21. Banking

21.1.1 Bank's external financing friction and its lending: Paravisini (2008). Uses Argentina data to study the sensitivity of bank lending to external financing. The intensity (waves) of MYPES (a 1993–1999 SME credit program funded by the Inter-American Development Bank, giving 3-year financing at the average deposit rate, requiring banks to lend 1 USD per 0.75 USD of program funds) is exploited. The basic regression:

$$\ln L_{it}-\ln L_{i,t-1}=\alpha_i+\alpha_t+\beta_0(\ln F_{it}-\ln F_{i,t-1})+\boldsymbol\gamma'\mathbf X_{it}+\varepsilon_{it} \tag{21.1}$$

\(F_{it}\) all funding net of regulatory reserves, \(L_{it}\) loans, \(\alpha_i/\alpha_t\) bank/time FE. Endogeneity: \(\ln F\) correlates with economic conditions (better conditions → more deposits and loans) → use 2SLS. Instrument: actual MYPES \(C_{iw}\) is endogenous (bank self-selection), so construct an exogenous expected MYPES — first estimate by probit the participation probability \(\hat p_{iw}\) of all banks (including non-participants) in wave \(w\), then

$$\hat C_{iw}=A_w\left[\frac12\frac{\hat p_{iw}Zregion_i}{\sum_{j=1}^{n_w}\hat p_{iw}Zregion_j}+\frac12\frac{\hat p_{iw}Zsize_i}{\sum_{j=1}^{n_w}\hat p_{iw}Zsize_j}\right]$$

(\(A_w\) total wave funding; the administrative formula gives poorer provinces \(Zregion\) / smaller-loan banks \(Zsize\) higher fractions). 2SLS: first stage \(\ln F_{it}-\ln F_{i,t-1}=\phi_i+\phi_t+\delta_0(\ln\hat C_{it}-\ln\hat C_{i,t-1})+\boldsymbol\Gamma'\mathbf X_{it}+e_{it}\) (21.2), second stage plugging fitted values into (21.1) for the true \(\beta_0\) (21.3). Results: the first stage implies bank resources rise 1 USD per 1 USD of expected MYPES (no substitution, MYPES infra-marginal, banks still need capital); the second stage implies banks lend 0.66 USD per 1 USD (relabeling previous debt as program debt).

21.1.2 Khwaja-Mian (2008). Uses Pakistan's 1998 nuclear-test data: after the tests the government froze all dollar deposits, allowing withdrawal only in rupees at an unfavorable rate; USD depositors ran banks → banks relying more on USD deposits had worse liquidity problems (Figures 21.1, 21.2). Two questions: 1. Can banks easily raise financing? 2. If not, can firms offset the lending reduction? Bank lending channel (answers 1): \(\Delta L_{ij}=\beta_0+\beta_1\Delta D_i+\eta_j+\varepsilon_{ij}\) (21.4) (\(\Delta L_{ij}\) = bank \(i\)'s loan change to firm \(j\) = intensive margin, \(\Delta D_i\) demand-deposit change, \(\eta_j\) firm FE). OLS, with \(\text{Cov}(\Delta D_i,\eta_j)>0\) (worse economy → less deposit and less loan demand), gives \(\beta_0^{OLS}>\beta_1\), so we need firm FE (unbiased, but only multi-bank firms remain = sub-sample). Firm borrowing channel (answers 2): aggregating across banks gives \(\Delta\bar D_j\), \(\Delta Y_j=\beta_0^{Firm}+\beta_1^{Firm}\Delta\bar D_j+\eta_j\) (21.5); \(\text{Cov}(\Delta\bar D_j,\eta_j)<0\) (in emerging markets big banks have more USD exposure, more profitable firms borrow from big banks → high \(\Delta\bar D_j\) associates with profitable firms less likely to be cut), so OLS's \(\beta_1^{Firm,OLS}\) is a lower bound on the truth. Results: intensive margin (Figure 21.3) lower liquidity → smaller loans; extensive margin (Figure 21.4) → higher exit, lower entry probability; firm borrowing channel (Figure 21.5) large firms suffer less.

21.1.3 Chodorow-Reich (2014). Uses firm-level confidential employment data (BLS) + Dealscan bank-firm data to study lending frictions' effect on employment. Relies on two assumptions: bank-firm relationships are sticky (Figure 21.6: a previous lead bank has a 67% probability of being the next lead); the 2008 crisis originated outside the corporate loan sector. Construct firm \(i\)'s credit-availability change

$$\Delta\hat L_{i,s}=\sum_{b\in s}\alpha_{b,i,last}\Delta L_{-i,b} \tag{21.6}$$

(\(\alpha_{b,i,last}\) bank \(b\)'s share of \(i\)'s last pre-crisis loan syndicate, \(\Delta L_{-i,b}=\frac{\sum_{j\ne i}\alpha_{b,j,crisis}L_{b,j,crisis}}{0.5\sum_{j\ne i}\alpha_{b,j,normal}L_{b,j,normal}}\), the 0.5 because the normal period (18 months) is twice the crisis period (9 months)). Four instrument groups remove \(\Delta L_b\)'s endogeneity: 1. Lehman cosyndication exposure; 2. ABX (toxic MBS) exposure; 3. bank statement items; 4. their combination (Figure 21.7). Results: credit supply↑ → higher loan probability (Figure 21.8), lower spread (Figure 21.9), higher employment growth (Figure 21.10); especially strong for SMEs without bond access (Figure 21.11). The aggregate analysis is criticized (Dealscan firms are not representative of the whole sample).

21.2.1 Jimenez et al. (2012) the quantity effect of monetary policy on lending: Spain (small influence on ECB policy → nearly exogenous), using credit registry CIR and firm FE to compare lending decisions of different banks to the same firm. Main regression

$$\text{Loan Granted}_{ibt}=\beta_I\Delta IR_t+\beta_G\Delta GDP_t+\beta_{IC}(\Delta IR_t\times\Delta CAP_{b,t-1})+\beta_{IL}(\Delta IR_t\times\Delta LIQ_{b,t-1})+\beta_{GC}(\Delta GDP_t\times\Delta CAP)+\beta_{GL}(\Delta GDP_t\times\Delta LIQ)+\cdots \tag{21.7}$$

(\(\Delta IR\) short-rate change, \(\Delta CAP/\Delta LIQ\) bank capital/liquidity ratios). Hypothesis 1: \(\beta_I<0\) (tightening → tighter lending), \(\beta_G>0\). Hypothesis 2: \(\beta_{IC},\beta_{IL}>0\) (high-capital/liquidity banks dampen \(\beta_I<0\)), \(\beta_{GC},\beta_{GL}<0\). Results (Figure 21.12): both hypotheses supported (only \(\beta_{GL}<0\) not significant).

21.2.2 Jimenez et al. (2014) the risk-appetite effect of monetary policy (same data, focusing on loan risk):

$$\text{Loan Granted}_{ibt}=\alpha_t+\alpha_i+\alpha_b+\beta\mathbf 1\{Firm\ Risk_{it}\}+\delta(\Delta\text{Overnight Rate}_{t-1}\times\mathbf 1\{Firm\ Risk\})+\gamma(\Delta\text{Overnight Rate}\times\mathbf 1\{Firm\ Risk\}\times\ln(\text{Bank Capital}_{b,t-1}))+\cdots \tag{21.8}$$

and a log-loan-amount version (21.9). \(\mathbf 1\{Firm\ Risk\}=1\) if the firm had non-performing loans in the prior 4 years. Results (Figure 21.13): \(\delta,\delta'\) significantly negative (tightening → lending to safer firms, both extensive and intensive margins); \(\gamma,\gamma'\) significantly positive (well-capitalized banks take more risk in tight periods).

21.2.3 Ioannidou et al. (2015) looser policy → worse loan quality: Bolivia (peso pegged to USD, open account → exogenous policy, 90% deposits in USD), focusing on ex-post loan performance. Main finding: a lower short rate (loosening) → granting riskier loans (worse ex-ante credit history, lower ex-ante ratings, weaker ex-post performance, i.e. higher default); effects stronger for small firms borrowing from multiple banks.

Uses FDIC deposit quantities (1994–2014) + Ratewatch deposit rates (1997–2014). Core idea: banks have market power → deposit rates are insensitive to policy (Figure 21.14); monetary contraction (short rate up) → deposits become unfavorable → deposits flow out of the banking system (Figure 21.15); banks respond by cutting loans (a traditional monopoly profit-maximization result) — this is the deposit channel of monetary policy.

Model: one period, no risk. The representative household maximizes \(u(W_0)=\max\left(W^{\frac{\varepsilon-1}\varepsilon}+\lambda^{\frac1\varepsilon}\ell^{\frac{\varepsilon-1}\varepsilon}\right)^{\frac\varepsilon{\varepsilon-1}}\) (\(\lambda\) a share, \(\varepsilon\) the elasticity of substitution between wealth \(W\) and liquidity service \(\ell\)). Bank \(i\)'s deposit spread at \(t\) is \(s_{it}=f_t-r_{it}^{dep}\) (\(f_t\) the Fed funds rate, \(r^{dep}\) the deposit rate). Proposition: with \(\rho<1<\varepsilon,\eta\) (\(\eta\) the cross-bank elasticity), \(\mathcal M=1-(\eta-1)(N-1)\) capturing banks' deposit market power, as \(\lambda\to0\), if \(\mathcal M<\rho\) the spread is zero, otherwise

$$s=\delta^{\frac1{\varepsilon-1}}\left(\frac{\mathcal M-\rho}{\varepsilon-\mathcal M}\right)^{\frac1{\varepsilon-1}}f$$

the spread \(s\) increases in market power \(\mathcal M\), increases in the Fed funds rate \(f\), and increases more in \(f\) when \(\mathcal M\) is higher.

Empirics: (1) a time-series regression \(\Delta y_{it}=\alpha_i+\beta_i\Delta f_t+\xi_{it}\) by county Herfindahl percentile (Figure 21.16: higher market power → more insensitive deposit rate = more sensitive spread); (2) a within-bank regression \(\Delta y_{ijct}=\alpha_i+\delta_{jt}+\zeta_c+\lambda_{st}+\gamma\Delta f_t\times HHI_c+\varepsilon_{ijct}\) (the bank-time FE \(\delta_{jt}\) makes \(HHI_c\) exogenous: multi-branch banks shift funds across branches to meet demand, ruling out demand factors; Figures 21.17, 21.18 contraction → higher spread and deposit outflow); (3) a county-level \(y_{ct}=\alpha_c+\delta_t+\beta HHI_{c,t-1}+\gamma\Delta f_t\times HHI_{c,t-1}+\varepsilon_{ct}\) (Figure 21.19 contraction → less new lending, slower employment/wage growth); (4) a bank-level \(\Delta y_{it}=\alpha_i+\delta_t+\beta HHI_{i,t-1}+\gamma\Delta f_t\times HHI_{i,t-1}+\varepsilon_{it}\) (Figure 21.20 contraction → less deposit, more wholesale funding, smaller assets, less real-estate lending). An interesting question: where does the deposit outflow go during contraction? (Relevant to shadow banking.)

Schenone (2010) defines a new measure of bank-firm relationship and analyzes how it affects the loan interest rate (price of credit). Data: U.S. corporate IPOs (SDC) + loans (Dealscan, 100,000 USD or more) 1998–2003. Relationship measure: \(\text{Prior by Lead}_{i,l}\) = the total number of firm \(i\)'s loans the lead bank of loan \(l\) has previously participated in; \(\text{Loans to Data}_{i,l}\) = firm \(i\)'s cumulative loans up to loan \(l\); the relationship intensity

$$\text{Intensity}_{i,l}\equiv\frac{\text{Prior by Lead}_{i,l}}{\text{Loans to Data}_{i,l}} \tag{21.10}$$

Two opposite forces: (1) the information-processing effect — as the relationship develops the lender's information advantage rises and the firm's credit competition falls → the relationship lender extracts more information rent → the borrowing rate rises in intensity; (2) the information-spillover effect — firm-specific information spills over to non-lenders → adverse selection falls, competition rises → the borrowing rate falls in intensity. Design: exploit the "exogenous" information shock from an IPO (before the IPO there's no credible disclosure channel, adverse selection is severe, the relationship lender exploits its info monopoly; after the IPO firms are subject to disclosure regulation, can't keep info private). Main regression \(\text{Spread}_{i,l}=\beta_0+\beta_1(\text{Intensity}\times\text{After IPO})+\beta_2(\text{Intensity}\times\text{After IPO})^2+\beta_3(\text{Intensity}\times\text{Before IPO})+\beta_4(\text{Intensity}\times\text{Before IPO})^2+\beta_5\text{After IPO}+\beta_6\text{Switched Lenders}+\beta_7\text{First Loan}+\boldsymbol\gamma'\mathbf X+\phi_{l,t}+\phi_i+\varepsilon\) (\(\text{Spread}\) = All-in Spread Drawn). Results (Figure 21.21): pre-IPO the spread is "U"-shaped in intensity (\(\beta_3<0,\beta_4>0\): at low intensity spillover dominates → lower rate to maintain the relationship; at high intensity processing dominates → extract more rent); post-IPO it monotonically declines (\(\beta_1<0\), \(\beta_2\) insignificant: asymmetry disappears, the relationship is frequent only when the borrowing cost is low).

Giannetti et al. (2017) show that before sharing data with competitors via a public credit registry, banks manipulate borrowers' ratings (downgrade high-quality, upgrade low-quality) to protect their own information rent. Credit registry (public, run by central banks/supervisors) vs credit bureau (private, commercial). Background: Argentina's public-credit-registry expansion — established 1991, institutions report loan amount, collateral, private ratings (integers 1–5, 1 best; 1/2 privately decided, very informative: default 3.6% for rating 1, 21% for rating 2); the central bank originally used magnetic tapes (costly) to report monthly only borrowers with total loans more than 200,000 USD or ratings 3/4/5. In April 1998 it announced a switch to CD-ROMs (cheaper) sharing all borrowers' monthly info → greatly improved sharing. Focus on the interim period (after announcement, before implementation, April–June 1998, when banks haven't observed others' shared info, no learning).

Design: focus on borrowers with total loans 150,000–250,000 USD before the announcement (the affected group = high-quality borrowers with total borrowing below 200,000 USD; info on 200,000–250,000 USD was already public → control). Two groups: Group 1 single-relationship, rating 1 (highest rent, strongest downgrade incentive); Group 2 multiple-relationship, rating 2. DiD (OLS):

$$\text{Downgrade}_{i,b,t}^{1\to2}=\beta_0+\beta_1\text{Treated}_i+\beta_2\text{Treated}_i\times\text{Interim}_t+\beta_3\text{Treated}_i\times\text{Post}_t+\beta_4\text{Treated}_i\times\text{Interim}_t\times\text{Local}_b+\beta_5\text{Treated}_i\times\text{Post}_t\times\text{Local}_b+\boldsymbol\gamma_1'\mathbf X_{b,t}+\boldsymbol\gamma_2'\mathbf X_{i,t}+\xi_t+\lambda_b+\phi_{b,t}+\varepsilon \tag{21.11}$$

strategic upgrade (Group 2) has the same form (21.12); the lending-relationship change \(y_t=\beta_0+\beta_1\text{Treated}_i+\beta_2\text{Treated}_i\times\text{Post}_t+\psi_i+\chi_{d,t}+\boldsymbol\gamma'\mathbf X_{i,t}\) (21.13) (\(y\) = total debt, number of relationships, local/foreign bank shares). Results: (1) strategic downgrade of high-quality borrowers — banks are more likely to downgrade single-relationship rating-1 borrowers in the interim to protect rent (not post-implementation = learning/competition start; primarily driven by local banks; Figures 21.22, 21.23); (2) strategic upgrade of low-quality borrowers — banks are more likely to upgrade multiple-relationship rating-2 borrowers in the interim to avoid a creditor run (after implementation all creditors see the bad rating and would run, so the bank giving the bad rating hides it; Figure 21.24); (3) relationship changes — after implementation, high-quality single-relationship borrowers shift from local (relationship) to foreign (transactional) banks, and multiple-relationship borrowers increase their number of relationships (Figure 21.25).