On Applications Using Credit Registers · External Validity . Internal Validity (Identification)...

On Applications Using Credit Registers

Steven Ongena

University of Zurich, Swiss Finance Institute, KU Leuven, CEPR

CONFERENCE ON THE USE OF CREDIT REGISTER DATA

FOR FINANCIAL STABILITY PURPOSES

DANMARKS NATIONALBANK, COPENHAGEN, OCTOBER 24, 2019

Policy Evaluation Financial Stability Board

Page 2

Source: Proposed Framework for Post-Implementation Evaluation of the Effects of the G20 Financial Regulatory Reforms, Consultation Document on Main Elements

Dimensions of Assessment

Page 4

Heterogeneity General Equilibrium

Attribution

External Validity

Internal Validity (Identification)

Source: Proposed Framework for Post-Implementation Evaluation of the Effects of the G20 Financial Regulatory Reforms, Consultation Document on Main Elements

Heterogeneity General Equilibrium

Attribution

How To Climb to «the Top»?

Page 6

Attribution

Heterogeneity General Equilibrium

Heterogeneity

Attribution

General Equilibrium

Data Skill

Page 8

Dimensions of Data

Page 9

Cross-sectional Comprehensiveness

Time-series Duration

Granularity/Frequency

Heterogeneity General Equilibrium

Attribution

Theme Objective Paper

«Max that match» to make it to within-firm heaven

Bonfim, Nogueira, Ongena “Sorry, We're Closed." Loan Conditions When Due to Branch Closure Firms Transfer to Another Bank

«Open to many shocks» but only one credit register Célérier, Kick, Ongena Taxing Bank Leverage

«Match that data» to get that shock Gunduz, Ongena, Tümer-Alkan, Yu CDS and Credit

Theme Objective Paper

«Max that match» to make it to within-firm heaven

Bonfim, Nogueira, Ongena “Sorry, We're Closed." Loan Conditions When Due to Branch Closure Firms Transfer to Another Bank

«Open to many shocks» but only one credit register Célérier, Kick, Ongena Taxing Bank Leverage

«Match that data» to get that shock Gunduz, Ongena, Tümer-Alkan, Yu CDS and Credit

Motivation

We study loan conditions when bank branches close

and firms subsequently transfer to a branch of another

bank in the vicinity.

To observe the conditions granted when banks “pool” price new applicants.

Motivation

• What happens when firms switch banks after a branch closes?

• There is a loss of information.

• Outside banks deal with many new applicants at once, about whom they know very little.

• Theory suggests that the outside bank will pool-price the new loans (von Thadden, FRL 2004).

• This setting allows us to test (for the first time in the literature) if the discounts are driven by «shoe leather» costs or by information asymmetries.

Bank 1 branch

Bank 2 branch

Switch

Switch: If the firm gets a new

loan from a bank from whom it hasn’t borrowed in the last 12 months (outside bank).

The firm had a

relationship with at least one other bank for at least 12 months (inside bank).

Definitions

Bank 1 branch

Bank 2 branch

5km

Transfer

Transfer loan: subgroup of switching loans.

A switching loan is a

transfer loan if the closest branch of one of the inside banks closes before a new loan is granted by an outside bank.

- After the closure, the

closest branch from the inside bank must be more than 5 km away from the firm.

Definitions

Branch closure

• There are 839 branch closures in our sample.

• Quasi-natural experimental setting

• Some of the largest banks were recapitalized with funds from the bailout package agreed with the IMF, the ECB and the European Commission

- These banks had to submit restructuring plans, with the aim of improving profitability and solvency.

- Prime cost-cutting measures: reductions in branches and staff members, implemented in a very short time frame.

Data

• Credit register: monthly loan data on all exposures.

• New operations database: monthly information on interest rates on all new loans granted by the largest banks.

• Branch register: list of all bank branches of resident financial institutions with postal codes, opening day and closing day.

• Period: 2012:06 to 2015:05 .

Matching

Ideal setting: we would need to know the interest rate offered to the firm for a non-switching loan. Solution: matching on observables (coarsened exact matching) • quarter • firm characteristics (credit rating, region and sector) • loan characteristics (collateral, maturity, loan amount, floating rate loan)

similar to Ioannidou and Ongena (JF, 2010)

Why Matching and Why Not Regressing?

• A regression model works if either of two assumptions (“double robustness”) is satisfied:

• if the linear model is true • if the two groups are balanced (so that you’re getting an

average treatment effect)

• Look at matching as “a tool for making a regression

more effective” (also in this application) • Angrist, J.B. & J.-S. Pischke, “Mostly Harmless Econometrics.”

19

Matching

• Empirical strategy: we match all switching/transfer loans with non-switching loans that have the same characteristics and calculate the spread between the interest rate on these loans.

• We regress the spread on a constant and weigh by one over the total number of comparable nonswitching loans per switching loan.

• For instance, if transfer i has 6 matches, each match will have a weight of 1/6 in the regression.

• We cluster at the switching-firm level.

Matching

We match on:

1. Inside bank: compare the rates on switching or transfer loans with non-switching loans being granted by the inside bank (columns I and II).

2. Outside bank: compare the rates on switching or transfer loans with non-switching loans being granted by the outside bank (column III) – baseline approach.

3. Firm: compare the rates on switching or transfer loans with other loans being granted at the same time to the same firm (column IV) – ideal approach, but few observations.

BenchmarkMatching Variables I II III IV

Quarter Yes Yes Yes YesInside bank Yes YesOutside bank YesForeign bank YesFirm YesCredit rating Yes Yes Yes YesRegion Yes Yes Yes YesIndustry Yes Yes Yes YesLegal structure Yes Yes Yes YesCollateral Yes Yes Yes YesLoan maturity Yes Yes Yes YesLoan amount Yes Yes Yes YesFloating loan rate Yes Yes Yes Yes

Number of switching loans 6.265 4.231 6.931 1.639Number of nonswitching loans 31.560 20.531 23.892 3.382Number of observations (matched pairs) 50.915 28.181 33.274 12.906

Interest rate difference with matching -122.37*** -88.96*** -58.53*** -91.93***(-7.87) (7.00) (4.60) (12.37)

Interest rate difference without matching -149.07*** -107.83*** -53.28*** -64.67**(8.25) (9.01) (8.60) (31.56)

Switching

Period since the branch closure Before1-6 months

after7-12 months

after>12 months

after

Number of switching / transfer loans 230 68 78 236Number of nonswitching loans 878 295 338 986Number of observations (matched pairs) 1.050 305 535 1.371

Interest rate difference with matching -62.81*** 15.62 -57.30* -94.21***(23.66) (29.55) (33.85) (16.84)

Interest rate difference without matching -79.73*** -180.55*** -209.16*** -263.39***(21.07) (29.88) (28.61) (21.78)

Transfer

Take-away: No discount on transfer loans Transferring after 6 months similar to switching

Tons of Robustness: Match on

•Firm

• Firm size

• Municipality

• Local branch density.

• Switching and transfer firms arriving at the same bank

Tons of Robustness: Are branch closures really exogenous?

• We include only branch closures by banks that were recapitalized with bailout funds (more externally imposed).

• We estimate a model to derive the likelihood of branch closure (exploring information on bank size, local branch density and branch portfolio quality).

•Re-estimate our main results for the sub-sample of branches that were less likely to close (first of three quantiles)

Tons of Robustness and Exploration: Do branch closures affect competition? What is the impact of branch closures?

• etc.

Summing up • Switching loans get lower interest rates than

nonswitching loans (58 bps)

• After the closure of a branch of an inside bank, firms that transfer to another bank close by do not get lower interest rates (evidence of pool pricing)

• for later transfers, the switching discount is again observed

• rhis evidence is consistent with the information asymmetry hypothesis:

• under competitive conditions, shoe-leather switching costs would also yield discounts for transfer loans.

Matching: Trade Off

precision of matching (high-quality observations)

with

external validity

(selected set of observations) and

statistical power (number of observations left)

Theme Objective Paper

«Max that match» to make it to within-firm heaven

Bonfim, Nogueira, Ongena “Sorry, We're Closed." Loan Conditions When Due to Branch Closure Firms Transfer to Another Bank

«Open to many shocks» but only one credit register Célérier, Kick, Ongena Taxing Bank Leverage

«Match that data» to get that shock Gunduz, Ongena, Tümer-Alkan, Yu CDS and Credit

Research Questions

• Can bank equity subsidization / bank leverage taxation, by making equity relatively cheaper and/or leverage more costly, be a complementary regulatory tool to control bank leverage?

• What would be the effect on bank capital structure,

bank lending and risk taking?

Allowance for Corporate Equity (ACE) Schepens (JFE, 2016)

What?

Same tax advantage to equity as to debt by deducting a notional interest on equity from taxes

How?

Where?

The regulator defines a notional interest rate R R × Book Value of Equity is deducted from income before taxes

In Belgium in 2005, triggered by European Commission

Altering the Relative Cost of Equity? ACE

Nice Quasi-experimental Set-up

Schepens (JFE 2016)

(1) No other simultaneous major tax reforms

(2) Applies only to a subset of banks (active in Belgium) subject to the same European regulatory framework

(3) Did not affect the demand for credit from most Belgian firms

(4) Applied to banks that were actively lending abroad • We can investigate the effect of the reforms on bank lending in

markets that these tax reforms did not affect.

Key Findings: Economic Relevancy Setup

1.1 percentage point (pp) increase in the relative cost of leverage

≈ The Belgian (full) ACE Corporate tax rate = 35%

Cost of equity = 3.5% = 0.035 * 0.35

(WACC decreases by 7.5 bps)

Key Findings: Bank Level

A 1.1 pp increase in the relative cost of bank liabilities increases the bank equity ratio by 1 pp

• As in Schepens (JFE 2016) • Large: Basel III is supposed to raise minimum capital

requirements from 4.5% to 6% over 6 years • Driven by an increase in the level of equity • Larger for banks with an ex-ante low level of equity

• So not driven by the subsidy effect

• Robust • Difference-in-Differences: Various Propensity Score Matching

and including bank size quintiles x year and bank leverage quintiles x year fixed effects

Key Findings: Bank Level

A 1.1 pp increase in the relative cost of bank liabilities increases the loan (-to-asset) ratio by 5 pp

• Large: an additional $35 billion of credit supply, i.e., 9% of

the GDP in Belgium in 2005 • Larger for banks with an ex-ante low level of “leverage

ratio” (=equity over assets)

Bank balance sheets Bank-firm Credit supply

Bank-firm Risk-taking

Identification Strategy : Credit Supply

Firms

Domestic banks

Germany

Identification Strategy: Credit Supply

Firms

Tax policy reform

3 Belgian banks

Domestic banks

Germany

Euro-area banks

Firm Fixed Effects (in a Difference-in-Differences)

Firms 3 Belgian banks

Domestic banks

Germany

Identification Strategy: Credit Supply

Euro-area banks

Tax policy reform

Khwaja & Mian (AER 2008) Most firms are large and use multiple banks (Degryse, De Jonghe, Jakovljevic, Mulier & Schepens, JFI 2019)

• All quarterly bank-firm exposures initially above 1.5 million euros

• Firms that borrowed at least once from banks in two countries, 2003-2007 • We back-fill exposures to create a balanced panel

e.g., Schertler, Buch & Westernhagen (IEEP, 2006), Hayden, Porath & Westernhagen (JFSR, 2007), Ongena, Tümer-Alkan & Westernhagen (EER, 2012), Behn, Haselmann & Wachtel (JF 2016), Haselmann, Schoenherr & Vig (JPE 2017), Ongena, Tümer-Alkan & Westernhagen (RF 2018), …

Lending Analysis: German Credit Register Data

Key Findings: Bank-Firm Level

(1) no German firms are affected by the introduction of the ACE (2) a subset of Belgian banks are lending actively in Germany

(3) the German economy is strong and stable, so it can absorb a positive supply shock

A 1.1 pp increase in the relative cost of bank liabilities increases the `supply` of credit by affected banks

to firms abroad , i.e., in Germany, by 16-30 pp (“share of wallet”)

Robust Difference-in-Differences: • Include at least bank, bank-firm and firm characteristics (incl. industry fixed effects)

• Can compare to foreign lending-only in Germany • Can saturate with firm fixed effects (Khwaja & Mian, QJE 2005)

Intensive Margin: Conditioning on bank-firm exposure in 2003 > 0

Why negative growth: Repayment of credit, no extensive margin, overall no credit growth Figure 3

Bertrand, Dufflo & Mullainathan (QJE 2004)

Davis & Haltiwanger (QJE 1992)

Table V

TAX

Liability Tax Devereux, Johannesen & Vella (2017)

What?

Taxes on bank total liabilities minus the value of equity

Why?

Where?

Ensure that banks make a contribution that reflects the potential risk to the financial system Encourage banks to move away from riskier funding

Staggered introduction in 6 European countries from 2010 to 2012, triggered by International Monetary Fund: Austria, Belgium, Germany, the Netherlands, Portugal and Slovakia

Altering the Relative Cost of Equity?

Table II

WACC decreases by 7.5 bps

Tentative Conclusion

• The paper studies the effect of a change in the (relative) fiscal cost of leverage on bank balance sheets and bank lending

• We show that increasing the fiscal cost of bank leverage has a positive effect on bank equity ratios and lending

• Related to deviation between regulatory risk weights and actual balance sheet item risk

• Fiscal policy might therefore be part of a solution for

financial stability and a credible complement to capital requirements

• Fiscal policy not remit of regulators • ACE leads to decrease in government revenues

• Trade-off with more banking sector stability and lower future bail-out costs?

One Credit Register

Can be used to study many shocks “coming in” from abroad

Internal validity = very good

Shocks likely exogenous and

with credit register many possibilities (although there may be spillovers)

External validity = maybe an issue

if cross-border credit is very different from domestic credit

Theme Objective Paper

«Max that match» to make it to within-firm heaven

Bonfim, Nogueira, Ongena “Sorry, We're Closed." Loan Conditions When Due to Branch Closure Firms Transfer to Another Bank

«Open to many shocks» but only one credit register Célérier, Kick, Ongena Taxing Bank Leverage

«Match that data» to get that shock Gunduz, Ongena, Tümer-Alkan, Yu CDS and Credit

«Bigly» Data

1. Credit register • Granular and comprehensive, with (some)

duration

2. Trading data • High-frequency, with (some) duration

Page 49

Motivation

Yalin Gündüz, CDS and Credit

Page 50

The default risk of a debt instrument should determine the purchase of

protection to hedge credit risk

• banks also trade for speculative purposes

• a CDS contract can be naked = without any underlying credit exposure

for the buyer

The ease of hedging could also affect lending

• Empty creditor problem (e.g. Bolton and Oehmke, RFS 2011) Empirical challenge: identify causal effects

CDS - credit nexus

Page 51

Bank level:

− Larger gross positions in credit derivatives lower loan spreads but no impact of net positions (Norden, Silva Buston and Wagner, JEDC 2014)

− More aggressive risk taking after using CDS: Loans to CDS-referenced borrowers larger and have higher yield spreads

(Shan, Tang, and Yan, 2014)

Firm level:

− No evidence for lower cost of debt; adverse effects on risky firms, such as rating downgrades and bankruptcies (Ashcraft and Santos, JME 2009; Subrahmanyam, Tang and Wang, RFS 2014)

− Higher leverage ratios and longer debt maturities (Saretto and Tookes, RFS 2013)

Bank-Firm level:

− Usage of CDS complements syndicated loan sales (Hasan and Wu, 2015)

What we do

Page 52

And does that affect the availability of new credit?

DTCC’s confidential dataset on bank’s CDS holdings on

European firms

Bundesbank’s credit register on bank exposures to firms

Does hedging motivate CDS trading?

Yalin Gündüz, CDS and Credit

we couple comprehensive bank-firm level data

to investigate:

Small Bang

Page 53

Contract and convention changes a higher degree of standardization − higher flexibility for dealers − a central decision maker − improved liquidity (Fulop and Lescourret, 2016)

“On March 11, 2009 major European dealers made a commitment to European regulators to begin clearing index and single name CDS trades through a European central clearing party by July 31, 2009” (Markit, 2009)

− trading with fixed coupons plus an

upfront fee − creating an event determination

committee − an auction mechanism that supports a

binding settlement

What happened Its effects

What we find

1. After the Small Bang,

exposures to riskier firms held by banks

protection-purchasing on these firms by these banks

2. Banks with CDS holdings also re-allocated their credit () maintaining

lending to safer firms despite a lending contraction

3. Only banks properly hedged take more risk!

Page 54

Yalin Gündüz, CDS and Credit

Main Contributions

1. First paper to use bank-firm CDS trading + bank-firm credit exposure

2. Our identification strategy relies on the Small Bang leading and foremost

affecting CDS trading

3. First hand evidence on the benefits of financial innovation

Page 55

Yalin Gündüz, CDS and Credit

Credit Register Matched with Other Data

Can be used to study market linkages and cross-market phenomena

The frontier, really …

To Summarize

• Use matching to obtain high-quality observations • Many data points, no fear

• Use shocks abroad • How exerternally valid?

• Use bigly data: the sky is the limit

CREDIT REGISTERS