Zambia): results robust.

- Alternative definitions and windows
  - Alternative SENS: two-year rolling-window betas on daily stock prices — results unaffected.
  - Loan growth in real USD: results unaffected.
  - Alternative commodity index CEP_alt (export-price weights ẑω_{i,j,t} = (1/3) Σ_{s=1}^3 x_{i,j,t−s} / Σ_{J j=1} x_{i,j,t−s}): main positive relationship remains significant.

- Robustness summary
  - Main findings robust to:
    - Excluding lagged dependent variable.
    - Excluding bank fixed effects.
    - Clustering at country level.
    - Restricting sample to countries with listed banks (sample drop: from over 495 to 239 banks).
    - Excluding countries where a single commodity accounts for > 5 percent average export weight (Cote d’Ivoire, The Gambia, Mauritania, Niger, Tajikistan and Zambia).
    - Two-year rolling-window SENS.
    - Loan growth in real USD.
    - Replacing CNEP with CEP_alt.

### Policy-relevant synthesis and conclusions (Section 6)
- Core synthesis
  - Lending by banks more exposed to the commodity sector reacts more to changes in commodity prices; this vulnerability is limited to low-income countries and stronger for countries with greater dependence on commodity exports.
  - Evidence favors supply-side mechanisms: commodity price shocks negatively affect deposit funding and asset quality, associated with a contraction of credit supply.
  - Banks with low deposits-to-assets ratios and with high NPLs reduce loan growth relatively more in response to a fall in commodity prices.
  - The bank lending channel can reinforce the direct negative effects of commodity price shocks on economic activity by further constraining firm growth and investment.

- Policy recommendation
  - "these findings underscore the importance to put in place strong financial regulation and macroprudential polices in developing countries, to strengthen the resilience of banking systems to commodity price shocks."

*Source: wp17279 - Section 5 concludes.*

### Section 5 concludes.

### wp17279 - Section 5 concludes.

### Data and stylized facts
- Commodity net export price (CNEP) index
  - Country-specific CNEP index constructed from 45 commodities using the methodology of Gruss (2014) on monthly data.
  - CNEP formula: CNEP_i,t = Σ_{j=1}^J P_{j,τ} ω_{i,j,t}, where P_{j,τ} is the logarithm of the relative price of commodity j (in U.S. dollars and divided by the IMF’s unit value index for manufactured exports).
  - Weights ω_{i,j,t} are pre-determined and calculated as a three-year average:
    - ω_{i,j,t} = (1/3) Σ_{s=1}^3 [ x_{i,j,t−s} − m_{i,j,t−s} ] / [ Σ_{J} x_{i,j,t−s} + Σ_{J} m_{i,j,t−s} ]
    - where x_{i,j,t−s} is export value (USD) of commodity j from country i in year t−s, and m_{i,j,t−s} is import value (USD) of commodity j by country i in year t−s.
  - Rationale: country-specific weights capture heterogeneous exposure to different commodity price movements and reflect changing export composition via pre-determined (three-year average) net-export shares.

- Macroeconomic and institutional data sources
  - GDP, domestic interest rates, inflation, exchange rates, credit to private sector: IMF’s World Economic Outlook and International Financial Statistics Database; World Bank’s World Development Indicators.
  - Index of capital account openness: Chinn and Ito (2006).
  - Variable definitions and data sources summarized in Table A2 (source document).

- Bank-level data (Bankscope sample)
  - Sample construction constraints: developing countries with at least 5 active banks in any year; banks with at least one non-missing value for key bank-level variables.
  - Final sample: 78 countries with 1,642 active banks for 2004-2015.
    - Out of 1,642 banks, 495 are headquartered in low-income countries and 249 banks are listed.
  - Bank-level variables used from Bankscope: growth rate of total loans (nominal USD), ratio of equity over assets, ratio of deposits over assets, liquidity (liquid assets over deposit and short-term funding), size (log of total assets), return on assets (ROA), ratio of non-performing loans to gross loans.
  - Stock price data for listed banks: Datastream and Bloomberg; country stock market indices: Bloomberg.
  - Summary loan-growth statistics:
    - Average loan growth: 18.6 percent.
    - Median loan growth: 12.6 percent.
    - Loan growth is about one percentage point higher for domestic versus foreign banks and for better capitalized banks.
    - Banks with low deposit funding and low NPL ratios show much higher loan growth than banks with higher deposits and NPL ratios.
  - Commodity dependence across countries/banks:
    - Median country share of commodity exports to GDP: around 7 percent.
    - Third quartile: above 15 percent.
    - In a number of countries, commodity exports account for more than 20 percent of GDP (examples in sample: Bolivia, Congo, Mauritania, Mongolia, Nigeria, Tajikistan, Vietnam, Yemen, Zambia).
  - Roughly 15 percent of banks in the sample are listed.

### Empirics: identification and sensitivity measures
- Bank sensitivity (SENS) to commodity prices
  - Direct observation of loans to commodity exporters is not available; instead construct bank-level time-varying sensitivity using listed-bank stock returns.
  - Daily rolling-window regression (one-year window) per listed bank:
    - ln(P)_{bd} = α + β ln(COMMODITY_PRICE)_{d} + γ ln(MARKET_INDEX)_{d}
    - ln(P)_{bd} is log of daily stock price of bank b on day d; ln(COMMODITY_PRICE)_{d} is log of commodity price index; ln(MARKET_INDEX)_{d} is log of country stock market index.
    - Sensitivity measure SENS for bank b in year t is the estimated β from the t−1 to t daily regression.
  - Distribution of estimated β across 249 listed banks:
    - Estimated β are approximately normally distributed.
    - Mean (and median) β larger in LICs than MICs: 0.284 versus 0.025.
    - Interpretation: greater dependence of banks headquartered in low-income countries on the commodity sector.

- Imputation of SENS for unlisted banks
  - Two-step imputation:
    1. For each country-year, impute unlisted banks’ SENS as the average SENS of listed banks in that country-year.
    2. For countries without listed banks, use the share of commodity exports in GDP as the measure of SENS.
  - Assumptions: banks in commodity-dependent countries are more affected by commodity price shocks and all banks in the country are equally exposed.
  - Robustness: Section 4 shows results robust to excluding banks headquartered in countries without listed banks.

- Outcome regression for loan growth (ΔLOANS)
  - Baseline panel specification (annual frequency, bank-country-year observations):
    - ΔLOANS_{bct} = α1 (CNEP_{ct−1} × SENS_{bct−1}) + α2 CNEP_{ct−1} + α3 SENS_{bct−1} + COUNTRY_{ct−1} + BANK_{bct−1} + δ_b + τ_t + e_{bct}
    - ΔLOANS_{bct}: growth rate of outstanding loans (nominal USD) of bank b in country c in year t.
    - COUNTRY_{ct−1}: country controls including real GDP growth and log domestic interest rates.
    - BANK_{bct−1}: lagged bank-specific controls (past loan growth, liquidity, size, capitalization, deposits-to-assets).
    - δ_b and τ_t: bank and year fixed effects.
    - Standard errors clustered at bank level.
  - Identification challenges:
    - Distinguishing credit supply versus credit demand channels: commodity price falls may reduce demand for credit (exporters slow down) and/or impair bank balance sheets (asset quality, higher NPLs) and thus supply.
    - Bank fixed effects control for systematic bank–firm matching (e.g., large banks lending to exporters).
    - Year fixed effects absorb global shocks; GDP growth and interest rates absorb part of country-specific demand.
    - Additional robustness with time-varying country fixed effects to absorb differential shifts in credit demand; however, bank-specific demand heterogeneity may remain.

### Results
- Baseline sample and fixed effects
  - Baseline sample: 1,642 banks in 78 countries.
  - Sequential specifications include year, country, and bank fixed effects; bank-level and country-level controls added progressively.

- Main findings on commodity prices and bank lending
  - Unconditional association: positive and statistically precise relation between commodity net export prices and bank lending in developing countries.
  - Magnitude:
    - One standard deviation in the CNEP index (in the sample corresponding broadly to the average decline between 2010 and 2013) is associated with a contraction of 1.5 percentage points in bank lending one year ahead (specification with all controls and fixed effects).
  - Control variable effects (signs and significance)
    - Loan growth is faster for more liquid, better capitalized, and smaller banks, and for banks with a higher deposits-to-assets ratio.
    - Positive and significant coefficient on lagged loan growth indicates persistence.
    - Bank lending is stronger in countries with faster growth (effect decreases comparing contemporaneous and lagged growth).
    - Bank lending slows in response to a contemporaneous increase in interest rates.

- Heterogeneity by bank sensitivity (interaction CNEP × SENS)
  - Interaction term CNEP × SENS is positive and significant: banks more sensitive to commodity price shocks curtail lending more when commodity prices fall.
  - Economic magnitudes:
    - One standard deviation decline of the CNEP index is associated with:
      - a 1.8 percentage points contraction in loan growth for banks highly exposed to the commodity sector (SENS at the 90th percentile).
      - a 1.1 percentage points reduction for weakly exposed banks (SENS at the 10th percentile).

- Heterogeneity by country income group
  - Splitting the interaction by low-income countries (LICs) versus middle-income countries (MICs):
    - The interaction term is significant and positive only for LICs.
    - Running the baseline model on the subsample of banks headquartered in LICs yields a coefficient on the interaction term similar, albeit marginally higher, than the estimate on the whole sample.
  - Interpretation:
    - Possible reasons: less developed and shallow banking systems in LICs are more vulnerable to commodity price fluctuations; higher dependence of LICs on commodity exports implies stronger transmission from prices to credit.

- Analytical focus going forward
  - Subsequent analysis in the report focuses on the sub-sample of banks located in LICs to explore heterogeneous effects across bank and country characteristics and to investigate transmission channels in greater detail.

*Source: wp17279 - Section 5 concludes.*

### 4.2    Bank heterogeneity

### 4.2 Bank heterogeneity

### Transmission channels: credit demand vs credit supply
- Commodity price declines can affect bank lending through two channels:
  - Credit demand channel: exporters postpone investment and reduce demand for credit.
  - Credit supply channel: highly exposed banks experience liquidity shocks (retail deposits fall) or deterioration in asset quality (NPLs rise), leading to curtailed lending.
- If the credit supply channel operates, the transmission should be stronger for banks with weak deposit-to-assets ratio and worse asset quality (NPLs to total loans).

### Empirical strategy for heterogeneity
- Banks are split into "low" (lowest decile of the distribution for the bank characteristic) and "high" (remaining banks) groups for three balance-sheet measures: bank capital, deposits, and non-performing loans (NPLs).
- The model interacts CNEP (commodity net export price index) with bank sensitivity and bank-high / bank-low indicators to estimate differential effects.

### Key findings on bank-level heterogeneity (summary of Table 3)
- Bank capital: no significant role in differential transmission.
- Bank deposits:
  - Banks with low deposits-to-assets ratio exhibit a stronger transmission of commodity price shocks to loan growth.
  - Quantitatively: in response to a one standard deviation decline of the commodity net export price index, a bank with high exposure (90th percentile) and low-deposit contracts lending by 4.1 percentage points.
  - Median loan growth for banks with low deposits-to-assets ratio is 19.8 percent.
  - By contrast, the credit contraction for a bank with high deposits is 1.1 percentage points.
- Non-performing loans:
  - Banks with high non-performing loans to gross loans show stronger transmission than those with low NPLs.
  - Quantitatively: a one standard deviation decline in CNEP leads to a 1.7 percentage points lower credit growth for banks with relatively higher NPLs.
  - Median growth for these banks is 15.6 percent.
- Statistical significance: p-values reported in Table 3 indicate the high vs low category differences for deposits and NPLs are statistically significant.
- Interpretation: findings lend support to the credit supply channel—exposure to commodity prices affects credit supply when a bank’s deposit base dries up or asset quality deteriorates.

### Bank ownership (foreign vs domestic)
- No significant difference in the transmission behavior of foreign and domestic banks; point estimates are extremely close (Table 3, column 4).
- Possible explanation: similarity of business models between domestic and foreign banks operating in LICs; many foreign banks headquartered in LICs rely on retail funding similar to domestic banks.
  - For banks located in LICs: retail funding is equal to 85 percent of total funding for the average foreign bank and 83 percent for the average domestic bank.
  - By contrast, in emerging markets the average share of retail funding for foreign banks is 61 percent.
- Robustness: results replicated in the sub-sample of domestic banks (columns 5 and 6) — deposits-to-assets ratio and non-performing loans remain the key channels.

### Economic relevance
- Example magnitudes preserved exactly:
  - One standard deviation decline in the commodity net export price index → high-exposure, low-deposit bank: lending contracts by 4.1 percentage points.
  - Median loan growth (low deposits-to-assets): 19.8 percent.
  - High-deposit bank: credit contraction of 1.1 percentage points.
  - High-NPL banks: one standard deviation decline in CNEP → 1.7 percentage points lower credit growth; median growth 15.6 percent.

*Source: wp17279 - 4.2    Bank heterogeneity*

### 5.3    Other robustness checks

### 5.3    Other robustness checks

### Robustness tests on model specification and clustering
- Estimating a static model (excluding the lagged dependent variable) leaves the coefficient on the interaction term "almost unchanged and precisely estimated" (column 1).
- Excluding bank fixed effects does not change results (column 2).
- Clustering standard errors at the country level (instead of bank level) yields standard errors "around 10-25 percent larger than in the baseline," but the key estimated coefficients remain significant at least at the 5 percent level (column 3).

### Sample construction and measurement-error checks
- The sensitivity measure SENS merges bank- and country-level exposure to the commodity sector to enable a sufficiently large sample.
- To address potential measurement error from country-level exposure, the sample is restricted to countries with at least one listed bank (thereby excluding banks for which the sovereign measure—the share of commodity exports over GDP—was used). In this sub-sample the measure of sensitivity varies only at the bank-level.
  - Sample size falls "from over 495 to 239 banks"; the coefficient on the interaction term retains statistical significance (column 4).
- To mitigate endogeneity concerns from large country events that could affect world commodity prices, countries where a single commodity has the highest average weight greater than "5 percent" are excluded. The excluded countries are: Cote d’Ivoire, The Gambia, Mauritania, Niger, Tajikistan and Zambia. Results are robust to this exclusion (column 5).

### Alternative variable definitions and windows
- Alternative sensitivity measure: estimating beta coefficients from bank-level regressions on daily stock prices using a two-year rolling window. Results are unaffected (column 6).
- Alternative loan growth definition: calculating loan growth in real USD rather than in nominal terms. Results are unaffected (column 7).
- Alternative country-specific commodity price index: commodity export prices (CEP) defined as
  - CEP_alt_i,t = Σ_{j=1}^J P_{j,τ} * ẑω_{i,j,t}
  - where P_{j,τ} is the logarithm of the relative price of commodity j in period t in U.S. dollars (divided by the IMF’s unit value index for manufactured exports);
  - weights ẑω_{i,j,t} are pre-determined three-year averages: ẑω_{i,j,t} = (1/3) Σ_{s=1}^3 x_{i,j,t−s} / Σ_{J j=1} x_{i,j,t−s}, with x_{i,j,t−s} the export of commodity j from country i in year t−s.
- Estimating the baseline model using CEP_alt shows the positive relationship between commodity prices and loan growth remains significant (column 8).

### Key empirical conclusions from robustness checks
- Main findings are robust to:
  - Excluding the lagged dependent variable.
  - Excluding bank fixed effects.
  - Clustering at the country level (despite 10-25 percent larger standard errors).
  - Restricting the sample to countries with at least one listed bank (sample drop: from over 495 to 239 banks).
  - Excluding countries where a single commodity accounts for more than 5 percent average export weight (Cote d’Ivoire, The Gambia, Mauritania, Niger, Tajikistan and Zambia).
  - Using alternative sensitivity measures (two-year rolling-window betas on daily stock prices).
  - Measuring loan growth in real USD.
  - Using an alternative commodity price index based on commodity export prices (CEP_alt).

### Policy-relevant synthesis (from Section 6 Conclusions)
- Lending by banks more exposed to the commodity sector reacts more to changes in commodity prices; this vulnerability is limited to low-income countries and stronger for countries with more dependence on commodity exports.
- Evidence suggests supply-side mechanisms: commodity price shocks negatively affect deposit funding and asset quality, associated with contraction of credit supply.
- Banks with low deposits-to-assets ratios and with high non-performing loans reduce loan growth relatively more in response to a fall in commodity prices.
- Implication: the bank lending channel can reinforce the direct negative effects of commodity price shocks on economic activity by further constraining firm growth and investment.
- Policy recommendation: "these findings underscore the importance to put in place strong financial regulation and macroprudential polices in developing countries, to strengthen the resilience of banking systems to commodity price shocks."

*Source: IMF Working Paper — section 5.3 "Other robustness checks" and Section 6 "Conclusions" (wp17279).*

### References

### wp17279 - References

### Major bibliographic themes
- Empirical literature on bank lending, monetary policy transmission, and internal capital markets:
  - ACHARYA, V. V. and STEFFEN, S. (2015). The “greatest” carry trade ever? understanding eurozone bank risks. Journal of Financial Economics, 115(2), 215–236.
  - CAMPBELL, M. (2002). Internal capital markets in financial conglomerates: Evidence from small bank responses to monetary policy. Journal of Finance, 57(6), 2773–2805.
  - CETORELLI, N. and GOLDBERG, L. S. (2011). Global banks and international shock transmission: Evidence from the crisis. IMF Economic Review, 59(1), 41–76.
  - IVASHINA, V., SCHARFSTEIN, D. S. and STEIN, J. C. (2015). Dollar funding and the lending behavior of global banks. The Quarterly Journal of Economics, 130(3), 1241–1281.
- Commodity prices, growth, and macroeconomic volatility in developing countries:
  - DEATON, A. (1999). Commodity Prices and Growth in Africa. Journal of Economic Perspectives, 13(3), 23–40.
  - HAMILTON, J. D. (2003). What is an oil shock? Journal of Econometrics, 113(2), 363–398.
  - GRUSS, B. (2014). After the boom–commodity prices and economic growth in Latin America and the Caribbean. IMF Working Paper 14/154.
  - MLACHILA, M. and OUEDRAOGO, R. (2017). Financial Development Resource Curse in Resource-Rich Countries: The Role of Commodity Price Shocks. IMF Working Paper 17/163.
- Financial development, inclusion, and firm-level credit access:
  - ALLEN, F., CARLETTI, E., CULL, R., QIAN, J. Q., SENBET, L. and VALENZUELA, P. (2014). The African financial development and financial inclusion gaps. Journal of African Economies, 23(5), 614–642.
  - BANERJEE, A. V. and DUFLO, E. (2014). Do Firms Want to Borrow More? Testing Credit Constraints Using a Directed Lending Program. Review of Economic Studies, 81(2), 572–607.
  - BURGER, A. N. and UDELL, G. F. (2002). Small business credit availability and relationship lending: The importance of bank organisational structure. The Economic Journal, 112(477), F32–F53.
- Cross-border banking, foreign bank participation, and international transmission:
  - CLAESSENS, S. and VANHOREN, N. (2014). Foreign Banks: Trends and Impact. Journal of Money, Credit and Banking, 46(s1), 295–326.
  - CULL, R. and MARTINEZPERIA, M. S. (2010). Foreign bank participation in developing countries: Policy Research Working Paper Series 5398, The World Bank.
  - PEEK, J. and ROSENGREN, E. S. (1997). The international transmission of financial shocks: The case of Japan. American Economic Review, 87(4), 495–505.
- Measurement, methods, and data sources cited in the references:
  - Gruss (2014) commodity price index definitions used for country-specific commodity price index.
  - Bankscope used extensively for bank-level variables (Loan Growth, Size, NPL, Deposits/Assets, Equity, Liquidity, ROA).

### Figures (captions and notes)
- Figure 1: Commodity prices and bank lending: transmission channels
  - Conceptual flow: Fall in commodity prices → Fall in government revenues and/or firm profits → Lower credit demand (Demand Channel); Fall in commodity prices → Liquidity shock and Deterioration of bank health → Reduction in bank lending (Supply Channel).
- Figure 2: Commodity prices and loan growth
  - Loan growth (annual, nominal USD) plotted against Commodity price index (CNEP); separate series for Middle-income countries and Low-income countries. Notes: CNEP grouped into 20 equal-sized bins; data on loan growth from Bankscope.
- Figure 3: Commodity prices and bank returns
  - Commodity price index = BCOM Index from Bloomberg. Stock returns for FBN Holdings and Skye Bank from Datastream.
- Figure 4: Distribution of the bank-specific sensitivity measure to commodity prices
  - Kernel density, 249 banks, 1,204 bank-year observations. Notes: Sensitivity measured as factor loadings from regression of daily bank stock prices on commodity price index over one-year rolling window controlling for overall stock market index (equation 3).
- Figure 5: Share of commodity exports over GDP, sample distribution
  - Time series median and quartiles of Export share of commodities in GDP (%) for years 2004–2014.
- Figure 6: Distribution of the bank-specific specialization in commodities
  - Kernel density, 75 banks, 462 bank-year observations. Specialization measure computed as difference in R-squared of regressions of daily bank stock prices with and without broad stock market index (equations 6 and 7).

### Tables — key statistics and highlighted empirical results
- Sample and summary statistics:
  - Full sample: Observations for loan growth = 9,540; Number of banks = 1,642 (full sample across tables).
  - Low-income countries subset: Observations = 2,848; Number of banks = 495.
  - Loan growth (∆LOAN) full sample: Obs 9,540; Mean 18.68; S.D. 33.04; 25th -0.34; Median 12.57; 75th 29.04.
  - Loan growth (∆LOAN) low-income countries: Obs 2,848; Mean 20.89; S.D. 29.73; 25th 3.22; Median 15.85; 75th 32.61.
  - Liquidity (full sample): Mean 33.34.
  - Size (full sample): Mean 13.21.
  - Equity (full sample): Mean 15.26.
  - Deposits/Assets (full sample): Mean 71.11.
  - NPLs (full sample): Mean 7.52; NPLs (low-income) mean 6.82.
- Baseline regression results (Table 2; loan growth dependent variable, annual 2004-2015):
  - CNEP_{t−1} coefficients across columns: 0.1246***; 0.1162**; 0.1447***; 0.1481***; 0.1025**; 0.1199***; 0.0561 (column-specific).
  - CNEP_{t−1} × SENS_{t−1}: 0.2637* (column 5); 0.2974*** (column 6).
  - CNEP_{t−1} × SENS_{t−1}, MIC: 0.0822 (std err 0.264).
  - CNEP_{t−1} × SENS_{t−1}, LIC: 0.2938** (std err 0.126).
  - SENS_{t−1}: -26.917**; -29.801**; -30.32*** (various columns).
  - Key controls with significance: LIQUIDITY_{t−1} ~ 0.3294***; SIZE_{t−1} ~ -22.05***; EQUITY_{t−1} ~ 0.3359**; DEPOSITS/ASSETS_{t−1} ~ 0.2418***.
  - Observations: 9,540 (all) and 2,848 (LICs); Number of banks: 1,642 (all) and 495 (LICs). R^2 ranges 0.198–0.488 depending on specification.
- Heterogeneous effects across bank characteristics (Table 3; LICs sample):
  - CNEP_{t−1} × SENS_{t−1}, high equity: 0.2935***.
  - CNEP_{t−1} × SENS_{t−1}, low equity: 0.2945*** (p-value for difference 0.104).
  - CNEP_{t−1} × SENS_{t−1}, high deposits: 0.2917***; low deposits: 1.0158*** (p-value 0.012).
  - CNEP_{t−1} × SENS_{t−1}, high NPLs: 0.2850***; low NPLs: 0.1459 (p-value 0.000).
  - Domestic vs foreign banks: both ~0.2952*** and 0.2974*** (p-value 0.901).
- Bank health and transmission channels (Table 4; LICs sample):
  - Dependent variables: Deposits/Assets, NPLs, ROA, Equity/Assets (columns 1–4).
  - SENS_{t−1} coefficients: -0.0598 (Deposits/Assets); 7.9000*** (NPLs); -0.4439 (ROA); 3.9417** (Equity/Assets).
  - CNEP_{t−1} × SENS_{t−1}: 0.0006* (Deposits/Assets); -0.0787*** (NPLs); 0.0045 (ROA); -0.0390** (Equity/Assets).
  - Observations: 2,842; 2,111; 2,838; 2,848 across columns; Number of banks: 494; 395; 495; 495.
- Asymmetry and cross-country heterogeneity (Table 5; LICs sample):
  - CNEP_{t−1} × SENS_{t−1} for High Exports/GDP: 0.3556**; Low Exports/GDP: 0.1971.
  - Asymmetry (∆CNEP_{t−1} positive vs negative): positive 0.1699; negative 0.7501***.
  - SENS_{t−1} varies: -38.3470** (High); -19.0514 (Low); -17.0016 (Positive); -76.297*** (Negative).
  - Observations: 841; 1,981; 599; 1,660 (columns respectively); Number of banks: 170; 354; 317; 394.
- Robustness exercises (Tables 6–8):
  - Alternative specialization measure (SPEC) interactions: CNEP_{t−1} × SPEC_{t−1} 0.8967* (column 1); asymmetry and heterogeneity results in Tables 6 and 7 show similar qualitative findings.
  - Expanded macro controls: CNEP_{t−1} × SENS_{t−1} remains significant in multiple specifications (e.g., 0.3112***; 0.2840***; 0.2943***; 0.4700***; 0.2600*** across columns in Table 7).
  - Alternative sample/specifications (Table 8): CTOT_{t−1} × SENS_{t−1} around 0.2962***; SENS_{t−1} negative and significant across many specifications (e.g., -30.2055***).
  - Robustness samples and definitions include: excluding lagged loan growth, excluding bank fixed effects, clustering at country level, restricting to countries with listed banks, excluding countries with average share of commodity exports > 5 percent, two-year rolling window for SENS, loan growth in real USD, replacing CNEP with CEP.
- Data and variable sources (Table A2):
  - Loan Growth: Bankscope.
  - Size, NPL, Deposits/Assets, Equity, Liquidity, ROA: Bankscope.
  - Price index: Gruss (2014).
  - Exposure (commodity exports to GDP): Gruss (2014) and authors’ calculations.
  - Growth: World Economic Outlook (IMF).
  - IR: International Financial Statistics (IMF) and World Development Indicators (World Bank).
  - Private Credit/GDP: International Financial Statistics (IMF) and World Development Indicators (World Bank).
  - Exchange Rate: International Financial Statistics (IMF).
  - Growth Forecast: World Economic Outlook (IMF), different vintages.
  - KA Open: Chinn and Ito (2006), updated.

### Online appendix and sample composition
- Table A1: Country coverage and banks
  - Total sample: 1,642 banks in 78 countries, of which 495 banks are in LICs; 249 banks are listed (kernel for bank-specific sensitivity measure).
  - Country-by-country counts provided (examples in list include AFGHANISTAN 19 banks, BANGLADESH 138 banks with 26 listed, NIGERIA 123 banks with 13 listed, etc.).
- Figure A1: Number of banks in the sample (map shows quantile distribution).
- Figure A2: Commodity exposure of countries in the sample (map shows quantile distribution of share of commodity exports to GDP; darkest shaded countries most exposed).

*References, figures, tables, and appendix materials as listed in the source document.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2017/wp17279.pdf_