Banks, Government Bonds and Default: What do the Data · PDF fileBanks, Government Bonds and Default: What do the Data Say? Nicola Gennaioli, Alberto Martin and Stefano Rossi ... Bankscope

Banks, Government Bonds and Default: What do theData Say?

Nicola Gennaioli, Alberto Martin and Stefano Rossi

Bocconi, CREI, and Purdue

December 13, 2016

GMR (Bocconi, CREI, and Purdue) Sovereign Debt and Risks December 13, 2016 1 / 32

Introduction

What are the costs of sovereign risk/defaults?I Recent emphasis on importance of banking channel

Center stage during recent European crisisI Europe’s troubled banks and broke governments are in a dangerous embrace(The Economist, 2011)

Different potential channels:I Balance sheet channel: banks hold substantial amounts of government bondsI Safety net channel: banks backed by government guaranteesI Macroeconomic channel: austerity policies might hurt economic activity andthus banks


This paper

Focus on balance sheet channelI Banks hold substantial amounts of domestic government bonds

F Liquidity (Bolton and Jeanne 2011, Gennaioli et al. 2014)F Risk-taking (Farhi and Tirole 2015)

I Default hurts balance sheets of banks → ⇓ in lending → ⇓ in output

Believed to have played major role in recent European crisis

Yet scant systematic evidence: existing factsI Aggregate (Gennaioli et al. 2014)I European: based on stress tests 2010-12 and syndicated lending (Popov andVan Horen (2014), De Marco (2016))

F Limited time period, small lending market


What we do

Analyze sovereign default-bank nexus using bank-level dataI Across many countries, periods and default episodes

Stylized facts (no causality)

Two key questions:I Which banks, in which countries, hold government bonds?

F Do they hold bonds all of the time or mostly during sovereign defaults?

I Do banks that hold more government bonds reduce their lending the mostduring defaults?


The Data

Bankscope database

Advantage: wide coverageI Characteristics of over 20.000 banks in 191 countries between 1998-2012

F Bondholdings at the bank level

I 20 default episodes: mostly in developing countries

Disadvantage:I Does not report nationality of bonds

F Domestic or foreign government bonds?


Main results

Negative, significant correlation between bank’s bondholdings duringsovereign default and subsequent lending

I $1 increase in bondholdings ⇔ $0.50 decrease in lendingI Result robust to controlling for country shocks and bank characteristics

Banks hold large amount of government bonds (around 9% of assets) innormal times

I In particular, banks that make fewer loans and are located in financiallyunderdeveloped countries

I Bondholdings increase slightly during default episodesF Especially in larger (and more profitable) banks

Findings consistent with “supply” channelI Banks hold substantial bonds in normal times (liquidity services)I Government defaults hurt banks and reduce lending

F Conceptually: result consistent with imperfect discrimination (Broner et al.2010)


Related literature

Costs of sovereign defaultsI Recent theories based on non-discrimination: collateral damage of defaults

F Basu (2010), Broner and Ventura (2011), Brutti (2011), Gennaioli et al.(2014), Mengus (2015), Farhi and Tirole (2015)

I Relationship between sovereign risk and private creditF Arteta and Hale (2008), Borensztein and Panizza (2008), Baskaya andKalemli-Ozcan (2016)

Relationship between bank lending and sovereign riskI Acharya and Steffen (2013), Brutti and Saure (2013), Popov and Van Horen(2014), De Marco (2016)

Demand for government bondsI Krishnamurthy and Vissing-Jorgensen (2012), Greenwood and Vayanos (2014)


Data sources: bank-level variables

Bank-level data from Bankscope dataset (Bureau van Dijk)I Information on broad range of bank characteristicsI Suitable for international comparisons because data is harmonized

Crucial: Bankscope reports banks’holdings of government bondsI However, no information on nationality of bondsI Use IMF/EU/Argentine data to validate information

Main sample: 7,391 banks in 160 countries, 36,449 bank-year observationsI Commercial banks (33.2%), cooperatives (38.2%), savings banks (20.6%),investment banks (1.6%)

I Sample construction: filter outF duplicate recordsF banks with total assets <$100KF years < 1997F banks without two consecutive years of data


Data sources: aggregate variables

Macroeconomic conditions:I IMF’s International Financial Statistics (IFS) and WB’s World DevelopmentIndicators (WDI)

I Proxy financial development with private credit to GDP

Sovereign default:I Dummy variable based on Standard and Poor’s

F Default is failure to meet principal or interest payment in original termsF Greek bond swap of 2012: defaultF 19 sovereign defaults in 16 countries

I complement analysis with alternative measure of defaultF “haircuts” (Cruces and Trebesch 2013)F S&P measure “augmented” with spreads


Data sources: aggregate variables

Supplement Bankscope data with realized and expected bond returns

Realized returns:I JP Morgan’s Emerging Market Bond Index (EMBIG+) for emerging countries,and on JP Morgan’s Global Bond Index (GBI) for developed countries

I Kim (2010) and Levy-Yeyati, Martinez-Peria, and Schmukler (2010)

Expected returns: problematic, because not observedI construct it through two-step processI first stage: regress returns on country-specific economic, financial and politicalrisk factors

Rc ,t = γt + β0 + β1Zc ,t−1 + uc ,t

whereF Rc ,t is realized returns of public bonds in country c at time tF γt are time-dummies, which capture variations in global risk-free rateF Zc ,t−1 are risk ratings compiled by ICRG, shown to negatively predict returns(e.g. Comelli 2012)

I second stage: define expected returns as fitted values of this first-stageregression


Sample of sovereign defaults

Country Default S&P Haircut No BankYears No Banks

Argentina 20012004 76.8% 231 87Ecuador 19982000 38.3% 32 30Ecuador 2009 67.7% 8 8Ethiopia 19981999 92.0% 2 1Greece 20112012 64.8% 12 6Guyana 19982004 91.0% 20 3Honduras 19982004 82.0% 79 21Ireland 7 7Indonesia 19982000; 2002 17 13Jamaica 2010 5 5Kenya 19982004 45.7% 160 33Nigeria 2002 41 41Portugal 12 12Russia 19982000 51.1% 40 31Serbia 19982004 70.9% 2 2Seychelles 20002002; 2010 56.2% 1 1Sudan 19982004 2 1Ukraine 19982000 14.8% 14 7Venezuela 34 26Zimbabwe 20002004 6 3No Banks 725 338No Countries 16 11No Episodes 19 12


Data on bondholdings: is it reliable?

Compare with IMF aggregate dataI Financial Institutions’Net Claims to the Government(IFS)

Average measures look quite similar


Data on bondholdings: is it reliable?

Compare with bank-level dataI European stress tests (2010, 2011, 2012)I Data on domestic bondholdings from Argentine Central Bank (2002-2004)

Table I – Bank’s Holdings of Government Bonds from BANKSCOPE and Other Sources

The table reports summary statistics of bank bondholdings as a percentage of total assets for selected samples.

Sample EU Banks GIIPS Banks Argentine BanksSource BANKSCOPE Stress Test BANKSCOPE Stress Test BANKSCOPE Central Bank

Mean 8.16 5.12 9.43 6.22 14.23 11.34Median 7.68 4.44 8.22 5.64 10.73 8.09Correlation 0.69 0.76 0.77

Sample Period 20102012 20102012 19972004No Obs. 126 65 589No Banks 66 33 142


Descriptive statistics: bondholdings


Descriptive statistics: bank characteristics

Panel A – Bankscope, Constantcontinuing sample

Mean Median Std Deviation No Countries No Observations

Assets ($/M) 9,922.0 725.6 81,400.0 160 36,449Noncash assets 95.8 97.6 5.6 160 36,449Leverage 91.0 93.3 8.4 160 36,449Loans 57.1 60.0 17.0 160 36,449Profitability 0.9 0.7 2.1 160 36,449Exposure to Central Bank 3.3 1.5 4.9 160 36,449Interbank Balances 12.2 9.2 12.5 160 36,449Government Owned 2.5 0.0 15.7 160 36,449


Descriptive statistics: sovereign bond returns

Figure 2. Sovereign Bond Prices in Defaulting Countries. The figure plots the average bond prices over 7 default episodesin 6 countries (Argentina 20012004, Russia 19982000, Cote d’Ivoire 20002004, Ecuador 19982000, Ecuador 2009,Nigeria 2002, Greece 2012), from day 1,000 to +1,000, whereby day 0 is the day in which default is announced.


Government bonds and lending

Do bondholdings matter for lending?

Letting Λi ,c ,t denote change in loans/assets for bank i , in country c ,between time t − 1 and t, we run the following regression:

Λi ,c ,t = γ0 + γ1 · Bi ,c ,t−1 + γ2 ·Defc ,t−1 ++γ3 ·Defc ,t−1 · Bi ,c ,t−1 + γ4 · Xi ,c ,t−1 ++γ5 ·Defc ,t−1 · Xc ,t−1 + γ6 · Xc ,t−1 + γ7 ·Defc ,t−1 · Xc ,t−1+µi ,c ,t

where:I Bi ,c ,t−1 is the bond/asset ratio for bank i , c and Defc ,t−1 is a default dummyI Xi ,c ,t−1 and Xc ,t−1 are vectors of bank and country level characteristics

F Bank variables: loans outstanding, non-cash assets, exposure to central bank,interbank balances, government ownership

F Country variables: financial development (private credit/GDP and bankingcrises), growth, inflation, expected bond returns



Letting Λi ,c ,t denote change in loans/assets for bank i , in country c ,between time t − 1 and t, we run the following regression:

Λi ,c ,t = γ0 + γ1 · Bi ,c ,t−1 + γ2 ·Defc ,t−1 ++γ3 ·Defc ,t−1 · Bi ,c ,t−1 + γ4 · Xi ,c ,t−1 ++γ5 ·Defc ,t−1 · Xc ,t−1 + γ6 · Xc ,t−1 + γ7 ·Defc ,t−1 · Xc ,t−1+µi ,c ,t

Main focus: γ3I γ1 captures average effect of bondholdings on bank loansI γ3 captures differential effect of bondholdings on loans during default

γ3 < 0: consistent with hypothesis that, during defaults, bank lending isaffected through bondholdings


Table IV – Bondholdings, Sovereign Default, and Changes in LoansThe table presents coefficient estimates from pooled OLS regressions. Standard errors (in parentheses below the coefficient estimates) are adjustedfor heteroskedasticity using the Huber (1967) and White (1980) correction, as well as for clustering at the bank level using the Huber (1967)correction. *** indicates significance at the 1% level; ** indicates significance at the 5% level; * indicates significance at the 10% level.

(1) (2) (3) (4) (5)Bank Bondholdingsi,c,t–1 * 0.126** 0.129** 0.096* 0.148** 0.133***

Sovereign Defaultc,t–1 (0.057) (0.057) (0.058) (0.060) (0.045)Sovereign Bond Returnc,t–1 * 0.072*** 0.068*** 0.071***Sovereign Defaultc,t–1 (0.014) (0.015) (0.015)Bank Bondholdingsi,c,t–1 0.032*** 0.034*** 0.009 0.009 0.018**

(0.009) (0.009) (0.011) (0.011) (0.008)Sovereign Defaultc,t–1 0.038 0.035 0.019

(0.026) (0.025) (0.024)Sovereign Bond Returnc,t–1 0.005 0.011* 0.004

(0.005) (0.006) (0.007)Leveragei,c,t–1 * 0.115** 0.107** 0.084 0.035 0.028

Sovereign Defaultc,t–1 (0.054) (0.054) (0.053) (0.057) (0.048)Loansi,c,t–1 * 0.180*** 0.189*** 0.169*** 0.202*** 0.189***

Sovereign Defaultc,t–1 (0.050) (0.050) (0.049) (0.054) (0.041)Bank Sizei,c,t–1 0.001*** 0.001*** 0.001*** 0.001*** 0.001***

(0.000) (0.000) (0.000) (0.000) (0.000)Noncash assetsi,c,t–1 0.029* 0.032* 0.025 0.011 0.021

(0.017) (0.017) (0.020) (0.020) (0.016)Loansi,c,t–1 0.043*** 0.041*** 0.061*** 0.054*** 0.049***

(0.005) (0.005) (0.005) (0.005) (0.004)Profitabilityi,c,t–1 0.083 0.089 0.078 0.094* 0.087**

(0.060) (0.060) (0.056) (0.053) (0.042)Exposure to Central Banki,c,t–1 0.006 0.005 0.072*** 0.048** 0.047***

(0.019) (0.019) (0.024) (0.023) (0.016)Interbank Balancesi,c,t–1 0.016** 0.019*** 0.004 0.010 0.004

(0.007) (0.007) (0.007) (0.007) (0.005)(0.004) (0.004) (0.004) (0.004) (0.003)

Year Dummies? Yes Yes Yes YesCountry Dummies? Yes Yes YesCountry x Year Dummies? Yes YesConstant 0.041** 0.033* 0.026 0.184 0.078

(0.018) (0.018) (0.021) (102.387) (177.873)No Observations 14,074 14,074 14,074 14,074 27,408No Banks 3,722 3,722 3,722 3,722 5,218No Countries 60 60 60 60 158Rsquared 0.061 0.072 0.106 0.204 0.224



Main result and robustness

Effect of bondholdings on lending quantitatively largeI $1 increase in bonds held by average bank → $0.50 decline in lending

Result is robust to:I Alternative definitions of lending

Li ,c ,t − Li ,c ,t−1Ai ,c ,t

and ∆ log(Li ,c ,t )

I Exclusion of government-owned banksI Exclusion of small countries (GDP per capita and size of default)I Exclusion of countries with few banksI Alternative definitions of default

F Haircut measure of Cruces and Trebesch (2013)F S&P dummy “augmented” with high spreads



Previous results come from all observationsI Bonds held at the time of defaultI Bonds accumulated during defaultI What drives the results?

Focus on bonds at time of default

Analyze change in lending in two years post-default and:I Bonds held in pre-default yearI Average bonds held in three years prior to default


Government bonds and lendingTable VII – Bondholdings and Changes in Loans: Normal Times v Default Years Bonds

The table presents coefficient estimates from pooled OLS regressions. The dependent variable changes in loans is computed as loans outstanding inyear t minus loans outstanding in year t1, divided by total assets. The main independent variables are predefault bank bondholdings, coimputedas bondholdings in the year prior to the first year of a sovereign default, divided by total assets; average predefault bank bondholdings, computedas the average of bondholdings divided by total assets in the last three years prior to the first year of a sovereign default; bank average nondefaultyears bondholdings, computed as the average of bank bondholdings in all the nondefault years prior to and including year t–1 , bank timevaryingbondholdings, computed as bank bondholdings minus bank average nondefault years bondholdings. Standard errors (in parentheses below thecoefficient estimates) are adjusted for heteroskedasticity using the Huber (1967) and White (1980) correction, as well as for clustering at the banklevel using the Huber (1967) correction. *** indicates significance at the 1% level; ** indicates significance at the 5% level; * indicates significance atthe 10% level.

(1) (2)

PreDefault Bank Bondholdings 0.281***(0.080)

Avg PreDefault Bank Bondholdings 0.361***(0.028)

Sovereign Bond Returnt–1 *

BankLevel Controls and Interactions? Yes YesYear Dummies? Yes YesCountry Dummies? Yes YesCountry x Year Dummies?

Constant 0.780** 0.874**(0.275) (0.272)

No Observations 105 105No Banks 105 105No Countries 5 5Rsquared 0.439 0.442



What can we conclude?I In default, significant relationship between bondholdings and lending

Consistent with recent findings for the European crisis:I Banks’costs of funds correlated with their bondholdings during crisis

F Significant, negative correlation between bank stock price/CDS and countryCDS

F Correlation stronger for banks with higher bondholdings

I Bank (syndicated) lending correlated with bondholdings during crisisF Banks in core countries with differential exposure to periphery debt (Popov andvan Horen, forthcoming)

F De Marco (2016) finds similar evidence


Determinants of bondholdings

We know:I Banks hold substantial amount of bonds in non-default yearsI Banks hold slightly more bonds in default years

Delving deeper: letting Bi ,c ,t denote bonds/assets held by bank i , in countryc , at time t, we run

Bi ,c ,t = α0 + α1 · Xi ,c ,t−1 + α2 · Xc ,t−1 + α3 ·Defc ,t−1 +α4 ·Defc ,t−1 · Xi ,c ,t−1 + α5 ·Defc ,t−1 · Xc ,t−1 + εi ,c ,t

where:I Defc ,t−1 is a default dummyI Xi ,c ,t−1 and Xc ,t−1 are vectors of bank and country level characteristics

F Bank variables: loans outstanding, non-cash assets, exposure to central bank,interbank balances, government ownership

F Country variables: financial development (private credit/GDP and bankingcrises), growth, inflation, expected bond returns


(1) (2) (3)Sovereign Defaultt–1* 0.009*** 0.011*** 0.007***

Sizet–1 (0.003) (0.003) (0.003)Sovereign Defaultt–1* 0.013 0.042 0.041

Loanst–1 (0.032) (0.037) (0.029)Sovereign Defaultt–1 * 0.107***

Expected Sovereign Bond Returnt–1 (0.029)Sovereign Defaultt–1 * 0.027 1.758***

GDP Growtht–1 (0.170) (0.432)Sovereign Defaultt–1 * 0.035* 0.172***

Banking Crisist–1 (0.021) (0.045)Sovereign Defaultt–1 * 0.448* 2.048***

Private Creditt–1 (0.230) (0.438)Sovereign Defaultt–1 0.123 1.501*** 0.091*

(0.158) (0.344) (0.055)Sizet–1 0.001*** 0.001 0.000

(0.000) (0.000) (0.000)Loanst–1 0.027*** 0.047*** 0.041***

(0.004) (0.007) (0.004)Expected Sovereign Bond Returnt–1 0.027***

(0.008)GDP Growtht–1 0.164** 0.134

(0.066) (0.096)Banking Crisist–1 0.030*** 0.022

(0.005) (0.019)Private Creditt–1 0.021*** 0.038**

(0.004) (0.018)

Other controls? Yes Yes YesYear Dummies? Yes Yes YesCountry Dummies? Yes YesCountry x Year Dummies? Yes

No Observations 13,082 5,341 26,549No Banks 2,896 2,103 5,124No Countries 38 29 157Rsquared 0.801 0.739 0.814


Bondholdings: main results

In non-default years: bondholdingsI Negatively correlated with lendingI Negatively correlated with financial development

In default years: bondholdings, increaseI Especially in large (and profitable) banksI Especially in countries with higher financial development

Consistent with following narrative:I In non-default years, bonds held for liquidity provision

F Banks without good investment opportunities or alternative assets

I In default years, subset of banks increases bondholdingsF Average increase slightF No indication that it is concentrated in “bad” banks


Conclusion

Evidence from 19 default episodes in 16 countries between 1998 and 2012

Main findings:I Strong negative correlation between bondholdings and subsequent lending

F $1 increase in bondholdings → $.50 fall in loans

I Banks hold large amount of bonds in non-default yearsF Especially banks with fewer loans, in less financially developed countriesF During default, bondholdings increase slightly: particularly in large banks

What do we learn?I Link between sovereign default and bank lending seems to be prevalent

F Consistent non-discrimination view (Broner et al. 2010, etc..)

I Differences between emerging and advanced economiesF Bonholdings in emerging markets higher than in OECD (12.7% vs. 5%)F Implications for regulation (e.g. risk weights)


Panel A – Bankscope, Constantcontinuing sample

Mean Median Std Deviation No Countries No Observations

Assets ($/M) 9,922.0 725.6 81,400.0 160 36,449Noncash assets 95.8 97.6 5.6 160 36,449Leverage 91.0 93.3 8.4 160 36,449Loans 57.1 60.0 17.0 160 36,449Profitability 0.9 0.7 2.1 160 36,449Exposure to Central Bank 3.3 1.5 4.9 160 36,449Interbank Balances 12.2 9.2 12.5 160 36,449Government Owned 2.5 0.0 15.7 160 36,449


Bonds Bank Size NoncashAssets Leverage Loans Profitability Exposure Balances

Banks size 0.063***

Noncash assets 0.835*** 0.202***

Leverage 0.141*** 0.335*** 0.207***

Loans 0.376*** 0.016*** 0.202*** 0.238***

Profitability 0.102*** 0.059*** 0.071*** 0.286*** 0.100***

Exposure to Central Bank 0.096*** 0.209*** 0.374*** 0.218*** 0.231*** 0.140***

Interbank Balances 0.136*** 0.087*** 0.117*** 0.173*** 0.553*** 0.061*** 0.367***

Government Owned 0.082*** 0.141*** 0.026*** 0.031*** 0.073*** 0.009*** 0.027*** 0.022***


Table V – Bondholdings, Country Shocks, and Changes in Loans(1) (2) (3) (4)

Bank Bondholdingst–1 * 0.144** 0.117** 0.131* 0.107*Sovereign Defaultt–1 (0.062) (0.047) (0.068) (0.064)

Bank Bondholdingst–1 * 0.156 0.285**GDP Growtht–1 (0.140) (0.137)

Bank Bondholdingst–1 * 0.027 0.025Exchange Rate Devaluationt–1 (0.040) (0.039)

Sovereign Bond Return * 0.091 0.010Sovereign Defaultt–1 (0.077) (0.059)

Bank Bondholdingst–1 0.001 0.003 0.008 0.013(0.012) (0.009) (0.011) (0.008)

BankLevel Controls and Interactions Yes Yes Yes Yeswith Sovereign Default?

BankLevel Controls and Interactions Yes Yeswith GDP Growth?

BankLevel Controls and Interactions Yes Yeswith Exchange Rate Devaluation?

Year Dummies? Yes Yes Yes YesCountry Dummies? Yes Yes Yes YesCountry * Year Dummies? Yes Yes Yes Yes

Constant 0.229 0.040 0.118 0.141(0.147) (3.715) (0.087) (130.540)

No Observations 13,873 26,467 13,908 24,982No Banks 3,649 4,967 3,646 4,645No Countries 56 129 54 97Rsquared 0.205 0.214 0.205 0.204


Documents

Banks, Government Bonds and Default: What do the Data · PDF fileBanks, Government Bonds and Default: What do the Data Say? Nicola Gennaioli, Alberto Martin and Stefano Rossi ... Bankscope