26. Financial Regulation

26.1.1 Narrow banking. The extreme regulatory idea is Henry Simons' "Chicago Plan": require banks to hold 100% reserves (or safe assets) against their demand deposits, so that deposits are fully backed and runs are eliminated outright. The Banking Act of 1935 institutionalized federal deposit insurance and separated commercial from investment banking. Real-world regulation is far less extreme, but its spirit — restraining bank fragility through regulation — runs throughout.

26.1.2 Why regulate banks. Banks are special because of their maturity/liquidity mismatch: they finance long-term risky investment with on-demand demand deposits (fragile liabilities).

• Deposit insurance prevents runs (Diamond-Dybvig 1983): the deposit contract itself admits a "run" and a "no-run" equilibrium; deposit insurance eliminates the bad one.
• Fragility is a feature, not a bug (Diamond-Rajan 2001): on-demand deposits are the device that disciplines the banker and forces liquidity creation; but this also makes banks inherently run-prone, so regulation/insurance must backstop them.

26.1.3 Major types of regulation. Modern regulation centers on the Basel III accord, with three classes of constraint.

(i) Capital adequacy. A bank must hold minimum capital relative to its risk-weighted assets (RWAs). The minimum Tier 1 ratio is

$$\frac{\text{Tier 1 Capital}}{\text{Risk-Weighted Assets}}\ge 4.5\%+1.5\%=6\% \tag{26.1a}$$

where 4.5% is common equity Tier 1 (CET1) plus an additional 1.5% of other Tier 1. Riskier assets carry larger weights and require more capital.

(ii) Leverage ratio. A risk-weight-independent "backstop" constraint:

$$\text{Leverage Ratio}=\frac{\text{Tier 1 Capital}}{\text{Total Exposure}}\ge 3\% \tag{26.1b}$$

(iii) Liquidity. Two ratios: the liquidity coverage ratio (LCR) ensures enough high-quality liquid assets under a 30-day stress; the net stable funding ratio (NSFR) ensures a stable longer-term funding structure.

$$\text{LCR}=\frac{\text{High Quality Liquid Assets}}{\text{Total Net Liquidity Outflow over 30 Days}}>100\% \tag{26.1c}$$

$$\text{NSFR}=\frac{\text{Available Stable Funding}}{\text{Required Stable Funding}}>100\% \tag{26.1d}$$

There is also a reserve requirement (U.S. as of January 2020, tiered on net transaction accounts: 0% / 3% / 10%).

26.1.4 The Glass-Steagall Act (GSA). Four sections of the Banking Act of 1933, the core of which is the separation of commercial and investment banking — barring deposit-taking commercial banks from securities underwriting/proprietary investment-banking activities, to insulate depositors' money from high-risk market activity. The GSA was repealed in November 1999 by the Financial Services Modernization Act (Gramm-Leach-Bliley), after which universal banking re-emerged, often blamed for contributing to the 2008 crisis.

26.1.5 Other regulation. Also includes: large-exposure limits (capping concentration to a single counterparty), credit-rating requirements, financial reporting and disclosure (to the SEC), and corporate-governance regulatory requirements.

26.2.1 Question and institutional background. Behn, Haselmann, and Wachtel (2016) study how model-based capital adequacy amplifies the pro-cyclicality of lending.

• Basel I: assets are placed into a few risk buckets with fixed risk weights, but the differentiation is too coarse — e.g., all corporate loans share the same weight. This gives banks an incentive to hold the riskiest asset within each bucket (under equal weights, riskier assets earn higher returns).
• Basel II offers two approaches:
• Standard Approach (SA): fixed risk weights much like Basel I.
• Internal Ratings-Based (IRB): the bank uses its own model to assess credit risk — first classifying assets under the Basel II framework, then computing RWAs via risk-weight functions whose inputs are probability of default (PD), loss given default (LGD), exposure, and maturity.

The German banking sector implemented Basel II from January 2007, choosing SA or IRB (but cannot switch back later). In the early stage (around September 2008), a bank that chose IRB could use IRB for some loan portfolios while keeping SA for the rest.

26.2.2 Data and identification. The primary data is the German credit register (compiled by the central bank, Deutsche Bundesbank), with quarterly data on all loans above 1.5 million Euros: each loan's regulatory approach, credit-risk estimates (e.g., PD), lender/borrower identity, corresponding RWAs, and collateral. Supplementary data is bank balance-sheet information from the central bank's BAKIS database. Sample period 2008 Q1–2011 Q3; pre-event 2008 Q1–Q2, post-event 2008 Q4–2011 Q3. Classification: banks using only SA are SA banks; banks using both SA and IRB are IRB banks; loans computed with IRB are IRB loans, those with SA are SA loans.

Exogenous variation (quasi-experiment): the failure of Lehman Brothers in September 2008 is treated as an exogenous credit-risk shock to the German banking sector (exogeneity comes from Lehman's failure being an unexpected event outside Germany). The shock exogenously raises the internal risk estimates and capital charges of loan portfolios under IRB, while capital charges under SA are unchanged. This is the exogenous variation in capital requirements — Figure 26.1 shows the RWAs/loans ratio for IRB loans rising tremendously because of Lehman (using IRB loans that exist throughout the sample, so the loan level itself is roughly flat).

26.2.3 Three bank-firm-level tests. The dependent variable is always the change in log loans across pre/post periods, \(\Delta\ln(\text{loans})_{ij}\) (bank \(j\) to firm \(i\)). The book labels the key explanatory variable "Dependent Variable" (i.e., an IRB-exposure proxy); \(\mathbf X\) are control variables.

Test 1: how IRB banks adjust lending relative to SA banks.

$$\Delta\ln(\text{loans})_{ij}=\alpha_i+\beta\,(\text{Dependent Variable}_j)+\boldsymbol\gamma'\mathbf X_{ij}+\epsilon_{ij} \tag{26.1}$$

\(\alpha_i\) is firm \(i\)'s loan-demand fixed effect (so only firms borrowing from at least two banks are used, following Khwaja-Mian 2008); the key variable \(\text{Dependent Variable}_j\) is either \(\text{Share IRB}_j\) (the share of IRB loans in bank \(j\)) or \(\text{IRB Bank}_j\) (a dummy for whether bank \(j\) is an IRB bank). Observations are collapsed to pre/post averages, with one observation per bank-firm relationship.

Test 2: restricting the sample to SA loans, comparing SA loans from IRB banks vs SA loans from SA banks.

$$\Delta\ln(\text{loans})_{ij}=\alpha_i+\beta\,(\text{Dependent Variable}_j)+\boldsymbol\gamma'\mathbf X_{ij}+\epsilon_{ij} \tag{26.2}$$

Test 3: restricting the sample to firms that borrow from at least two IRB banks (one IRB loan from an IRB bank, and one SA loan from another IRB bank), comparing the same firm's loans from the IRB pool vs the SA pool of different IRB banks.

$$\Delta\ln(\text{loans})_{ij}=\alpha_i+\alpha_j+\delta\,\text{IRB Loan}_{ij}+\boldsymbol\gamma'\mathbf X_{ij}+\epsilon_{ij} \tag{26.3}$$

This includes both a firm fixed effect \(\alpha_i\) and a bank fixed effect \(\alpha_j\); \(\text{IRB Loan}_{ij}\) is a dummy (=1 in bank \(j\)'s IRB pool, =0 in the SA pool).

Results (Figure 26.2): the three tests agree — a higher capital charge (tightening capital adequacy) shrinks the loan size.

• Columns (1)–(3) (Test 1): banks with a higher IRB-loan share cut lending more; adding more control variables (columns (5)(6)) removes the significant effect of \(\text{Share IRB}\) (column (4)), indicating no significant fundamental difference between IRB and SA banks — which resolves the selection-bias concern from banks self-selecting into IRB.
• Columns (4)–(6) (Test 2): SA loans from IRB banks and from SA banks respond similarly to the Lehman shock (SA loan capital charges did not change, hence no difference).
• Columns (7)–(9) (Test 3): with bank fixed effects + firm fixed effects, the same firm's loans in the IRB pool shrink more severely than in the SA pool.

26.2.4 Firm-level test: can firms offset by borrowing elsewhere? Testing the total effect of higher capital requirements on a firm's access to funds — whether firms can offset the IRB-loan contraction by borrowing from other banks. The firm-level regression:

$$\Delta\ln(\text{firm loans})_i=\alpha+\beta\,\overline{\text{IRB Loan}}_i+\boldsymbol\gamma'\mathbf X_i+\epsilon_i \tag{26.4}$$

The dependent variable is the change in log of firm \(i\)'s total bank loans across pre/post. The key variable (the book's "Dependent Variable") \(\overline{\text{IRB Loan}}_i\) takes two measures: the share of firm \(i\)'s loans (from all banks) that were IRB loans prior to the shock; or \(\overline{\text{IRB Loan}}_i^{\,*}\), the share of firm \(i\)'s loans from IRB banks that were IRB loans prior to the shock (the two have similar economic implications and serve as robustness for each other).

Results (Figure 26.3): firms with larger pre-shock exposure to IRB loans see a larger drop in total loans when capital adequacy tightens — firms cannot fully offset the effect of tightening bank capital regulation. Controlling for bank and firm fixed effects (the Test 3 columns of Figure 26.3), loans in the IRB pool shrink more severely than in the SA pool.

26.2.5 Intensive margin vs extensive margin vs price. Decomposing on which margin the adjustment happens:

• Extensive margin (Figure 26.4): IRB loans that existed before the shock do not have a significantly higher probability of exit afterward — i.e., lending relationships are not easily severed.
• Price (Figure 26.5): the difference in loan interest rates between pre- and post-event periods is not significant — i.e., loan price barely adjusts to the change in capital requirements.

Conclusion: when capital adequacy requirements change, loans adjust mostly on the intensive margin of quantity (the loan amount within existing relationships), not on the extensive margin of quantity or on price. The intensive margin is more important; the extensive margin is not.

26.3.1 ABCP conduits and regulatory arbitrage. Acharya, Schnabl, and Suarez (2013) use asset-backed commercial paper (ABCP) conduits to illustrate how regulatory arbitrage occurs. The ABCP market grew from 650 billion USD in January 2003 to 1.3 trillion USD in July 2007. The basic logic:

• To avoid capital-adequacy regulation, banks set up off-balance-sheet conduits.
• The conduit buys long-term risky assets (such as mortgages) and issues short-term commercial paper to finance the purchase.
• The bank, as the conduit's sponsor, provides it a liquidity guarantee.

26.3.2 The "liquidity" guarantee is really a "credit" guarantee. The crux is the design of this liquidity guarantee:

• The liquidity guarantee requires the sponsor to buy back ABCP at par as long as the conduit assets are not in default.
• But technically, ABCP holders can always withdraw their money from the sponsor before the conduit goes into default.
• So this nominal "liquidity" guarantee is effectively a "credit" guarantee — the sponsor in fact bears the credit risk of the conduit's assets.

Because the conduit is off-balance-sheet, the bank need not set aside any capital for the transaction. Ideally, such financial innovation should make the bank an equity holder of the conduit and limit the bank's exposure to the conduit's risk to the equity level. But this is not true: in a run, the bank ends up buying back the commercial paper and bearing all the risk. So ABCP is a way for banks' arbitrage from regulation, but not a way to actually get rid of the fundamental risk.

Data: the authors use the panel data set on the universe of conduits from January 2001 to December 2009.

26.3.3 Documented facts.

• Conduits with commercial banks as sponsor and a liquidity guarantee (i.e., the regulatory-arbitrage type) rocketed before the 2008 financial crisis (Figure 26.6: ABCP outstanding by type of guarantee with a commercial-bank sponsor; the liquidity-guarantee type peaks at about 800 billion USD in mid-2007, far above credit, extendible, structured-investment-vehicle, and other types).
• This regulatory arbitrage results in a huge amount of missing capital (capital that should have been prepared but actually was not), large both in levels (Figure 26.7, third column) and as a percentage of bank equity (fourth column). The missing-capital level (third column) is computed by multiplying ABCP face value (in billions of USD, second column) by 8%; the first column is the bank's Tier 1 capital.
• The bank with the largest proportion of missing regulatory capital (e.g., Sachsen Landesbank, missing 79.9%) is the first bank that cannot be bailed out if it cannot provide the guarantees extended to its conduits.
• Conclusion: in a run, smaller banks are completely wiped out by the conduit losses; larger banks survive the conduit losses but are weakened.

26.4.1 The inconsistent-regulator problem. Agarwal, Lucca, Seru, and Trebbi (2014) study the inconsistency of regulators when implementing identical regulation rules.

Institutional background: U.S. banks can choose between a national charter and a state charter:

• Nationally chartered banks are supervised only by federal regulators (the Office of the Comptroller of the Currency, OCC).
• State-chartered banks are supervised jointly by both state and federal banking regulators, accounting for about 70% of U.S. commercial banks. The specific federal regulator depends on the bank's membership in the Federal Reserve system: the Fed supervises state member banks (SMBs), the FDIC supervises nonmember banks (NMBs).
• Alternate examination programs (AEP): state-chartered commercial banks' onsite examinations are fixed to rotate between state and federal regulators every 12 or 18 months (since 1997, banks with assets below 250 million USD and a CAMELS rating of 1 or 2 are examined every 18 months; others every 12 months). Each spell's onsite examination produces a written report and a CAMELS rating; the rating stays constant between two rotation spells.

26.4.2 Data. The focus is banks subject to the AEP. The primary data is from the National Information Center of the Federal Reserve, containing all onsite examination results (e.g., CAMELS) conducted by U.S. banking regulators. Supplementary information: balance sheet, profitability, and asset quality from quarterly Call Reports (the Consolidated Report of Condition and Income); budget and other information on state banking departments (from annual reports of the Conference of State Banking Supervisors); and state-level economic measures and indicators of banking stress such as failure rates. The sample period is 1996 Q1 to 2010 Q4.

26.4.3 Identification. The exogenous variation comes from the exogenous timing of the supervisory rotations — the timing of rotation is exogenous to the banks. Almost all 18-month-spell banks have examination spells below the threshold (assets < 250 million, rating 1/2); but some 12-month-spell banks have the spell rotating at 5 or 6 quarters.

26.4.4 Main regression.

$$Y_{it}=\alpha+\boldsymbol\beta'\mathbf B_{it}+\boldsymbol\sigma'\mathbf S_{it}+\theta_i+\lambda_t+\epsilon_{it} \tag{26.5}$$

• \(Y_{it}\) is a regulatory outcome of interest (such as the CAMELS rating).
• \(\mathbf B_{it}\) is a vector of characteristic variables of bank \(i\) at quarter \(t\).
• \(\mathbf S_{it}\) is a vector of characteristic variables of bank \(i\)'s supervisor at quarter \(t\); it can reduce to a single dummy indicating the regulator's identity (=1 if a state regulator).
• \(\theta_i\) is a bank fixed effect, \(\lambda_t\) a quarter fixed effect; both drop in within-bank analysis.

26.4.5 Results.

• State regulators are more lenient: banks receive more lenient outcomes from state regulators (more upgrades, fewer downgrades, and less harsh downgrades). See Figure 26.8 (the impact of regulator identity on CAMELS upgrades and downgrades under SMB/FDIC/Federal analyses, state vs federal rotating).
• Banks "behave better" under federal supervision (Figure 26.9): under a federal regulator spell, the bank has a higher Tier 1 risk-weighted capital ratio (more compliant, column (1)), a higher leverage ratio (column (2)), a higher expense ratio (higher costs, harder to pursue risky activities, column (3)), a lower return on assets (lower profitability, mostly from greater difficulty in pursuing risky-but-profitable activities, column (4)), a higher nonperforming-loan share (more frank about bad loans, column (5)), higher delinquency rates (more bad behavior identified, column (6)), and lower new loan growth (column (7), but insignificant).

26.4.6 The cost of inconsistency. State regulators consistently give lower (more lenient) CAMELS ratings than federal regulators, a result robust across specifications (Figure 26.10: the cost and benefit of inconsistent regulation, state-level analysis). The CAMELS spread (state regulators being more lenient) is larger when:

• the bank failure rate is high;
• the bank problem rate is high;
• the TARP (Troubled Asset Relief Program) repayment rate is low;
• the asset-sale discount of troubled banks liquidated by the FDIC is high.

Other results: the spread is larger when the state has a smaller budget for supervision or less qualified supervisors (insufficient regulatory capacity yields more leniency); there is no evidence that corruption or the "revolving door" (regulators being lenient to land a job at the regulated institution later) explains the magnitude of the spread.

Takeaway: state regulators' greater leniency magnifies the cost precisely when local conditions deteriorate — nominally identical regulatory rules produce systematic differences depending on who implements them.

References: Acharya, Schnabl, and Suarez (2013, JFE) Securitization without risk transfer; Agarwal, Lucca, Seru, and Trebbi (2014, QJE) Inconsistent regulators: Evidence from banking; Behn, Haselmann, and Wachtel (2016, JF) Procyclical capital regulation and lending; Diamond and Dybvig (1983, JPE) Bank runs, deposit insurance, and liquidity; Diamond and Rajan (2001, JPE) Liquidity risk, liquidity creation, and financial fragility: A theory of banking; Khwaja and Mian (2008, AER) Tracing the impact of bank liquidity shocks: Evidence from an emerging market.

(End of chapter, and the conclusion of Part VI "Reduced Form Empirical Analysis.")