manuel buchholz. caps on banks’ leverage and domestic credit after the crisis

Caps on banks’ leverage and domestic credit after thecrisis

Manuel Buchholz1,2

1Visiting Researcher at Eesti Pank2Halle Institute for Economic Research (IWH)

Open SeminarEesti Pank

September 28, 2015

The views expressed in the presentation are those of the author and do not necessarilyreflect the views of Eesti Pank (Bank of Estonia) or the IWH.

M. Buchholz Leverage caps & domestic credit Sep 28, 2015 1 / 29

Motivation – The (not so) new era of macro pru

In the wake of the Global Financial Crisis, macroprudential policy hasbecome a key element of the ongoing regulatory debateIn several (advanced) economies, macroprudential policy instrumentshave recently been implemented or are to be implemented in the nearfuture (loan-to-value ratios, countercyclical capital buffers)Shift from microprudential to macroprudential perspective of financialsupervision: Focus is on safeguarding stability of financial system as awhole and not only of individual institutionsMany emerging market (and few advanced) countries have alreadyimplemented macroprudential tools in the pastThese experiences allow studying the effects of macroprudential policyon financial and real outcomes


Table of contents

1 IntroductionMotivationMacroprudential policyResearch questionRelated literature

2 DataCaps on banks’ leverageReal credit growth

3 Empirical specification and estimation resultsEmpirical specificationEstimation results

4 Robustness

5 Concluding remarks


The aim and effectiveness of macroprudential policy

The aims of macroprudential policyI Safeguarding stability of the financial system by means of reducing the

build-up of systemic risksI Ultimate target: Making financial crises less likely and reduce their

adverse impact on the real economyI Intermediate target: Smoothing the financial cycle by means of

reducing (excessive) credit growth in boom times and stabilizingprovision of credit in times of financial downturns (countercyclicality)

Results of empirical research suggest that macroprudential policy iseffective in reducing credit growth in boom timesEvidence is less clear regarding the potentially stabilizing role ofmacroprudential policy during financial downturns

⇒ Focus of this paper


A specific tool: Caps on banks’ leverage

A cap on leverage implies that banks have to hold a minimum amountof capital with respect to their total (non-risk-weighted) assetsSuch a leverage ratio is currently tested under the Basel III frameworkand set at 3% to complement the already existing capitalrequirements incorporating risk weights (BCBS 2014)The goals

I prevent the build-up of excessive leverage in the financial systemI function as a backstop for risk-weighted regulatory capital ratios

Possible channel how the cap might stabilize lending after the crisis:Higher pre-crisis capital buffer might relax the binding constraint dueto risk-weighted capital requirements in times of financial downturn


Research question

Did the implementation of a leverage cap prior to the crisis affectprovision of credit to the private sector after the crisis?Endogeneity: Pre-crisis implementation of leverage cap is not randombut rather conditional on several (potentially unobserved) economicfactorsEmpirical strategy: difference-in-differences approach

I The Global Financial Crisis affected the capability of banks to providecredit

I The years after the crisis are an example for a period of financialdownturn

I Investigate if real credit growth after the crisis relative to the pre-crisisperiod differs for countries which had implemented a cap on banks’leverage prior to the crisis


Overview of related literature (1)

Empirical studies based on individual countriesI Aiyvar et al. (2014a): Capital surcharges lead to lower lending by

UK-owned banks and foreign subsidiaries; leakage due to increasedcredit provision by foreign branches

I Buch et al. (2014): Banks affected by the German bank levyimplemented in 2011 reduce lending and increase deposit rates

I Jiménez et al. (2014): Dynamic provisioning regulation on Spanishbanks effective in smoothing credit cycle

I Danisewicz et al. (2015): Lending by foreign subsidiaries and brancheslocated in the UK differs significantly in response to theimplementation of macroprudential policy in their home countries


Overview of related literature (2)

Multi-country studiesI Claessens et al. (2013): Macroprudential policy decreases banks’

leverage and asset growth during boom times; limited evidence forstabilizing effects of countercyclical instruments in downturns

I Aiyvar et al. (2014b): Capital surcharges lead to slowdown in growth ofcross-border credit by UK banks

I Cerutti et al. (2015): Macroprudential policy is effective in dampeningreal credit growth but works less well in busts


Table of contents




4 Robustness



Data – Caps on banks’ leverage

Source: Cerutti et al. (2015) – based on the IMF survey on GlobalMacroprudential Policy Instruments (GMPI) – provide the year ofimplementation of the cap but not its intensity

CountryYear of

implementation1

Value of

leverage cap2

Advanced

Canada ≤ 2000 5%

United States ≤ 2000 3% / 4%

Emerging

Chile ≤ 2000 5%

Ecuador 2001 n.a.

Jordan 2003 6%

Paraguay ≤ 2000 8% / 12%

Saudi Arabia ≤ 2000 6.25%

St. Kitts and Nevis ≤ 2000 n.a.

1 Source is Cerutti et al. (2015). 2 Measured as the implied minimum ratio of capital to total assets. For illustrativepurposes only. Correctness cannot be guaranteed because sources include non-official documents.


Data – Real credit growth before and after crisis

05

1015

2025

Rea

l cre

dit t

o pr

ivat

e se

ctor

(gr

owth

rat

e in

%)

2002 2004 2006 2008 2010 2012 2014

Year

Cap on leverage: no Cap on leverage: yes

Source: Computation of growth rates of real credit to the private sector based on data from the Other Depository CorporationsSurvey (IMF) and the World Development Indicators (World Bank). Growth rates were winsorized at the 1% and 99% quantiles.Sample is based on the estimation sample of 69 countries.


Data – Summary statistics

Variable Country group Observations MeanStandard

deviationMin Max

Leverage cap: yes 89 11.62 11.44 -14.10 60.74

Leverage cap: no 730 13.81 16.86 -14.10 78.34

Total 819 13.57 16.37 -14.10 78.34

Real GDP growth rate (in %) Leverage cap: yes 89 3.87 3.35 -5.60 14.22

Leverage cap: no 730 3.30 3.76 -11.77 18.23

Total 819 3.36 3.72 -11.77 18.23

Monetary policy rate (in %) Leverage cap: yes 89 4.56 3.40 0.13 15.36

Leverage cap: no 730 5.95 9.97 0.08 150.00

Total 819 5.80 9.49 0.08 150.00

CPI inflation rate (in %) Leverage cap: yes 89 3.99 3.04 -0.73 14.99

Leverage cap: no 730 4.77 7.01 -1.09 109.59

Total 819 4.68 6.70 -1.09 109.59

Leverage cap: yes 89 53.39 28.57 14.62 134.99

Leverage cap: no 725 65.06 40.55 4.20 172.41

Total 814 63.78 39.57 4.20 172.41

Leverage cap: yes 82 3.01 1.31 1.00 5.00

Leverage cap: no 677 1.38 1.27 0.00 6.00

Total 759 1.55 1.37 0.00 6.00

Leverage cap: yes 82 0.34 0.57 0.00 2.00

Leverage cap: no 677 0.34 0.63 0.00 2.00

Total 759 0.34 0.62 0.00 2.00

Capital ratio (in %) Leverage cap: yes 63 8.34 3.78 0.00 14.03

Leverage cap: no 691 9.91 4.26 0.00 22.56

Total 754 9.77 4.24 0.00 22.56

Financial institutions-targeted

macroprudential index (0-10)

Borrower-targeted

macroprudential index (0-2)

Private credit-to-GDP ratio (in %)

Real credit to the private sector

(growth rate, in %)

All variables were winsorized at the 1% and 99% quantiles. Sample is based on the estimation sample of 69 countries.M. Buchholz Leverage caps & domestic credit Sep 28, 2015 12 / 29

Table of contents




4 Robustness



Empirical specification – Difference-in-differences

The effect of the leverage cap in the difference-in-differences setup:

∆ log RealCreditit = αi + γt + β[DLEV ×DPostCrisis] + X ′itδ + εit

∆ log RealCreditit : Growth rate of real credit (domestic currency)DLEV : Leverage cap dummy (0/1); 1 if implemented prior to 2008DPostCrisis: Post-crisis dummy (0/1); 1 from 2009 onXit : Control variablesi: country, t: year, αi : country fixed effects, γt : time fixed effectsβ: Effect of leverage cap implemented prior to the crisis on post-crisiscredit growth; stabilizing role of leverage cap if β > 0Estimation period: 2002 to 2014; Frequency: yearly; Countries: 73emerging and advanced countries or less, 69 in baseline specification(depending on availability of controls)


Identifying assumptions

Common trends: Same pattern of real credit growth after the crisis inboth groups of countries if none of them had implemented theleverage cap prior to the crisis

I Assumption not testable but pre-treatment analysis can shed light onplausibility

The crisis has to be an exogenous event in the sense that it did notaffect implementation of leverage cap through anticipation effects

I Viewing the financial crisis as a generally unanticipated event appearsplausible

I Narrative evidence that leverage caps where implemented primarily formicroprudential reasons, i.e. to reduce leverage at the individual banklevel (Bordeleau et al. 2009, Lim et al. 2011)

I The caps on leverage might have been implemented for differentreasons such as a general preference for financial stability, quality ofinstitutions, etc.

I The difference-in-differences approach allows to control for suchcountry-specific characteristics even if they are unobservable


Test for pre-treatment differential effects

-10

010

2030

-2 -1 0Year since crisis

Coefficient 95% Confidence interval

Graph shows the difference-in-difference estimate over time. The crisis dummy is put to one already three years before the actualpost-crisis period, i.e. from 2006 on. Point estimates are surrounded by 95% confidence intervals.


The effect of the leverage cap on real credit growth

Dependent variable:

Real credit growth (%)(1) (2) (3)

DLEV x DPostCrisis 7.044** 6.659*** 6.029***

(2.966) (2.181) (2.123)

Real GDP growth (%) 1.568*** 1.638***

(0.236) (0.248)

Monetary policy rate (%) 0.197***

(0.073)

Country FE y y y

Year FE y y y

Countries 73 72 69

Observations 885 868 819

R2

0.28 0.37 0.38

Robust standard errors clustered by country given in parantheses. ∗∗∗,∗∗,∗ denotes significance at the 1, 5, and 10 percentlevel, respectively. All variables were winsorized at the 1% and 99% quantiles.


The influence of additional time-varying controlsDependent variable:

Real credit growth (%)(1) (2) (3) (4) (5) (6)

DLEV x DPostCrisis 6.029*** 5.943*** 6.178*** 5.373** 6.028** 5.792**

(2.123) (2.065) (2.129) (2.322) (2.382) (2.229)

Real GDP growth (%) 1.638*** 1.629*** 1.643*** 1.562*** 1.566***

(0.248) (0.248) (0.253) (0.242) (0.248)

Monetary policy rate (%) 0.197*** 0.297 0.197*** 0.153** 0.176**

(0.073) (0.190) (0.073) (0.076) (0.075)

Inflation rate (%) -0.148

(0.281)

Private credit-to-GDP (%) 0.012

(0.044)

MacroPru index (fin. sector) -1.808

(1.848)

MacroPru index (borrower) -2.054

(1.605)

Leverage cap (0/1) -4.585

(12.568)

Real GDP growth (lag, %) 1.144***

(0.208)

Monetary policy rate (lag, %) 0.284***

(0.048)

Country FE y y y y y y

Year FE y y y y y y

Countries 69 69 69 69 69 69

Observations 819 819 814 759 759 824

R2

0.38 0.38 0.38 0.37 0.37 0.34



The role of country-specific characteristics

Decision to implement leverage cap is likely to depend oncountry-specific characteristics

I If these characteristics also affect real credit growth, we have to controlfor them to get unbiased results

I This has been achieved by our specification which controls for country(and time) fixed effects

How do key (observable) country-specific variables affect creditgrowth?To answer this question, we use the correlated random effectsapproach, which allows both controlling for fixed effects and includingconstant country-specific variables (Mundlak 1978, Wooldridge 2010)In particular, we consider

I the credit-to-GDP gap (pre-crisis avg. 2002-07)I the capital and deposit-to-assets ratio (pre-crisis avg. 2002-07)I the dummy variable DLEV indicating implementation of the leverage

cap prior to the crisis


The role of country-specific characteristics – ResultsDependent variable:

Real credit growth (%)(1) (2) (3)

DLEV x DPostCrisis 6.029*** 6.147** 6.029***

(2.140) (2.329) (2.142)

Real GDP growth (%) 1.638*** 1.624*** 1.638***

(0.250) (0.278) (0.250)

Monetary policy rate (%) 0.197*** 0.202*** 0.197***

(0.073) (0.073) (0.073)

Avg. real GDP growth 2002-07 (%) 1.101** 1.050* 1.106**

(0.514) (0.619) (0.514)

Avg. monetary policy rate 2002-07 (%) 0.043 0.245 0.060

(0.363) (0.417) (0.363)

Avg. credit-to-GDP ratio 2002-07 (%) -0.076*** -0.078*** -0.076***

(0.017) (0.017) (0.017)

Avg. capital ratio 2006/07 (%) 0.260*

(0.154)

Avg. deposit ratio 2006/07 (%) 0.003

(0.039)

DLEV (0/1) -3.991

(7.062)

Country FE y y y

Year FE y y y

Countries 69 61 69

Observations 819 746 819

R2

0.54 0.54 0.54



Does the effect work through the pre-crisis capital ratio?

The stabilizing effect of the leverage cap might work through highercapital buffers built up by banks prior to the crisisBanks might draw on these buffers after the crisis to continue lendingTherefore, we expect the effect of leverage caps on real credit growthto be stronger for higher pre-crisis capital ratiosWe can test if the effect of the leverage cap on real credit growthworks through the pre-crisis capital ratioIn the empirical specification, we introduce an additional interactionof the the capital ratio with the interaction term of the post-crisis andleverage cap dummy:

∆ log RealCreditit = αi + γt + β1[DLEV ×DPostCrisis]+ β2[DPostCrisis × CapRatioi ]+ β3[DLEV ×DPostCrisis × CapRatioi ] + X ′

itδ + εit


Does the effect work through the pre-crisis capital ratio? –Results 1

Dependent variable:

Real credit growth (%)(1)

DLEV x DPostCrisis -1.059

(3.923)

DPostCrisis x Capital ratio -0.368

(0.349)

DLEV x DPostCrisis x Capital ratio 0.862**

(0.416)

Real GDP growth (%) 1.562***

(0.282)

Monetary policy rate (%) 0.195**

(0.081)

Country FE y

Year FE y

Countries 61

Observations 746

R2

0.61



Does the effect work through the pre-crisis capital ratio? –Results 2

-10

-50

510

15

0 2 4 6 8 10 12

Pre-crisis capital ratio (%)The figure shows the effect of the leverage cap on real credit growth (solid line, in percentage points) for different values of thepre-crisis capital ratio (vertical axis). The dashed lines show 95% confidence bands based on robust standard errors clustered bycountry. All variables were winsorized at the 1% and 99% quantiles.


Table of contents




4 Robustness



Robustness: Competing explanations

There are other possible explanations for the resultPre-crisis credit boom: pattern of real credit growth rates after crisisreflects standard adjustment to the stance of the financial cycle priorto the crisisSeverity of crisis: Countries which experienced a more severe crisis arelikely to have depressed credit growth rates afterwards


Robustness: Competing explanations – ResultsDependent variable:

Real credit growth (%)(1) (2)

DLEV x DPostCrisis 6.250*** 5.256**

(2.250) (2.371)

Pre-crisis credit boom x DPostCrisis 0.011

(0.032)

Severity of crisis x DPostCrisis -0.461

(0.433)

Real GDP growth (%) 1.633*** 1.541***

(0.245) (0.257)

Monetary policy rate (%) 0.193*** 0.205***

(0.070) (0.075)

Country FE y y

Year FE y y

Countries 69 69

Observations 819 819

R2 0.38 0.39

Pre-crisis credit boom: private credit-to-GDP prior to the crisis (average of 2006/07). Severity of crisis: GDP growth rate of2009 multiplied by minus one. Robust standard errors clustered by country given in parantheses. ∗∗∗,∗∗,∗ denotes significanceat the 1, 5, and 10 percent level, respectively. All variables were winsorized at the 1% and 99% quantiles.


Robustness: Subsample analysis

Role of other macroprudential instruments: include only countries inthe control group which did not implement any instrument prior tothe crisisRun analysis for sample of emerging market countries onlySymmetric window of four pre and post-crisis years: Run for years2005-12Clear-cut pre versus post-crisis period: Exclude years 2008/09Excluding countries from treatment group (see Appendix)


Robustness: Subsample analysis – Results

(1) (2) (3) (4)

No MacroPruEmerging

economies2005-12

Exclude

2008/09

DLEV x DPostCrisis 7.800* 7.759*** 6.191** 7.232***

(4.145) (2.598) (2.720) (2.353)

Real GDP growth (%) 1.223*** 1.664*** 1.625*** 1.823***

(0.257) (0.301) (0.227) (0.297)

Monetary policy rate (%) -1.013 0.211*** -0.194 0.231***

(0.741) (0.066) (0.177) (0.070)

Country FE y y y y

Year FE y y y y

Countries 24 44 69 69

Observations 275 529 521 689

R2

0.41 0.39 0.43 0.40

Dependent variable:

Real credit growth (%)



Concluding remarks

Evidence on stabilizing role of macroprudential policy gained bystudying the example of a cap on banks’ leverage

I The effect is significant in economic and statistical termsI It is robust with respect to various alternative specifications controlling

for different compositions of control groups and competing explanations

This stabilizing dimension should be incorporated in anycomprehensive cost-benefit analysis of macroprudential policyFurther research: Is there a stabilizing role of other macroprudentialpolicy instruments?


Appendix – The effect on total asset growth andcontributions by sectoral components

(1) (2) (3) (4) (5)

Total Assets Private Sector Non-residents Central bank Public sector

DLEV x DPostCrisis 4.559* 3.205** 1.367 0.565 -0.076

(2.476) (1.570) (0.854) (0.752) (0.507)

Real GDP growth (%) 1.016*** 0.690*** 0.118 0.111 -0.012

(0.272) (0.127) (0.074) (0.090) (0.009)

Monetary policy rate (%) 0.244*** 0.075** 0.101*** 0.095*** 0.001

(0.032) (0.033) (0.009) (0.016) (0.003)

Country FE y y y y y

Year FE y y y y y

Countries 66 66 66 66 65

Observations 761 761 761 739 748

R2

0.33 0.33 0.33 0.33 0.33

Dependent variable:

Change in claims subcategory

relative to total assets of

previous period

Column header indicates the sectoral component of total claims. Robust standard errors clustered by country given inparantheses. ∗∗∗,∗∗,∗ denotes significance at the 1, 5, and 10 percent level, respectively. All variables were winsorized at the1% and 99% quantiles.


Appendix – Robustness: Excluding countries fromtreatment group

(1) (2) (3) (4) (5) (6) (7) (8)

Canada Chile Ecuador Jordan St. Kitts and Nevis Paraguay Saudi Arabia United States

DLEV x DPostCrisis 6.030*** 6.026*** 6.339*** 5.420** 5.422** 5.012** 7.043*** 6.790***

(2.123) (2.123) (2.338) (2.271) (2.263) (2.084) (2.065) (2.200)

Real GDP growth (%) 1.636*** 1.640*** 1.641*** 1.648*** 1.700*** 1.627*** 1.642*** 1.638***

(0.248) (0.248) (0.250) (0.252) (0.251) (0.261) (0.251) (0.248)

Monetary policy rate (%) 0.198*** 0.198*** 0.198*** 0.197*** 0.198*** 0.207*** 0.197*** 0.197***

(0.0728) (0.0727) (0.0726) (0.0733) (0.0724) (0.0645) (0.0737) (0.0731)

Country FE y y y y y y y y

Year FE y y y y y y y y

Countries 68 68 68 68 68 68 68 68

Observations 812 815 806 806 806 806 806 806

R2

0.38 0.38 0.38 0.38 0.39 0.39 0.38 0.38

Dependent variable:

Real credit growth (%)

Column header indicates country excluded from estimation. Robust standard errors clustered by country given in parantheses.∗∗∗,∗∗,∗ denotes significance at the 1, 5, and 10 percent level, respectively. All variables were winsorized at the 1% and 99%quantiles.


manuel buchholz. caps on banks’ leverage and domestic credit after the crisis

Economy & Finance