## _wp15270 ## Source details **Canonical URL:** [_wp15270](https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15270.pdf) ## Other formats - [Markdown version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15270.pdf.md) - [Structured JSON version](/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15270.pdf.json) --- ### Introduction and research question - Assess the effectiveness of monetary policy transmission—particularly the bank lending channel—in Uganda, a bank-dependent developing economy that transitioned to an "inflation targeting lite (IT-lite)" framework. - Empirical challenges and approaches: - Endogeneity of monetary policy addressed by controlling for macro variables (real GDP growth rate, inflation), using time fixed effects, and exploiting unusually large short-term interest rate variation during the 2011 tightening as having an exogenous element. - Separating loan supply from loan demand addressed using granular loan-level data from the Compuscan Uganda CRB Ltd. credit register containing individual loan applications (acceptance/rejection) and loans granted to non-financial firms by the 15 largest Ugandan banks during 2010:Q3–2014:Q2 (these banks account for 95 percent of total banking sector assets); focus on local currency loans (Ugandan shillings), which represent almost 90 percent of loan applications. - Empirical strategy components: - Test bank lending channel predictions on extensive margin (probability of loan granting) and intensive margin (loan volume for granted loans). - Estimate pass-through of short-term interest rates to retail lending rates on new loans. - Test real economy transmission using district-level outcomes (nightlights intensity) and evaluate dependence of transmission on bank balance sheet characteristics (capital and liquidity). ### Institutional background and policy regime - Macroeconomic and structural context: - Flexible exchange rate regime; foreign currency assets share in 2013 was 31.6 percent. - Food share in CPI basket: 27 percent. - Financial market characteristics: illiquid financial markets, concentrated banking industry, large microfinance sector, significant agricultural employment, weak institutional/legal environment; banks favor government securities and hold excess reserves. - Monetary policy regime change: - Prior to 2011: de jure monetary aggregate targeting. - In July 2011: moved to IT-lite framework and introduced a policy rate; explicit inflation target set at 5 percent. ### The tightening (July–November 2011) and expansionary phases (2012 onward) - Drivers of policy actions: - Negative commodity price shock during 2010-2011 (food and fuel price rises), strong credit growth, weakening currency, and low real interest rates led to soaring inflation. - Coordinated tightening across four East African Community countries in second half of 2011. - Tightening phase (July–November 2011): - Policy rate raised by a cumulative 1,000 basis points (bps) over July-November 2011: 300 bps between July and September and an additional 700 bps between September and November. - Year-on-year credit growth peaked at more than 30 percent in early 2011; credit growth collapsed to negative levels by second half of 2012. - Tightening argued to be largely unanticipated by economic agents (prior dovish track record; pre-tightening communication emphasizing support for activity; as of October 2011 Bank of Uganda had not published a well-articulated intermediate inflation trajectory). - Expansionary phase (beginning January 2012): - Policy rate gradually reduced over next three quarters from 23 percent to 11 percent. - Non-performing loans rose from 1.8 percent in 2011:Q2 to 4.9 percent of total loans in 2013:Q1. - Average marginal lending rate on local currency loans increased during tightening from close to 15 percent to almost 25 percent and subsequently returned to about 20 percent. - Potential confounders: - Reserve requirement reduced in March 2011 by 1.5 percentage points to 8 percent of total deposits (would dampen estimated effects of interest rate hikes). - No other macroprudential policy changes during 2010-2014. - Foreign banks fund operations primarily with local deposits (limited foreign policy transmission). ### Data, coverage, and key banking system statistics - Credit register and microdata: - Uganda’s credit register set up in 2008, stabilized by mid-2010; collects monthly data on loan applications and granted loans from all commercial banks, microfinance deposit-taking institutions, and other credit institutions. - Analysis uses credit register data for the largest 15 banks (95 percent of total banking assets). - CRB records date, interest rate, maturity, currency, collateral description and resale value, and borrower performance for each application/granted loan. - Sample (local currency loans, cleaned): 26,363 loan applications and 25,948 granted loans during 2010:Q3–2014:Q2. - Borrower coverage: applications from 8,679 firms; loans granted to 8,718 firms across 66 districts and 9 industries. - Relationship concentration: 83 percent of firms apply to only one bank; 13 percent to two banks. In granted loans dataset, 87 percent borrow from one bank and 10 percent from two banks. - Banking system aggregates and balance-sheet averages (sample period context): - Bank credit to the private sector rose to 15 percent of GDP during 2001-2013. - Total banking sector assets represent 19 percent of GDP. - Largest 5 banks account for 73 percent of total banking system assets. - Average Tier 1 capital ratios: 20 percent. - Average total regulatory capital ratios: 24 percent. - Typical bank funding: 66 percent deposits; 30 percent shareholders’ equity; 11 percent market-based funding; 6 percent other. - Typical asset composition: 52 percent loans; 21 percent securities (mostly government bonds); 10 percent reserves at the central bank; 4 percent cash. - Average liquid-to-total assets ratio: 27 percent. - Key descriptive statistics (2010:Q3–2014:Q2 unless noted; preserve values as reported): - IR_t (7-day interbank rate): Obs. 161; Mean 13.65; S.D. 6.39; Min 3.36; p25 10.96; p50 11.61; p75 17.46; Max 26.23. - ∆IR_t: Obs. 160; Mean 0.51; S.D. 3.64; Min -4.31; p25 -0.95; p50 -0.16; p75 1.27; Max 8.13. - Policy rate: Obs. 121; Mean 4.85; S.D. 4.37; Min 1.33; p25 1.58; p50 2.25; p75 11.33; Max 22. - LIQUIDITY: Obs. 772; Mean 42.13; S.D. 26.46; Min 0.00; p25 29.21; p50 40.41; p75 57.19; Max 129.71. - CAPITAL: Obs. 772; Mean 24.88; S.D. 11.23; Min 11.67; p25 17.04; p50 22.06; p75 28.67; Max 78.00. - LOANGRANTED: Obs. 190,630; Mean 0.84; S.D. 0.37; Min 0; p25 1; p50 1; p75 1; Max 1. - LOANAMOUNT (all loans): Obs. 252,680; Mean 0.20; S.D. 0.87; Min 0.01; p25 0.03; p50 0.10; p75 0.29; Max 9.89. - LENDINGRATE (all loans): Obs. 189,442; Mean 25.27; S.D. 7.57; Min 12; p25 21.00; p50 24.00; p75 29.00; Max 255. ### Empirical identification and estimation approach - Extensive margin (loan acceptance): - Dependent variable: LOANGRANTED_ibt = 1 if loan application by firm i to bank b in quarter t was accepted. - Sample excludes pending (36.2 percent) and cancelled (0.2 percent); accepted 53.4 percent and rejected 10.2 percent of applications. - Baseline: linear probability model with firm (ηi) and bank (ψb) fixed effects; interactions of ∆IR_t with lagged bank liquidity and capital to test balance sheet channel. - Controls: ΔIR_t (7-day interbank rate changes), ΔGDP_t, ΔCPI_t; OLS with standard errors clustered at district level. - Intensive margin and pass-through: - Dependent variables: ln(LOANAMOUNT_jbt) for loan volumes aggregated at district-industry j by bank b; average lending rate at district-industry level for pass-through. - Main regressor: cumulative change in 7-day interbank rate over horizons ΔIR_t,t−z with z=1,2 quarters. - Fixed effects: bank and district-industry (φj); interactions with bank liquidity and capital to identify balance sheet channel. - Real effects via night lights: - Dependent variable: LIGHTS_dt = night-time luminosity in district d in quarter t (0-63 scale). - Monetary policy enters as cumulative change over past z quarters IR_t,t−z with lags up to 4 quarters. - District-level bank balance-sheet variables computed as market-share-weighted averages of bank capital and liquidity. ### Key empirical findings — extensive margin (probability of loan granting) - Baseline effect: - An increase in interest rates by half a standard deviation raises the probability of loan granting in the same quarter by 0.7-1.4 percentage points. (Reported as: half standard deviation increase in 7-day interbank rate over a quarter = 182 bps → reduces probability of loan granting by between 0.7 and 1.4 percentage points; coefficients in Table 4: -0.0076***, -0.0060***, -0.0041***.) - Asymmetry: - Higher interest rates associated with higher loan rejection rates during the monetary contraction (before 2012:Q1). - Monetary expansion does not have a statistically significant effect on loan granting. - Bank balance-sheet heterogeneity: - Differential effect of a rise in the interbank rate by half a standard deviation (182 bps) between a highly and a poorly capitalized bank (90th vs. 10th percentile) is 3.5 percentage points (90th and 10th percentiles of capital ratio distribution are 34 and 15 percent; differential computed as 182×(34−15)×0.0010=3.5). - Interpretation: higher capital dampens transmission (banks with high capital pass on increases in interest rates to the supply of credit less). - More liquid banks amplify the effect of interest rates (∆IR_t × LIQUIDITY_b,t−1: -0.0003***), consistent with a crowding-out effect where liquid banks shift into government securities when rates rise. ### Key empirical findings — intensive margin (loan volumes) and lending-rate pass-through - Intensive margin (quantity effects): - For ΔIR_t−2 (two-quarter cumulative change), half a standard deviation increase (322 bps) reduces bank credit by between 5.1 and 10.1 percent (reported ranges: 4.9-9.7 percent in summary and 5.1-10.1 percent in Table 6 context; Table 6 coefficients for ∆2IR_t: -0.0158***; -0.0314***; -0.0223***). - Negative effects stronger in contractionary period (p-value=0.017). - Bank balance-sheet heterogeneity: high-capital banks reduce the volume of new loans less; example: a half standard deviation increase in interest rates over two quarters (322 bps) leads high-capital banks (90th percentile) to reduce loan volume by 23.8 percent more than low-capital banks (10th percentile) in one quantified illustration. - Interest-rate pass-through to marginal lending rates: - A 100 bps increase in the 7-day interbank rate is associated with an increase in the lending rate of between 33 and 49 bps (columns 1-3); upper estimate (49 bps) not statistically different from 50 bps (pass-through of almost 50 percent). - Pass-through estimated using lending rates calculated from microdata on loan originations (marginal lending rate), not average lending rates. - Heterogeneity in pass-through: - High-capital banks charge 82 bps less than low-capital banks (90th vs. 10th percentile) for an increase of half a standard deviation in interest rates over one quarter (Table 7, column 5). - More liquid banks pass through the increase in the 7-day interbank rate less than banks with lower liquidity ratios in some specifications; interactions show nuanced patterns (∆IR_t × LIQUIDITY positive in some pass-through specs, LIQUIDITY level negative in lending-rate regressions). ### Key empirical findings — real effects (nightlights) - Nightlights evidence and timing: - Real activity proxied by night-time luminosity is less affected by monetary policy tightening in districts where banks are better capitalized and less liquid. - Column 7 example: for an increase of half a standard deviation in the interest rate over three quarters (about 400 bps), night lights in high-capital districts (90th percentile) are higher by 0.142 units than in low-capital districts (10th percentile). The 0.142 units represent almost one tenth of a standard deviation of the lights distribution. - Effects discernible after 3-4 quarters; results more precise when Kampala is excluded. - District-level interaction patterns (selected significant interactions from Table 9): - CAPITAL_d,t−1 × IR_t−z positive and significant for z=3,4 (including Kampala and stronger excluding Kampala). - LIQUID_d,t−1 × IR_t−z negative and significant at longer horizons (z=2,3,4) when excluding Kampala. - Interpretation: district-level bank capital cushions local real activity from tightening; higher district liquidity amplifies contractionary real effects. ### Robustness and sensitivity checks - Baseline results replicated using alternative interest rates: the policy rate introduced in July 2011, the 91-day T-bill rate, and the discount rate for emergency Bank of Uganda lending. - Alternative clustering: industry, quarter, and industry-district clustering. - Controlling for changes in the nominal exchange rate. - Results reported as robust across sensitivity tests (see regression tables in Online Appendix referenced in source). ### Contributions, novelty, and comparative context - First study to use loan-level credit register data to study real effects of the bank lending channel in a developing country. - Unique dataset features leveraged: - Loan application decisions (accept/reject) allow separation of supply and demand effects. - Loan-level pricing enables precise estimation of marginal lending rate pass-through. - Ability to control for lender and borrower heterogeneity via bank, borrower, and borrower×time fixed effects. - Comparative context: - Coefficient magnitudes suggest weaker monetary policy transmission in Uganda than in advanced economies; Jiménez et al. (2012) find a 100 bps increase in short-term interest rate reduces probability of loan granting to Spanish firms by 1.4 percentage points (larger than baseline estimates reported here). - Cross-country pass-through context cited: long-run correlation between policy and lending rates 0.29 for developing countries versus 0.35 for advanced economies; regional pass-through patterns reported (Advanced/G20 ≈ full; South America/East Asia ≈ 50 percent; E&E/MENA/Sub-Saharan Africa ≈ 40 percent). ### Policy-relevant conclusions and implications (reported) - The bank lending channel operates in Uganda but is weaker (by about 50 percent) than in advanced economies. - A strong bank balance-sheet channel is present: - High-capital banks transmit interest rate changes less to the probability of loan granting, to new loan volumes, and to retail lending rates. - High-liquidity banks pass through interest rate changes to the quantity and price of credit more than low-liquidity banks—implying a crowding-out effect of high liquidity on loans. - At district level, for a given increase in interest rates, nighttime luminosity is higher in high-capital and low-liquidity districts than elsewhere. - Microdata from credit registers provide a pathway to identify loan supply and real effects of the domestic bank lending channel beyond aggregate time-series identification challenges. *Source: IMF Working Paper (chapter excerpts and data from _wp15270).* ### Appendix available for download on: https://sites.google.com/site/presbitero/homepage/wp. This paper is part of a resear ### _wp15270 - Appendix available for download on: https://sites.google.com/site/presbitero/homepage/wp. This paper is part of a research ### Introduction - Research question: assess the effectiveness of monetary policy transmission—particularly the bank lending channel—in Uganda, a bank-dependent, developing economy that transitioned to an "inflation targeting lite (IT-lite)" framework. - Key empirical challenges: - Endogeneity of monetary policy: addressed by controlling for macro variables (real GDP growth rate, inflation) and using time fixed effects; exploit unusually large short-term interest rate variation during tightening as having an exogenous element. - Separating loan supply from loan demand: addressed using granular loan-level data from the Compuscan Uganda CRB Ltd. credit register containing individual loan applications (acceptance/rejection) and loans granted to non-financial firms by the 15 largest Ugandan banks during 2010:Q3–2014:Q2 (these banks account for 95 percent of total banking sector assets); focus on local currency loans (Ugandan shillings), which represent almost 90 percent of loan applications. - Empirical strategy components: - Test bank lending channel predictions on extensive margin (probability of loan granting) and intensive margin (loan volume for granted loans). - Estimate pass-through of short-term interest rates to retail lending rates on new loans. - Test real economy transmission using district-level outcomes (nightlights intensity), and evaluate dependence of transmission on bank balance sheet characteristics (capital and liquidity). ### Institutional background - Macroeconomic and structural context: - Uganda: flexible exchange rate regime, moderate dollarization (foreign currency assets share in 2013 was 31.6 percent). - Large share of food in CPI basket: 27 percent; inflation driven by external food and fuel shocks and domestic supply shocks (weather-related). - Banking system features relevant to transmission: illiquid financial markets, concentrated banking industry, large microfinance sector, significant agricultural employment, weak institutional/legal environment leading banks to favor government securities and hold excess reserves. - Monetary policy regime change: - Prior to 2011: de jure monetary aggregate targeting. - In July 2011: moved to IT-lite framework and introduced a policy rate; explicit inflation target set at 5 percent. ### The tightening and expansionary phases (2010–2014) - Drivers of policy actions: - Negative commodity price shock during 2010-2011 (food and fuel price rises) combined with strong credit growth, weakening currency, and low real interest rates led to soaring inflation. - Coordinated tightening across four East African Community countries in second half of 2011. - Tightening phase (July–November 2011): - Policy rate raised by a cumulative 1,000 basis points (bps) over July-November 2011: 300 bps between July and September and an additional 700 bps between September and November. - Year-on-year credit growth peaked at more than 30 percent in early 2011; credit growth collapsed to negative levels by second half of 2012. - Arguments that the magnitude and timing of tightening were largely unanticipated by economic agents: prior dovish track record, prior inaction during 2007–2008 shock, pre-tightening communication emphasizing support for activity, and transition to a new policy framework with limited guidance (as of October 2011 Bank of Uganda had not published a well-articulated intermediate inflation trajectory). - Expansionary phase (beginning January 2012): - Policy rate gradually reduced over next three quarters from 23 percent to 11 percent. - Loan quality deterioration: non-performing loans rose from 1.8 percent in 2011:Q2 to 4.9 percent of total loans in 2013:Q1. - Lending rate dynamics: average marginal lending rate on local currency loans increased during tightening from close to 15 percent to almost 25 percent and subsequently returned to about 20 percent. - Potential confounders and assessment: - Reserve requirements: Bank of Uganda does not use cash reserve requirements actively for monetary policy; reserve requirement reduced in March 2011 by 1.5 percentage points to 8 percent of total deposits (this would dampen estimated effects of interest rate hikes). - No other macroprudential policy changes during 2010-2014. - Foreign monetary policy transmission limited because foreign banks in Uganda fund operations primarily with local deposits. ### Data and empirical strategy (overview) - Data: - Credit register from Compuscan Uganda CRB Ltd.: individual loan applications (accept/reject) and loan contracts for the 15 largest Ugandan banks, 2010:Q3–2014:Q2. - Coverage: banks account for 95 percent of banking sector assets; almost 90 percent of loan applications are in local currency (Ugandan shillings). - Observables include loan pricing for each loan (rare feature in credit registers), enabling marginal lending rate pass-through estimation. - Identification strategy: - Extensive margin: estimate effect of short-term interest rate changes on probability of loan granting, controlling for borrower, bank, borrower×time fixed effects, and macro controls to net out demand influences. - Intensive margin and pass-through: estimate volume effects for granted loans and pass-through to retail lending rates using loan-level pricing information. - Real effects: use district-level nightlights intensity over time and exploit variation in local banking sector conditions to identify heterogeneous transmission. - Bank balance sheet channel: identify differential effects by banks’ capital and liquidity characteristics; conjectures include: - Higher-capital banks transmit policy changes less than lower-capital banks. - Higher-liquidity banks may amplify the lending channel in Uganda because liquid banks prefer government securities and may pull back from lending when rates rise. ### Key empirical findings (quantitative results preserved exactly as reported) - Extensive margin (probability of loan granting): - An increase in interest rates by half a standard deviation raises the probability of loan granting in the same quarter by 0.7-1.4 percentage points. - Intensive margin (loan volumes): - An increase in interest rates by half a standard deviation over two quarters reduces loan supply by 4.9-9.7 percent. - Interest rate pass-through: - About half of the variation in market interest rates translates into changes in retail lending rates on new loans. - Bank balance sheet (heterogeneity in transmission): - High-capital banks transmit changes in the monetary policy stance significantly less than low-capital banks. - Highly liquid banks pass through policy changes more than other banks. - Real effects: - Real activity, captured by nightlights intensity across districts and over time, is less affected by a monetary policy tightening in districts where banks are better capitalized and less liquid. - Comparative context: - Coefficient magnitudes suggest weaker monetary policy transmission in Uganda (a developing economy) than in advanced economies; for reference, Jiménez et al. (2012) find a 100 bps increase in short-term interest rate reduces probability of loan granting to Spanish firms by 1.4 percentage points (larger than baseline estimates reported here). ### Contributions and novelty - First study to use loan-level credit register data to study real effects of the bank lending channel in a developing country. - Unique features of the dataset leveraged: - Loan application decisions (accept/reject) allowing separation of supply and demand effects. - Loan-level pricing enabling precise estimation of marginal lending rate pass-through. - Ability to control for lender and borrower heterogeneity via bank, borrower, and borrower×time fixed effects. - Extends period of analysis through mid-2014 to capture both dramatic tightening and subsequent expansionary phases, enabling tests for asymmetric transmission. *IMF Working Paper (appendix available for download on: https://sites.google.com/site/presbitero/homepage/wp).* ### 2.2 The banking system in Uganda ### 2.2 The banking system in Uganda ### Financial deepening and market structure - Bank credit to the private sector more than doubled to 15 percent (of GDP) during 2001-2013 (International Monetary Fund, 2015). - Financial depth remains low by international standards and there is a large informal financial sector: - 74 percent of the Ugandan adult population used the financial services of an informal lender in 2013. - 15 percent used the services of both a formal and informal lender (Finscope, 2013). - The banking system comprises 25 (mostly foreign- and privately-owned) banks during the period of analysis. - Total banking sector assets represent 19 percent of GDP. - The largest 5 banks account for 73 percent of total banking system assets (GFDD, 2011). ### Bank balance sheets, funding, and liquidity - Capitalization: - Average Tier 1 capital ratios of 20 percent. - Average total regulatory capital ratios of 24 percent. - Funding sources (typical bank): - 66 percent in deposits. - 30 percent shareholders’ equity. - 11 percent market-based funding (from banks in Uganda and abroad). - 6 percent other sources. - Asset composition (average bank): - 52 percent loans. - 21 percent securities (mostly government bonds). - 10 percent reserves at the central bank. - 4 percent cash. - Liquidity: - Average liquid-to-total assets ratio of 27 percent. - The analysis exploits heterogeneity in bank balance sheet characteristics—capital and liquidity—and computes district-level bank capital and liquidity weighted by banks’ market shares in each district. ### The Ugandan credit register and microdata coverage - Uganda’s credit register was set up in 2008 and stabilized by mid-2010. - The credit register collects monthly data on loan applications and granted loans from all commercial banks, microfinance deposit-taking institutions, and other credit institutions. - Analysis uses credit register data for the largest 15 banks, which account for 95 percent of total banking assets. - The CRB collects comprehensive information on loan applications and originations with no minimum loan size; for each application/granted loan it records date, interest rate, maturity, currency, collateral description and resale value, and borrower performance. - Banks submit separate datasets for applications and for loans; there is limited overlap between them. - Sample (local currency loans, cleaned): - 26,363 loan applications during 2010:Q3–2014:Q2. - 25,948 granted loans during 2010:Q3–2014:Q2. - Borrower coverage: - Applications from 8,679 firms. - Loans granted to 8,718 firms. - Location across 66 districts and sector across 9 industries. - No firm balance sheet information available. - Relationship concentration: - 83 percent of firms apply for a loan to only one bank (could apply multiple times). - 13 percent apply to two banks; remainder to 3+ banks. - In granted loans dataset, 87 percent of firms borrow from one bank and 10 percent from two banks. ### Data merges and real-activity measurement - Micro-data merged with bank balance sheet variables and macroeconomic time series (interest rates, GDP growth, inflation) on monthly and quarterly basis. - Satellite night lights data (monthly, aggregated to district-quarter) used to examine real effects. - Variable definitions, sources, and descriptive statistics reported in Tables 2 and 3 (in source). ### Empirical strategy: overview - Main identification challenge: disentangle credit demand from credit supply effects. - Key bank-level measures: - LIQUIDITYb,t−1 = ratio of liquid assets to total deposits (lagged one quarter). - CAPITALb,t−1 = ratio of total regulatory capital to risk-weighted assets (lagged one quarter). - Macro controls include changes in the 7-day interbank rate (ΔIRt), real GDP growth (ΔGDPt), and inflation (ΔCPIt). - Standard estimation approach: OLS with clustering of standard errors at the district level. ### Empirical strategy: extensive margin (loan application acceptance) - Dependent variable: LOANGRANTEDibt = 1 if loan application by firm i to bank b in quarter t was accepted. - Sample: only applications that were either accepted (53.4 percent) or rejected (10.2 percent); pending (36.2 percent) and cancelled (0.2 percent) excluded from the main extensive-margin analysis. - Average acceptance rate during 2010-2014 is 84.6 percent. - Baseline specification (linear probability model) includes firm (ηi) and bank (ψb) fixed effects and interacts ΔIRt with bank liquidity and capital to test for a bank balance sheet channel. - Expectations from theory: - Better capitalized banks weaken the bank lending channel (higher capital → smaller pass-through). - More liquid banks may strengthen the bank lending channel via a “crowding-out effect” (preference for government securities → larger cut in lending when rates rise). ### Empirical strategy: intensive margin and interest rate pass-through - Dependent variable for intensive margin: ln(LOANAMOUNTjbt) — volume of credit granted to firms in district-industry j by bank b in quarter t (district-industry aggregation yields 290 pairs). - Main regressor: cumulative change in 7-day interbank rate over different horizons ΔIRt,t−z with z=1,2 quarters. - Fixed effects: bank fixed effects (ψb) and district-industry fixed effects (φj) to control for common demand shocks across firms in each district-industry cluster. - Also interact ΔIRt,t−z with bank liquidity and capital to identify the bank balance sheet channel. - Pass-through to lending rates examined by replacing dependent variable with average lending rate at the district-industry level. - Regressions estimated with OLS and standard errors clustered at the district level. ### Empirical strategy: real effects via night lights - Dependent variable: LIGHTSdt = night-time luminosity in district d in quarter t (district-quarter aggregation). - Reduced form specification allows monetary policy to affect real activity with lags up to 4 quarters (z=1,2,3,4). - Interest rate enters as cumulative change over past z quarters: IRt,t−z. - District-level bank balance sheet variables computed as averages of bank capital and liquidity weighted by market power (share of local currency loans extended to firms in each district). - Expectations: α2 < 0 (stronger negative IR × liquidity effect) and α3 > 0 (capital dampens transmission). ### Results — extensive margin - Impact of ΔIR on acceptance probability: - Half a standard deviation increase in the 7-day interbank rate over a quarter (182 bps) reduces the probability of loan granting by between 0.7 and 1.4 percentage points (columns 1-3, Table 4). - Estimates are lower than those for advanced economies (example comparison with Spain). - Asymmetry across contraction/expansion: - Higher interest rates associated with higher loan rejection rates during the monetary contraction (before 2012:Q1). - Monetary expansion does not have a statistically significant effect on loan granting. - Possible dampening factors during the expansionary period: high policy uncertainty surrounding the tightening episode and large non-performing bank loans (Figure A3). - Bank balance sheet channel (interaction results): - Differential effect of a rise in the interbank rate by half a standard deviation (182 bps) between a highly and a poorly capitalized bank (90th vs. 10th percentile) is 3.5 percentage points. - Note: 90th and 10th percentiles of capital ratio distribution are 34 and 15 percent; differential computed as 182×(34−15)×0.0010=3.5. - Interpretation: banks with high levels of capital pass on increases in interest rates to the supply of credit less than other banks (capital dampens transmission). - More liquid banks amplify the effect of interest rates (α2 < 0) — consistent with a crowding-out effect where higher rates raise demand for safe government assets by liquid banks, reducing private lending. ### Results — intensive margin - Quantity effects of ΔIR: - The 7-day interbank rate affects loan quantities both instantaneously and with lags; estimates are more precise with deeper lags (z=2). - For ΔIRt−2 (two-quarter cumulative change), half a standard deviation increase (322 bps) reduces bank credit by between 5.1 and 10.1 percent (columns 1-3, Table 6). - Subperiods: - Allowing distinct effects during contractionary and expansionary periods shows negative and statistically significant coefficients for both, but larger magnitude in the contractionary period (p-value=0.017). - Bank balance sheet channel (intensive margin with industry-district×quarter fixed effects): - Identification with comprehensive industry-district×quarter fixed effects shows: - Higher capital dampens the transmission of interest rate changes to credit supply. - Higher liquidity enhances the transmission. - Quantitative example: a half standard deviation increase in interest rates over two quarters (322 bps) leads high-capital banks (90th percentile) to reduce the volume of new loans by 23.8 percent more than low-capital banks (10th percentile). *Source: IMF working paper chapter "2.2 The banking system in Uganda" (content excerpt).* ### 5.3 Lending rates ### _wp15270 - 5.3 Lending rates ### Pass-through of policy rate to lending rates (main findings) - A 100 bps increase in the 7-day interbank rate is associated with an increase in the lending rate of between 33 and 49 bps (columns 1-3). - The upper estimate (49 bps) is not statistically different from 50 bps, indicating a pass-through of almost 50 percent. - The study uses lending rates calculated directly from micro data on loan originations (marginal lending rate), not average lending rates on existing loan claims. ### Heterogeneity by bank balance-sheet conditions - Differential effects of short-term interest rate changes on lending rates for high/low capital and liquidity banks are present in all specifications. - High-capital banks charge 82 bps less than low-capital banks (90th vs. 10th percentile of the capital distribution) for an increase of half a standard deviation in interest rates over one quarter (Table 7, column 5). - More liquid banks pass through the increase in the 7-day interbank rate less than banks with lower liquidity ratios. - At the district level, higher interest rates affect economic activity less in high-capital districts and more in high-liquidity districts (effects discernible after 3-4 quarters). ### Comparison with cross-country and literature evidence - The correlation between policy rates and lending rates in the long run is 0.29 for developing countries compared to 0.35 for advanced economies. - Saborowski and Weber (2013) find an average interest rate elasticity of 0.52 in a large sample over 2003-2008. - Regional/peer-group pass-through patterns reported: Advanced and G20 economies exhibit almost full pass-through; South American and East Asian countries have 50 percent; Eastern Europe, Middle East and North Africa, and sub-Saharan Africa have 40 percent. - The paper emphasizes that using loan-origination microdata captures the marginal lending rate, which adjusts more quickly than average lending rates from aggregate series. ### Real effects linked to lending-rate transmission (night lights evidence) - Night lights (satellite nighttime luminosity), averaged to quarterly frequency, are used as a district-level proxy for economic activity. - Regression results (panel of 66 districts) consistently support a bank balance sheet channel with a lag of about 3-4 quarters. - Column 7 coefficient: for an increase of half a standard deviation in the interest rate over three quarters (about 400 bps), night lights in high-capital districts (90th percentile) are higher by 0.142 units than in low-capital districts (10th percentile). - The 0.142 units represent almost one tenth of a standard deviation of the lights distribution. - Results are more precisely estimated when Kampala is excluded from the sample. ### Robustness checks - Baseline results replicated using three alternative interest rates: the policy rate introduced in July 2011, the 91-day T-bill rate, and the discount rate for emergency Bank of Uganda lending. - Alternative assumptions on the correlation structure of errors: clustering at the industry, quarter, and industry-district level. - Controlling for changes in the nominal exchange rate. - Results are robust to all sensitivity tests (see regression tables in the Online Appendix). ### Policy-relevant conclusions and implications - The bank lending channel operates in Uganda but is weaker (by about 50 percent) than in advanced economies. - A strong bank balance sheet channel is present: - High-capital banks transmit interest rate changes less to the probability of loan granting, to new loan volumes, and to retail lending rates. - High-liquidity banks pass through interest rate changes to the quantity and price of credit more than low-liquidity banks—implying a crowding out effect of high liquidity on loans. - At district level, for a given increase in interest rates, nighttime luminosity is higher in high-capital and low-liquidity districts than elsewhere. - Microdata from credit registers allow identification of loan supply and real effects of the domestic bank lending channel and provide a pathway beyond aggregate time-series identification challenges. *Source: _wp15270 - 5.3 Lending rates* ### Chapter II, Monetary and Exchange Rate Policies in Sub-Saharan Africa, International Mon- ### Chapter II, Monetary and Exchange Rate Policies in Sub-Saharan Africa ### Literature and data sources - Cited works include IMF Country Reports and academic articles (examples in source list: INTERNATIONAL MONETARY FUND (2011), (2012), (2015); JIMÉNEZ et al. (2012, 2014); KAPAN and MINOIU (2013); KASEKENDE and BROWNBRIDGE (2011); KHAN (2011); MISHRA and MONTIEL (2013); MISHRA et al. (2014); and others). - Data sources referenced in figures and tables: Bank of Uganda, Uganda Bureau of Statistics, International Financial Statistics (IFS), Compuscan Uganda CRB Ltd., National Geophysical Data Center at the National Oceanic and Atmospheric Administration (NOOA), and author’s calculations based on Bank of Uganda data. ### Macroeconomic and financial indicators (Uganda, sample periods noted) - Figures depict: - GDP growth (yoy), Core inflation (yoy), and 7-day interbank rate for 2009:Q1–2014:Q2. - Credit growth rate (%, y-o-y), policy rate, and 7-day interbank rate for 2010:M1–2014:M4. - Real credit growth (%, y-o-y) and real 7-day interbank rate for 2005:M1–2014:M9. - Rejection rate, weighted lending rate, and 7-day interbank rate for 2010:Q3–2014:Q2. - Bank capital (ratio of total regulatory capital to total risk-weighted assets) and bank liquidity (ratio of liquid assets to total deposits) for a sample of 15 banks, 2010:Q3–2014:Q2. - District-level nightlights intensity (2010, 2011, 2012), color coded red (highest luminosity) to green (lowest luminosity). ### Industry composition of borrowers (Compuscan Uganda CRB Ltd. data) - Panel A: Distribution of loans by borrower industry (Loan applications; Granted loans) - Agriculture: 2,560 (9.71%); 13,340 (12.87%) - Mining and Quarrying: 354 (1.34%); 27,01? (1.04%) — (note: source shows "2701" with formatting) - Manufacturing: 1,344 (5.12%); 2,713 (10.46%) - Trade: 4,595 (17.43%); 4,779 (18.42%) - Transport and Communication: 3,142 (11.92%); 1,965 (7.57%) - Electricity and Water: 860 (3?); 1330.51? — (source formatting) - Building, Construction and Real Estate: 2,660 (10.09%); 3,443 (13.27%) - Community, Social and Other Services: 3,573 (13.55%); 1,859 (7.16%) - Central and Local Government: 1,029 (3.93%); 121 (1.2%) - Other: 7,020 (26.63%); 7,134 (27.49%) - Total: 26,363 (100%); 25,948 (100%) - Panel B: Distribution of borrowers by industry (Applicant firms; Borrowing firms) - Agriculture: 697 (8.03%); 917 (10.52%) - Mining and Quarrying: 101 (1.16%); 46 (0.53%) - Manufacturing: 354 (4.08%); 454 (5.21%) - Trade: 1,320 (15.21%); 1,177 (13.5%) - Transport and Communication: 1,157 (13.33%); 568 (6.52%) - Electricity and Water: 320 (0.37%); 350 (0.4%) - Building, Construction and Real Estate: 790 (9.10%); 645 (7.4%) - Community, Social and Other Services: 1,029 (11.86%); 608 (6.97%) - Central and Local Government: 385 (4.44%); 13 (1.56%) - Other: 2,814 (32.42%); 4,132 (47.4%) - Total: 8,679 (100%); 8,718 (100%) ### Variable definitions and descriptive statistics (sampling and measurement) - Key variable definitions (selected): - IR (7-day interbank rate): Interest rate on interbank market with maturity of 7 days (Bank of Uganda). - Policy rate: Bank of Uganda policy rate (central bank rate or CBR) introduced in July 2011 (IFS). - LIGHTS: Night time luminosity measured on a 0-63 scale (National Geophysical Data Center at NOAA). - LIQUIDITY: Liquid assets to total deposits (Bank of Uganda). - CAPITAL: Total regulatory capital divided by risk weighted assets (Bank of Uganda). - LOANGRANTED: Dummy = 1 for accepted loan applications, 0 if rejected (Compuscan Uganda CRB Ltd.). - LOANAMOUNT: Total loan amount for granted loans (Compuscan Uganda CRB Ltd.). - Descriptive statistics (period 2010:Q3–2014:Q2 unless noted): - IR_t (7-day interbank rate): Obs. 161; Mean 13.65; S.D. 6.39; Min 3.36; p25 10.96; p50 11.61; p75 17.46; Max 26.23. - ∆IR_t: Obs. 160; Mean 0.51; S.D. 3.64; Min -4.31; p25 -0.95; p50 -0.16; p75 1.27; Max 8.13. - Policy rate: Obs. 121; Mean 4.85; S.D. 4.37; Min 1.33; p25 1.58; p50 2.25; p75 11.33; Max 22. - 91-day T-bill rate: Obs. 161; Mean 11.06; S.D. 4.11; Min 4.68; p25 8.99; p50 9.45; p75 13.89; Max 19.49. - LIQUIDITY: Obs. 772; Mean 42.13; S.D. 26.46; Min 0.00; p25 29.21; p50 40.41; p75 57.19; Max 129.71. - CAPITAL: Obs. 772; Mean 24.88; S.D. 11.23; Min 11.67; p25 17.04; p50 22.06; p75 28.67; Max 78.00. - LOANGRANTED: Obs. 190,630; Mean 0.84; S.D. 0.37; Min 0; p25 1; p50 1; p75 1; Max 1. - LOANAMOUNT (all loans): Obs. 252,680; Mean 0.20; S.D. 0.87; Min 0.01; p25 0.03; p50 0.10; p75 0.29; Max 9.89. - LENDINGRATE (all loans): Obs. 189,442; Mean 25.27; S.D. 7.57; Min 12; p25 21.00; p50 24.00; p75 29.00; Max 255. ### Empirical findings — Extensive margin (loan application acceptance) - Dependent variable: LOANGRANTED_ibt (indicator for successful loan applications). - Key estimated coefficients (Table 4, specifications vary): - ∆IR_t: coefficients -0.0076***, -0.0060***, -0.0041*** (standard errors reported). - ∆GDP_t: 0.0097***, 0.0084***. - ∆CPI_t: -0.0051***, -0.0026**. - LIQUIDITY_b,t−1: 0.0036***, 0.0036***, 0.0041***. - CAPITAL_b,t−1: 0.0077***, 0.0070***, 0.0045***. - Interaction effects: - ∆IR_t × contraction: -0.0041**. - ∆IR_t × expansion: -0.0015 (not significant). - ∆GDP_t × contraction: 0.0103***. - ∆GDP_t × expansion: 0.0062*. - ∆CPI_t × contraction: -0.0039**. - ∆CPI_t × expansion: 0.0029*. - ∆IR_t × LIQUIDITY_b,t−1: -0.0003***. - ∆IR_t × CAPITAL_b,t−1: 0.0010***. - ∆GDP_t × CAPITAL_b,t−1: -0.0021***. - ∆CPI_t × LIQUIDITY_b,t−1: 0.0003***. - Observations across specifications: 13,826; 15,763 in one specification. - R-squared reported: 0.403, 0.405, 0.410, 0.411, 0.274 (depending on model). - Controls: firm fixed effects (in most specs), bank fixed effects, and industry-district×year-quarter FE in one specification. Interpretation drawn from reported coefficients (as presented): - An increase in ∆IR_t is associated with a statistically significant negative effect on the probability a loan application is granted (negative ∆IR_t coefficients). - Higher bank liquidity and capital (lagged) are associated with higher probabilities of loan approval (positive LIQUIDITY and CAPITAL coefficients). - Interaction terms indicate differential effects during contraction versus expansion periods and heterogeneous amplification/moderation by bank balance sheet variables. ### Empirical findings — Intensive margin (loan amounts), contemporaneous and lagged effects - Dependent variable: LOANAMOUNT_jbt (total loan amount to borrowers in district-industry j by bank b at time t). Contemporaneous effect (Table 5): - Selected coefficients: - ∆IR_t: -0.0072; -0.0209**; -0.0120 (varied significance across specs). - ∆GDP_t: 0.1152***; 0.1054***. - LIQUIDITY_b,t−1: 0.0157***; 0.0162***; 0.0246***. - CAPITAL_b,t−1: 0.0354***; 0.0345**; 0.0092. - Interaction effects: - ∆IR_t × contraction: -0.0387**. - ∆IR_t × expansion: 0.0080 (not significant). - ∆GDP_t × contraction: 0.1111***. - ∆I R_t × CAPITAL_b,t−1: 0.0051**. - ∆GDP_t × LIQUIDITY_b,t−1: -0.0052**. - Observations: 3,563 (varies); one spec with 2,652 observations. - R-squared: 0.417, 0.422, 0.428, 0.429, 0.527. - Fixed effects: firm FE, bank FE; industry-district×year-quarter FE in one spec. Lagged effect (Table 6): - Selected coefficients for second differences (∆2 variables): - ∆2IR_t: -0.0158***; -0.0314***; -0.0223***. - ∆2GDP_t: 0.0821***; 0.0684***. - ∆2CPI_t: 0.0180*; 0.0167**. - LIQUIDITY_b,t−2: 0.0174***; 0.0162***; 0.0015. - CAPITAL_b,t−2: 0.0473***; 0.0477***; 0.0268. - Interaction effects: - ∆2IR_t × contraction: -0.0598***. - ∆2IR_t × expansion: -0.0162***. - ∆2GDP_t × contraction: 0.0696**. - ∆2CPI_t × contraction: 0.0376**. - ∆2IR_t × LIQUIDITY_b,t−2: -0.0022***. - ∆2IR_t × CAPITAL_b,t−2: 0.0039***. - ∆2CPI_t × LIQUIDITY_b,t−2: 0.0027***. - Observations: 3,563 (varies); one spec with 2,652. - R-squared: 0.418, 0.423, 0.431, 0.433, 0.529. - Fixed effects: bank FE, sector-district FE; industry-district×year-quarter FE in one specification. Interpretation drawn from reported coefficients (as presented): - Contemporaneous and lagged increases in policy interest measures (∆IR, ∆2IR) are associated with negative and often significant effects on loan amounts, especially pronounced during contraction periods (interaction terms). - Bank liquidity and capital generally have positive associations with loan amounts; interactions show nuanced moderating or amplifying roles. ### Interest rate pass-through to lending rates (contemporaneous and lagged) Contemporaneous effect (Table 7): - Dependent variable: ∆LENDINGRATE_jbt - Key coefficients: - ∆IR_t: 0.3343***; 0.4877***; 0.4722***. - ∆GDP_t: -0.4024***; -0.3401***. - ∆CPI_t: -0.2059***; -0.2477***. - LIQUIDITY_b,t−1: -0.0845***; -0.0843***; -0.1057***. - Interaction effects: - ∆IR_t × contraction: 0.5037***. - ∆IR_t × expansion: 0.5250***. - ∆IR_t × LIQUIDITY_b,t−1: 0.0239***. - ∆IR_t × CAPITAL_b,t−1: -0.0343***. - ∆GDP_t × contraction: -0.3268***. - ∆CPI_t × contraction: -0.2793***. - Observations: 1,516 (varied); one spec with 1,066. - R-squared: 0.089, 0.103, 0.109, 0.110, 0.196. - Fixed effects: firm FE, bank FE; industry-district×year-quarter FE in one spec. Lagged effect (Table 8): - Dependent variable: ∆LENDINGRATE_jbt with second differences (∆2) - Key coefficients: - ∆2IR_t: 0.2267***; 0.2787***; 0.2718***. - ∆2CPI_t: -0.0873**; -0.0849*. - LIQUIDITY_b,t−2: -0.0341***; -0.0264*; -0.0821. - Interaction effects: - ∆2IR_t × contraction: 0.3116***. - ∆2IR_t × expansion: 0.2257***. - ∆2IR_t × CAPITAL_b,t−2: -0.0181***. - ∆2GDP_t × CAPITAL_b,t−2: 0.0157***. - ∆2CPI_t × contraction: -0.2074***. - Observations: 1,516 (varied); one spec with 1,066. - R-squared: 0.100, 0.101, 0.102, 0.107, 0.188. - Fixed effects: bank FE, sector-district FE; industry-district×year-quarter FE in one spec. Interpretation drawn from reported coefficients (as presented): - There is positive and significant pass-through from changes in interbank rates to changes in lending rates (coefficients for ∆IR_t and ∆2IR_t positive and significant). - Bank liquidity is associated with negative levels in lending rate changes (liquidity dampens lending rate changes), but interaction terms indicate liquidity can increase pass-through (∆IR_t × LIQUIDITY positive). - Capital interacts negatively with pass-through in some lagged specifications (e.g., ∆2IR_t × CAPITAL_b,t−2: -0.0181***). ### Real effects of monetary policy using nightlights (Table 9) - Dependent variable: average night-time luminosity in district d in quarter t. - Regressions conducted for horizons z = 1, 2, 3, 4; results presented both including Kampala and excluding Kampala. - Selected coefficients (examples): - CAPITAL_d,t−1 × IR_t−z: - Including Kampala: z=3: 0.0024**; z=4: 0.0025**. - Excluding Kampala: z=3: 0.0028***; z=4: 0.0028***. - CAPITAL_d,t−1 × ∆GDP_t−z: - Including Kampala: z=2: -0.0112**; z=4: 0.0099***. - Excluding Kampala: z=2: -0.0088**; z=4: 0.0103***. - CAPITAL_d,t−1 × ∆CPI_t−z: - Including Kampala: z=1: 0.0030; z=2: -0.0092**. - Excluding Kampala: z=1: 0.0041***; z=2: -0.0087**. - LIQUID_d,t−1 × IR_t−z: - Including Kampala: z=4: -0.0016**. - Excluding Kampala: z=2: -0.0009**; z=3: -0.0017***; z=4: -0.0020***. - LIQUID_d,t−1 × ∆CPI_t−z: - Including Kampala: z=1: -0.0025*. - Excluding Kampala: z=2: 0.0043*. - Observations by specification range: 1,056–1,254 (including Kampala) and 1,040–1,235 (excluding Kampala). - R-squared range: 0.275–0.447 (including Kampala) and 0.300–0.447 (excluding Kampala). - Number of districts: 66 (including Kampala) and 65 (excluding Kampala). - Notes: CAPITAL and LIQUIDITY computed as weighted averages of bank capital and liquidity at district level using banks’ market power as weights. Monthly night lights series for Uganda discontinued in 2014; regression period is 2010Q1:2013Q4. Interpretation drawn from reported coefficients (as presented): - Interactions of district-level bank balance sheet measures (CAPITAL and LIQUID) with monetary variables show heterogeneous effects on night-time luminosity, with some coefficients positive and significant and others negative and significant depending on horizon and inclusion/exclusion of Kampala. - The results suggest bank capital and liquidity condition the transmission of monetary conditions to local real activity proxied by nightlights. ### Summary of empirical patterns (as reported) - In the extensive margin regressions, increases in short-term interbank rates (∆IR_t) are associated with decreases in the probability that loan applications are granted (negative and significant ∆IR_t coefficients). - Bank-level liquidity and capital are positively associated with loan approval probabilities and loan amounts (positive LIQUIDITY and CAPITAL coefficients, many significant). - Monetary tightening is associated with reductions in loan amounts, with stronger negative effects during contraction periods (interaction terms like ∆IR_t × contraction and ∆2IR_t × contraction are negative and often significant). - Interest rate pass-through to lending rates is positive and significant (∆IR_t and ∆2IR_t coefficients), but moderated by bank liquidity and capital (LIQUIDITY negative, interaction terms with LIQUIDITY/CAPITAL significant). - District-level analyses using nightlights indicate that bank capital and liquidity condition the local real effects of monetary policy, with heterogeneous signs across horizons and with/without Kampala. *International Monetary Fund. Chapter II, Monetary and Exchange Rate Policies in Sub-Saharan Africa (extracted content).* --- _Source: https://www.imf.org/-/media/websites/imf/imported-full-text-pdf/external/pubs/ft/wp/2015/_wp15270.pdf_