Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein,...

65
May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit 1 Bank dominance: Financial sector determinants of sovereign risk premia May 13, 2013 Mohit Thukral 1 Department of Economics Stanford University Stanford, CA 94305 Written under the direction of Darrell Duffie Abstract This paper investigates the persistence, extent and nature of the correlation between banking and sovereign credit risk in the European financial crisis. I use Credit Default Swap (CDS) premia on sovereigns as a measure of sovereign risk while using a specially constructed CDS Banking Risk Index (BRI) to measure the financial sector risk of each country. This paper finds that there is “bank dominance” in the determination of sovereign risk premia in the European financial crisis even when fiscal variables are included in the estimation. The Banking Risk Index is the primary statistically significant determinant of sovereign risk premia in the crisis period. Moreover, this paper finds that the correlation between banking and sovereign risk increases with an increase in the financial sector risk, measured by the BRI—in times of greater financial sector risk, banking and sovereign risk are jointly evaluated in the market, even more so than usual. 1 This article was written under the supervision of Darrell Duffie, whose guidance has been invaluable in all aspects of this project. I am incredibly grateful to him for his mentorship and support. I am also grateful to John Taylor, Paul Milgrom, Monika Piazzesi, Marcelo Clerici-Arias, Maxwell Wernecke, Geoffrey Rothwell, Tanya Beder, Ian Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments and suggestions over the course of many conversations. This project is dedicated to my parents, who have been a source of inspiration. All omissions and errors are those of the author alone.

Transcript of Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein,...

Page 1: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    1  

Bank dominance: Financial sector determinants of sovereign risk premia

May 13, 2013

Mohit Thukral1

Department of Economics Stanford University

Stanford, CA 94305

Written under the direction of Darrell Duffie

Abstract This paper investigates the persistence, extent and nature of the correlation between banking and sovereign credit risk in the European financial crisis. I use Credit Default Swap (CDS) premia on sovereigns as a measure of sovereign risk while using a specially constructed CDS Banking Risk Index (BRI) to measure the financial sector risk of each country. This paper finds that there is “bank dominance” in the determination of sovereign risk premia in the European financial crisis even when fiscal variables are included in the estimation. The Banking Risk Index is the primary statistically significant determinant of sovereign risk premia in the crisis period. Moreover, this paper finds that the correlation between banking and sovereign risk increases with an increase in the financial sector risk, measured by the BRI—in times of greater financial sector risk, banking and sovereign risk are jointly evaluated in the market, even more so than usual.

                                                                                                               1 This article was written under the supervision of Darrell Duffie, whose guidance has been invaluable in all aspects of this project. I am incredibly grateful to him for his mentorship and support. I am also grateful to John Taylor, Paul Milgrom, Monika Piazzesi, Marcelo Clerici-Arias, Maxwell Wernecke, Geoffrey Rothwell, Tanya Beder, Ian Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments and suggestions over the course of many conversations. This project is dedicated to my parents, who have been a source of inspiration. All omissions and errors are those of the author alone.

Page 2: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    2  

Section I: Introduction

Market assessments of credit risk have undergone a remarkable transformation since the

beginning of the financial crisis. This re-assessment of credit risk is especially stark in the market

for sovereign credit risk in the European Union. In 2007, the sovereign debt of fiscally unhealthy

Eurozone countries like Italy and Greece yielded interest rates similar to those of German or

French debt. For example, on 01/01/2007, the spread of Spanish 10-year bond over the 10-year

German Bund was less than 30 basis points (0.3%). Findings that internal factors (like fiscal

health) were inconsequential in determining sovereign bond yields were hailed as a mark of the

success of the monetary union. Perhaps even more surprising than the integration of sovereign

credit markets across Europe was the integration of banking credit markets across the monetary

union. At the beginning of 2007, the difference between the average CDS risk premia of Spanish

and German banks was about 40 basis points.

In the post financial crisis world, the credit spreads between the “periphery” and the

“core” (in fact, the “core” of the Eurozone is now arguably restricted to only Germany and

France) of Eurozone have widened significantly. In July 2010, the spread between Spanish and

German 10-year bond yields reached 250 basis points. Similarly, the spread between the average

financial sector CDS of the 2 countries also reached 260 basis points. Clearly, there was a re-

assessment of credit risk across various markets for these countries, including sovereign and

banking sector credit spreads.

The changes in credit risk conditions were also accompanied by increasing inter-

dependence between the markets for sovereign and credit risk. During the crisis period and

especially after the nationalization of Anglo-Irish bank in January 2009, each of the respective

financial sectors of European countries has come to represent a contingent liability for its

Page 3: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    3  

sovereign. For example, measures for extensive bank support in the United Kingdom could

represente 44% of the country’s GDP putting a large potential fiscal burden on government

finances (Panetta et al., 2009). The worsening of sovereign creditworthiness assessments also

adversely affected bank funding conditions, bank asset holdings and the value and probability of

potential bailouts. The anecdotal evidence above raises questions about the relationship between

banking and sovereign risk and about the predictors of the high credit spreads in the Eurozone.

The purpose of this paper is two-fold: first, to provide an explanation for changes in

sovereign credit risk in the European financial crisis and second to study the relationship

between sovereign and banking risk in the same period. Three hypotheses are developed for this

paper:

• The co-movement hypothesis: Changes in banking and sovereign risk in the European

financial crisis were correlated with each other.

• The Bank dominance hypothesis: Even after controlling for fiscal factors like the ratio

of debt-to-GDP and current account balance, banking sector risk variables remain the

primary statistically significant predictor of levels of sovereign credit risk levels.

• The joined-at-the-hip hypothesis: The correlation between banking and sovereign risk

increases with an increase in the levels of banking and/or sovereign credit risk.

To test the hypotheses outlined above, this paper uses data from the period between

10/13/2008 and 1/1/2013. Sovereign risk is measured in terms of 5-year Euro traded CDS risk

premia on the sovereigns. Figure 1 represents the mean and standard deviation of sovereign and

banking sector CDS risk premia across 4 countries in the Eurozone: France, Germany, Spain and

Italy. Clearly, the worsening of the credit conditions for sovereign debt is observable in the

increase in mean sovereign risk premium. The estimation of financial sector risk however, is

Page 4: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    4  

challenging—individual bank

CDS is vulnerable to

exogenous factors whereas

indices of stocks and bonds

make a direct comparison with

sovereign credit risk

impossible. A specially

devised Banking Risk Index

(BRI) is constructed for the

purpose of this paper. This index comprises an asset-weighted average of the CDS for 5 large

banks domiciled in each of the sovereigns. These are the largest banks, by assets, for which CDS

risk premia were available continuously throughout the sample period. Note that this index

makes financial sector risk comparable across countries, which enables us to make an apples-to-

apples comparison among the data.

Three separate specifications are developed to test the three hypotheses outlined above.

First, the changes in sovereign risk premia are regressed on changes in the banking risk index to

determine if the two variables co-move. Weekly averages of the CDS data are used for this

specification so as to ensure that data points have sufficient volume beneath them; daily quotes

may be vulnerable to lack of market demand. Various control variables for global financial

conditions and country specific factors are added. The first specification finds that there is a

statistically significant relationship between first differences of banking and sovereign risk. In

fact, a 1 basis point change in the BRI predicts a more than 1 basis point change in the sovereign

010

020

030

040

0 C

DS

risk

prem

ia0 50 100 150 200

Week

Mean of sovereign CDS Standard dev. of sovereign CDSMean of Banking risk Index Standard dev. of BRI

Figure 1: Mean & Standard dev. of CDS premia

Page 5: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    5  

CDS rate. In predicting sovereign CDS changes, lagged values of the BRI have a negative

coefficient, likely indicating some degree of market inefficiency or stale data.

The second specification estimates the factors determining the levels of sovereign risk

premia. This specification is developed to estimate whether financial sector factors or fiscal

factors are dominant in determining the sovereign CDS risk premia. Given constraints around the

availability of fiscal data, this specification is estimated using monthly averages rather than

weekly averages. Lagged values of current account balances and outstanding securities-to-GDP

ratios are used as variables to measure fiscal factors. Variables used to estimate banking and

sovereign risk remain the same as above. Therefore, the regression for this hypothesis uses the

level of sovereign risk premium as the dependent variable while using the Banking Risk Index

and the fiscal variables as the independent variables.

This specification provides the most salient results in this paper: banking factors are

dominant over fiscal factors in determining sovereign risk premia in the European financial

crisis. Both financial and fiscal factors remain statistically significant but the economic

magnitudes of the fiscal factors render them negligible determinants of sovereign risk premia. A

1 standard deviation increase in the current account balance leads to only a 0.7 basis points

predicted decrease in the sovereign CDS premium. On the other hand, even a small 1 basis point

increase in the Banking Risk Index contemporaneously leads to a 1.22 basis points increase in

the sovereign CDS rates (a 1 standard deviation increase leads to an 80 basis point increase).

Similar results are seen when economic magnitudes for the outstanding securities to GDP

variable are considered. Overall, this specification confirms the “Bank dominance” hypothesis,

that is, variables measuring financial sector risk are the primary statistically significant

determinant of sovereign CDS risk premia.

Page 6: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    6  

The salience of these results lies in its disagreement with the popular “fiscal dominance”

hypothesis that states that fiscal factors in the Eurozone, like debt-to-GDP ratios and current

account balances, are driving the high levels of sovereign credit risk premia (see Greenlaw,

Hamilton, Hooper and Mishkin for a detailed explanation). The results included here show that

when banking sector risk factors (like the BRI) are included, the high levels of sovereign credit

risk are predicted by banking sector factors rather than fiscal factors. This dominance of banking

sector risk factors and relative insignificance of fiscal factors might hold important lessons for

the management of the European financial crisis; the path to stability and prosperity might lie

through stable banks.

The third, “joined-at-the-hip” specification revolves around the correlation between

banking and sovereign risk. The trailing 6-month correlation between the BRI and the sovereign

CDS premium is used to measure the correlation between banking and sovereign risk. A linear-

log specification is used based on economic theory as well as graphical representations of the

relationship between the correlation coefficient and the BRI. The dependent variable is the

correlation coefficient mentioned above while the independent variable of interest is the log of

the BRI. The results of this specification do not reject the hypothesis that the correlation between

banking and sovereign risk increases in times of greater financial sector risk. It should be noted,

however, that the regressions around the third specification have low explanatory power, with R2

values of around 20%. Clearly, other factors determining the correlation between financial sector

and sovereign risk remain important but un-estimated.

Overall, the major conclusion of this paper is that financial sector risk is the primary

statistically significant determinant of sovereign risk premia. Measures of financial and

sovereign risk co-move in the financial crisis period and the correlation between banking and

Page 7: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    7  

sovereign risk is heightened in times of greater financial sector risk. The rest of the paper is

organized as followed: Section II reviews the relevant literature in this field. Section III provides

brief descriptions of the various hypotheses estimated in this paper. Section IV provides details

on the methodology and results for each of the hypotheses outlined in the previous section.

Section V concludes.

Section II. Literature review

The literature on the link between sovereign and banking risk can broadly be divided into

four general categories: papers showing the inter-dependence or co-movement (or lack thereof)

of sovereign and banking risk, those identifying sovereign determinants of banking risk, those

identifying financial-sector determinants of sovereign risk and papers that look into extent of this

relationship under various conditions. This paper first provides a brief analysis of the existing

literature on this subject while also establishing where this contribution fits into the existing

landscape.

Category 1: The co-movement of sovereign and banking risk

The correlation and co-movement between sovereign credit risk and banking sector risk

during the European financial crisis is described in this first category. Mody and Sandri (2011)

show that the health of the national banking sector and other domestic factors (including fiscal

factors), were an important determinant of sovereign credit risk during the crisis period (Mody

and Sandri, 2011). In this case, the authors regress the sovereign bond spread on the normalized

ratio of financial sector equity to the overall equity index of the country, in the form of the

following equation2:

                                                                                                               2 For a full description of the methodology, see Mody and Sandri (2011). The above equation is provided for reference to one part of the authors’ methodology and specification and does not seek to represent the whole of it.

Page 8: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    8  

Wherein:

• ∆ represents first differences

• Spread represents the difference the spread of the sovereign’s 10-year bond yield over the

10-year German Bund

• Bank Equity(F) represents the ratio of the equity of the country’s financial sector divided

by the overall equity index. Normalized to 100 at the beginning of the period.

• USTyields represents the yields on US government bonds. Proxy for a “flight to quality”.

Potential simultaneity with Bunds not discussed.

• US_Bank_CDS represents an index of CDS yields for US banks. Used as a proxy for

global financial conditions specific to the bond/credit markets.

The above measures selected by Mody and Sandri to estimate sovereign and banking risk

highlights some of the perils involved in making such a selection. A decrease in the financial

equity index signifies that bank equity for a particular company is performing worse than the

overall equity markets. If there is a corresponding decrease in financial equities as well as the

overall equity index, the financial equity index will remain unchanged. Therefore the use of this

ratio excludes some general stock market changes. However, a regression that explains changes

in sovereign spreads by using an equity index, inherently runs into the problem of comparing

apples and oranges. There is a significant body of literature (see Campbell and Ammer, 1991 for

example) that points to the differences between factors affecting the bond and equity markets. By

regressing a bond market measure of sovereign risk onto an equity market measure of financial

sector risk, the authors pre-suppose that the inherent differences between the 2 measures will not

ΔSpreadi,t = βs=0

p

∑ ΔSpreadi,t−s + λs=0

m

∑ ΔBankEquity i,t−s+ Φs=0

n

∑ D.USTyieldsi,t−s +

ϕs=0

p

∑ D.US _Bank _CDSi,t−s +Country_Dummyi + Period _Dummyt + ε i,t

Page 9: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    9  

confound the estimated results. Gray (2009) points out that the extent and nature of the linkages

between sovereign and banking risk are often mis-estimated partly due to the difficulty in

accurately estimating both the independent and the dependent variables.

Mody and Sandri (2011) finds that after the collapse of Bear Sterns, there was a

statistically significant correlation between changes in sovereign spreads and the lagged financial

equity index. Simply put, the authors argue that during the financial crisis, changes in sovereign

risk were correlated with preceding changes in the health of the financial sector of the country.

These results hold for the period between the collapse of Bear Sterns and that of Anglo Irish.

After the Anglo Irish credit event, the financial equity index became contemporaneously

correlated with the sovereign bond spread.

Regardless of the problems with the estimation methodology, Mody and Sandri (2011)

show that there is a statistically significant link between measures of sovereign and banking risk.

This result is confirmed by other papers including Mody (2009), Demirguc-Kunt and Huizinga

(2010) and Sgherri and Zoli (2009). To one extent or another, these papers posit that there is a

statistically significant correlation between various distinct measures of sovereign credit risk and

banking risk3, respectively. This correlation between banking and sovereign credit risk described

above does not imply a time-invariable or stationary relationship; the hypothesis of correlation

between banking and sovereign risk in advanced economies is rejected in the periods prior to the

financial crisis. Pagano and von Thadden (2004) study the European bond markets from 2000-

2004, that is, in the period after the monetary union. This study discovers that in the sample

period, sovereign and private sector bond markets had become “integrated” across countries

                                                                                                               3 Note that the papers mentioned here vary significantly in their choice of measures for estimating sovereign and banking risk. Demirguc-Kunt and Huizinga (2010), for example, focus somewhat on the size of the banking sector. Having said that, all papers highlighted above find that there was a statistically significant correlation between banking and sovereign risk measures.

Page 10: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    10  

regardless of internal factors. This means that the sovereign bond market was not materially

affected by the fiscal health of the sovereign or by other country-specific factors like the health

of the country’s financial sector. This result is extrapolated to mean that there is no link between

internal factors such as the health of the financial sector and the country’s public finance system

and the performance of the sovereign in the sovereign debt market (Pagano and van Thadden,

2004). This strand of literature also suggests that the extent of the correlation between financial

and sovereign risk may be a function of the levels of these two individual measures—that is, in

times when sovereign risk high, the link or the feedback loop between sovereign and banking

risk is particularly strong4.

Overall, this category of literature points to a strong and statistically significant

relationship between sovereign and banking sector risk in the financial crisis and post-financial

crisis period.

Category 2: Sovereign determinants of banking risk

This strand of literature identifies four primary mechanisms through which a decrease in

the creditworthiness of the domestic sovereign (and of other related sovereigns) can adversely

affect the health of the financial system in the country (and other countries) (Bank of

International Settlements, 2011). Papers explaining and showing each of these four mechanisms

are outlined below.

                                                                                                               4 Monfort and Renne (2011) shows that changes in euro area sovereign yield spreads were driven by liquidity variables during the financial crisis period. They argue that after subtracting liquidity pricing effects from the sovereign yield spreads, the “actual default probabilities” (not driven by liquidity effects) are significantly lower that the risk neutral default probabilities (which include liquidity as well as credit risk). Therefore, adverse liquidity shock, which obviously influence bank liquidity and health also influence sovereign credit risk thereby representing increased correlation between the two.

Page 11: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    11  

A. Banks’ holdings of sovereign debt

Firstly, banks’ holdings of sovereign debt can cause losses for the banks’ assets in cases

where the market value of the sovereign debt decreases. Simply put, since bank hold some of

their assets in the form of sovereign debt, a decrease in the values of sovereign debt causes losses

(and therefore, increases riskiness) for the banks. Thus, a decrease in the price of sovereign

bonds (possibly due to worsening creditworthiness of the sovereign) leads to losses for banks and

thereby increases the risk in the financial system. There is some initial evidence to suggest that

this mechanism might have been particularly harmful during the European financial crisis—not

only did European banks hold a significant amount of sovereign debt to begin with but also they

continued to purchase more during the course of the crisis. For example, after the Long-term

Refinancing Operations (LTRO)5 by the European Central Bank (ECB), banks were allegedly

asked to use these funds to purchase government bonds to control sovereign bond yields (which

is now, rather infamously, referred to as the “Sarkozy trade”) (Reuters, 2012).

Angelloni and Wolff (2012), however, find that the level of bank holdings of sovereign

debt did not have a material impact on the equity valuation of bank between July and October of

2011. The authors use data from the two European stress tests to measure the changes in the

exposure of various banks to sovereign debt between July and October and then evaluate the

banks’ performance in light of these data. However, Eurozone stress tests measure only the

“trading-books” of the banks included in the sample. Most of the sovereign debt holdings of

                                                                                                               5 In December 2011, the ECB announced its first Long-Term Refinancing Operation (LTRO) for banks to gain access to funding from the ECB with a 1% interest rate and a 3-year term. This provision of long-term low interest rate loans to the banks using the banks’ portfolio as collateral provided a significant amount of fresh liquidity to the banking system. More than €500 billion were allocated within the first three months. Some banks allegedgly used this fresh liquidity to buy more of their own sovereign’s debt thereby improving conditions in the market for European sovereign credit risk as well. See ECB, 2012 and Hodson, 2011 for more details on the ECB’s LTROs.

Page 12: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    12  

European banks are held on their banking books6 rather than on the trading books estimated by

the stress tests (Blundell-Wignall and Slovik, 2010 and BIS, 2011). Blundell-Wignall and Slovik

(2010) further argue that banking book exposures should not be ignored by the market given

their importance in determining banking health and recovery rates in the case of default.

However, when considering only the trading-book exposures measured by the stress tests, the

results of Angeloni and Wolff are robust and may be extrapolated further to signify that

differences in bank holdings of sovereign debt do not change market perceptions of their

respective riskiness.

On the other hand, BIS (2011) offers some evidence stating that the “asset holdings”

transmission channel does indeed lead to market perception of greater risk in the banking sector.

Holdings of domestic government bonds as a percentage of bank capital remain higher in

countries with higher levels of public debt. Moreover, in countries where the banking sector has

larger claims on the sovereign debts of the PIGS countries, sovereign CDS premia (of the

countries where these banks are domiciled) co-move more closely with the sovereign CDS

premia of the PIGS countries (BIS, 2011). This suggests that bank holdings of the debt of the

domestic as well as foreign sovereigns can impact the health of the sovereign through the “asset

holdings” channel. Overall, the evidence in this case remains mixed.

B. Bank asset values and bank funding conditions

Higher sovereign risk manifested in higher spreads for the sovereign bonds or higher

CDS rates for sovereign credit default swaps reduces the values of bank assets that can be

pledged as collateral to obtain loans thereby adversely affecting bank funding conditions. The

                                                                                                               6 The Banking book includes all securities that are not actively traded by the financial institution but instead are supposed to be held till maturity. These securities are accounted differently than the securities on the trading book and are not necessarily marked-to-market or valued using the market price. The lack of market price valuation further increases the importance of considering these assets as the value of assets on the trading book might not be the “market price”. Losses born by financial institutions on the banking book may thus present a risk as well.

Page 13: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    13  

effect of the decrease in these asset values is multiplied across the financial sector given the

practice of re-hypothecation of securities. The repurchase (repo) markets provide an example of

a case in which decreasing value and increased uncertainty around asset value can cause a

deterioration of bank funding conditions (Duffie, 2010). If a bank holds some of its assets in the

form of sovereign debt of its own domestic or other sovereigns, then the decrease in the value of

these assets means that they can be pledged as collateral for lower amounts of funding.

Moreover, in case of an increase in the uncertainty around sovereign spreads and bond prices, the

“haircuts” imposed on sovereign bonds being pledged as collateral could increase7 (BIS, 2013).

These haircuts in the inter-bank and bank funding markets are primarily determined based on

uncertainty in collateral valuation, market liquidity and counterparty credit risk (CGFS, 2010).

Sovereign bonds, which typically have lower haircuts, are also used for benchmark rates

in the repo markets (Copeland et al., 2010). High valuation uncertainty in the sovereign markets

can thus lead to an overall deterioration of the bank funding market. There is significant evidence

that this mechanism has been observed in the European financial crisis. In June 2011, the

haircuts demanded by LCH.Clearnet (a major European clearinghouse) on Irish and Portuguese

banks were 75% and 65% respectively (BIS, 2011). Moreover, between 2009 and 2010, the share

of Irish, Greek and Portuguese bonds used as collateral as a share of the total European private

repo market had fallen by half (BIS, 2011), perhaps driven by rising haircuts. In this way,

changes in the creditworthiness of the sovereign may adversely affect funding conditions for the

banks within its jurisdiction. Theoretical models supporting this hypothesis can be found in

Brunnermeier and Pederson (2009) who show that deterioration in asset values and asset

liquidity adversely affects traders’ funding liquidity which may further affect asset prices.

Overall, evidence from the European financial crisis suggests that bank funding conditions have                                                                                                                7  Haircuts are the difference between the market value of the collateral pledged and the cash “loaned”).  

Page 14: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    14  

gotten worse with increases in sovereign risk. Whether this correlation represents a causality

remains an open question.

C. Sovereign ratings downgrade and bank funding costs and bank ROE

Many large banks in most advanced countries operate under the implicit or explicit

assumption of government bailouts in cases where default seems likely. In such a case, the

implicit or explicit bailout option acts like a contingent liability for the sovereign and a

contingent asset for the banks. Given that the markets assign a certain probability to the banks

being bailed out by the sovereign in case of default, the fiscal and financial health of the

sovereign may have a material impact on the health of the banks. The health of the sovereign

affects the probability, size and likelihood of success of the potential bailout in such cases.

Therefore, if the market’s perception of a sovereign’s health deteriorates, the market expectation

of the probability and extent of the bailout of the financial sector will also decrease. In the

starkest of cases, the sovereign may not have the capacity to come to the rescue of its banking

sector. There is considerable evidence that market participants price in the probability of

sovereign bailout while evaluating the health of the banking sector. For example, credit ratings

are “corrected” for the likelihood of sovereign bailouts. Credit ratings agencies assign two types

of ratings to banks (Moody’s Analytics, 2011):

i. Issuer rating or Bank Financial Strength Rating or BFSR represents the overall

probability that the bank will pay back its creditors, including external support.

ii. Individual or Baseline Credit Assessment (BCA) rating reflects the intrinsic capacity

of banks (without external help) to repay its debts.

The difference between the BFSR and the BCA represents the external (mostly sovereign)

“uplift” of bank ratings that stems mostly from the likelihood of external support (a “bailout”) if

Page 15: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    15  

the bank is unable to pay back its own debt. Rime (2005) found that proxies for too-big-to-fail

(TBTF) banks have a significant, positive impact on issuer ratings. On average, being TBTF

raises the banks’ credit ratings. In the case of the European financial crisis, the CDS implied

rating of bank debt between 2008 and 2011 has closely matched the Baseline Credit Assessment

or the banks’ individual ratings (Moody’s Analytics, 2011). This implies that after the onset of

the European financial crisis, market participants became skeptical of the probability and value

of sovereign bailouts. While the ratings uplift remained the same, the implicit market “uplift”

(measure by CDS rates) for financial sector debt based on the expectation of a sovereign

“bailout” went to zero. Clearly, the health of the banking sector (at least in cases of extreme

distress) is evaluated simultaneously with the health of the sovereign.

Further evidence for the effect of sovereign credit ratings on bank health and funding

performance comes from the equity markets. Correa, Lee, Sapriza and Suarez (2011) found that

a one-notch downgrade in the sovereign’s credit rating in the past 15 years has, on average, led

to a 2% reduction in bank equity returns in advanced economies. Overall, there is strong

evidence suggesting that changes in sovereign ratings had a significant impact on the market’s

perception of banking sector risk.

D. Sovereign “safety nets” and bank health

As argued above, the safety nets offered by the governments to large banks (in the form of

implicit or explicit guarantees of a bailout in cases where default seems likely) are included in

market participants’ assessments of the risk of the banking sector (See Rime, 2005). Sovereign

credit ratings present one prominent way in which participants evaluate the value of the “safety

net”. However, even without changes in the sovereign’s assigned credit rating, other news and

measures of sovereign risk could influence bank health. For example, if the fiscal health of a

Page 16: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    16  

nation deteriorates over the course of a few years then the value of the safety net goes down,

even if the sovereign credit rating of the country remains exactly the same. Specifically in the

context of the European Monetary Union, Arezki, Candelon and Sy (2011) found that news on

sovereign ratings had a material impact on bank stock prices in the EMU during 2007-20098. In

this way, the value of the safety net affects banking risk both directly and indirectly through the

sovereign credit ratings. Ejsing and Lemke, 2009 argue that bank bailouts (guarantee schemes)

reduce the risk spreads (proxy for funding costs) of banks at the cost of increased risk for

sovereigns.

Category 3: Financial sector determinants of sovereign risk

On January 15, 2009 Anglo Irish bank was nationalized due to its poor health and the

corresponding likelihood of default and subsequent damage to the Irish economy. On the

morning of January 16th, the share prices of the other 2 major Irish banks (Allied Irish and Bank

of Ireland) fell by approximately 13% each, reflecting a shock to the financial markets (Financial

Times, 2009). Irish sovereign spreads, in the same week increased by an average of 20% over the

previous week representing a rise of 32 basis points to 142 basis points from 110 bps (Mody,

2009). In this case, a clear market signal (the nationalization of Anglo Irish bank) of poor bank

health in Ireland prompted investors to re-evaluate the potential contingent liabilities of the Irish

government. The nationalization of Anglo-Irish signaled to the market that the banking sector of

the country could be a contingent liability for the Irish government--that is, they may have to

nationalize more banks.

                                                                                                               8 It should be noted here that “news” of sovereign credit ratings is not limited to a downgrade of sovereign debt. It can also include news on the trend of sovereign ratings and credit downgrade “watch” news. This implies that a change in sovereign ratings does not always work through the direct bank funding mechanisms described above. This further strengthens the argument in favor of “safety net” channels through which sovereign downgrades can impact financial sector health (see Category 2.D)

Page 17: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    17  

Similarly, various other countries in the European Union have also declared or re-iterated

the government’s support for the domestic banking sectors. Moreover, just like the Anglo-Irish

case mentioned above, various banks throughout Europe have either been re-capitalized or

nationalized or been given indirect government or central bank bailouts. The impact of this

sovereign assistance to banks can come in the form of two distinct pathways by raising both the

probability and the size of the contingent liabilities of the sovereigns. If the investors evaluate

that it is either more likely that the sovereign will bail out its banks or that in the case of a

bailout, that more resources would be required to achieve it, then the possible contingent

liabilities of the state increase. Attinasi, Checherita and Nickel (2009) argue that the re-

assessment of Euro area sovereign debt was caused by bank rescue packages that alerted the

markets to potential contingent liabilities of the sovereigns in the form of weak banking systems.

Measurement of the contingent liabilities described above proves a particularly vexing

estimation issue. One approach is to use the size of the financial sector as a proxy for the

contingent liabilities of the government. The rationale for such an approach states that since

measures of financial sector distress across banks are correlated (Duffie, 2010), when one bank is

in jeopardy, then the overall financial sector represents some sort of contingent liability for the

sovereign. Gerlach, Schulz and Wolff (2010) show that during periods of financial crisis (greater

aggregate and systemic risk), the sovereign risk premium increases with the size of the financial

sector. As the size of the financial sector relative to the rest of the economy increases, the

contingent liabilities of the sovereign from banking risk represent a greater fiscal challenge for

the government. Changes in CDS and CDS spread immediately following a bailout represent a

second way of measuring changes in market perceptions of contingent liabilities.

Page 18: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    18  

Another mechanism for the transmission of financial sector stress to sovereign risks is

through the economy of the country. Weakness in the financial sector not only cause banks to re-

consider lending to other banks but also causes changes in the country’s growth prospects,

investment projects and private sector lending market. In such a case, the country’s growth

prospects along with the sovereign’s fiscal situation suffer due to the financial crisis (Reinhardt

and Rogoff, 2012). Moreover, given the high holdings of sovereign debt by many Eurozone

banks, a worsening of bank assets may cause either a sale of sovereign debt by banks (perhaps

weakening sovereign bond prices). This would decrease their liquidity for investment in bonds of

their own government (softening the demand for sovereign bonds, especially in the primary

market). Especially in “peripheral’ countries (where the government debt is not seen as the

ultimate safe harbor and thus there is no flight to quality effect), this mechanism could

particularly harm sovereign credit risk conditions. Mody (2009) finds that the link between

banking and sovereign risk is particularly strong in countries with poor public finances.

Category 4: What factors affect the nature and extent of this relationship?

Perhaps as a result of the thinking prior to 2007 that sovereign spreads were not

dependent on “internal factors” but only on global ones, there is remarkably thin research on the

nature and extent of the relationship between banking and sovereign risk. The primary question

in this strand of literature revolves around the factors affecting the relationship between banking

and sovereign credit risk.

Firstly, part of the research shows that at least in the Eurozone, the relationship between

banking and sovereign risk has changed over time (Mody and Sandri (2011), Mody (2009),

Pagano and von Thadden (2004) and so on). Prior to the financial crisis, sovereign spreads of

various countries in the Eurozone were trading nearly at zero spreads regardless of the internal

Page 19: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    19  

fiscal or financial variables. In fact, on 6th July, 2007, Irish bond spreads were trading at a -30

bps yield spread against the German Bund (10 year yields). Clearly, these conditions have

changed since the onset of the financial crisis (see Section IV- Data and summary statistics).

Secondly, there is now significant research that shows that fiscal and trade policy are

significant determinants of the extent of the relationship between sovereign and banking risk.

Mody and Sandri (2011) find that countries with higher debt-to-GDP ratio saw a greater co-

movement in the relationship between sovereign and financial risks. This strand of the literature

argues that while the correlation between banking and sovereign risk persists across the

advanced economies, the correlation is higher in countries with worse fiscal health. Debt/GDP

ratios as well as current account balances have been used as a proxy for the fiscal health.

The finding that fiscal characteristics lead to changes in the level of correlation between

banking and sovereign risk implies that the correlation depends on the levels of banking and

sovereign risk of the country. The argument here is that worse fiscal health leads to higher

sovereign risk and consequently, the higher level of sovereign risk leads to a higher correlation

between sovereign and banking risk. The above hypothesis is also related to the “fiscal

dominance” hypothesis which contends that metrics of the fiscal health of the nation, like

Debt/GDP and current account deficit dominate the determinants of the sovereign’s risk

premium (see Greenlaw, Hamilton, Hooper and Mishkin, 2013 for a thorough explication of this

argument).

However, it remains unclear if the health of the banking sector is counted in the “fiscal”

factors outlined in this strand of the literature. While the banking sector does represent a

potential contingent liability for the sovereign, the health of the banking sector is not included in

the ‘fiscal’ measures used by the various authors referenced above. Only for countries that have

Page 20: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    20  

conducted outright nationalization of one or many of its banks, the health of the banking sector is

directly reflected on the “balance sheet” of the sovereign. In cases where the banking sector

remains weak and the likelihood of bailout remains high, but no nationalizations have been

conducted, fiscal metrics do not reflect the potential contingent liabilities in the form of banking

sector bailout risk. In this sense, it is worth some thought whether the fiscal dominance

hypothesis would hold up after the inclusion of measures for the health of the banking sector.

Section III. Hypotheses

This paper seeks to answer questions primarily in the first and fourth “categories” of the

literature outlined above—focusing on the factors determining sovereign risk premia and those

affecting the relationship between sovereign and financial sector risk. This section provides a

brief outline and explanation of the 3 main hypotheses that are tested in this paper, while detailed

description of the methodology used to test each of the hypotheses is provided in Section V after

a description of the basic estimators used to measure sovereign and banking risk in Section IV.

Note that the first two hypotheses outlined below revolve around the determinants of sovereign

risk premia in the European financial crisis. On the other hand, the third hypothesis focuses on

the correlation between sovereign and banking risk.

A. Hypothesis 1 or the “Co-movement” hypothesis: There exists a statistically significant correlation between changes in measures of banking and sovereign risk. This correlation is significant even after the inclusion of variables reflecting global financial conditions along with other control variables.

This hypothesis argues that changes in the banking risk measure are a statistically

significant determinant of changes in the sovereign risk measure. Further, this

relationship persists even after the inclusion of various control variables that are supposed

to reflect changes exogenous to the market for banking and sovereign risk.

Page 21: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    21  

B. Hypothesis 2 or “Bank Dominance” hypothesis: Banking sector risk is the dominant statistically significant predictor of sovereign CDS rates, even after controlling for fiscal variables like the ratio of debt-to-GDP and current account balances. This hypothesis is defined in stark contrast to the “fiscal dominance” hypothesis,

which states that fiscal factors overpower other variables in predicting sovereign CDS

rates. On the contrary, the bank dominance hypothesis posits that banking sector risk

measures are dominant predictors of sovereign credit spreads. This hypothesis seeks to

find the predictors of the levels of sovereign risk premia and thereby includes banking

sector and fiscal variables along with various liquidity and exogenous market control

variables.

C. Hypothesis 3 or “Joined-at-the-hip” hypothesis: The extent of the correlation between banking and sovereign risk is directly proportional to the either the level of banking risk or the combined level of banking and sovereign risk. Simply put, the correlation between banking and sovereign risk is higher for

higher levels of banking and sovereign risk. If the level of banking risk in a given country

rises, then not only does that lead to an increase in the measure of sovereign risk

(Hypothesis 1) but also strengthens the correlation between banking and sovereign risk.

This hypothesis implies that banking and sovereign risk are “joined-at-the-hip”; an

increase in banking risk of a given magnitude (say 10 basis points) is more likely to, on

average, lead to a higher increase in sovereign risk if banking risk levels are already

elevated.

Page 22: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    22  

Section IV: Data and summary statistics

The data for the research presented in the subsequent pages was obtained through

Datastream, Thomson One, Chicago Board Options Exchange and Bloomberg. The dataset

includes 20 banks across 4 Eurozone countries—France, Germany, Italy and Spain (a detailed

list of banks included across all sovereigns is attached in Appendix A). The data used stretches

from October 13, 2008 to January 1, 2013 and is based on daily quotes that are sometimes

aggregated into weekly and monthly statistics where appropriate. Sovereign risk is measured by

using CDS premia mid points for 5-year9 maturity Euro traded complete restructuring (CR)10

credit default swaps available on Datastream. Credit default swaps valuations have been shown

to be based on credit risk, liquidity risk as well as other factors that are not related to the

country’s fiscal situation (See Longstaff et al, 2011). Still, this paper uses CDS as a measure for

credit risk due to 2 primary reasons: firstly, other measures like bond spreads are found to be

confounded in the crisis period covered in this dataset. Take, for example, the sovereign bond

yield spread (as an alternative measure of sovereign credit risk) which is calculated by

subtracting the sovereign’s bond yield from a “risk-free” rate. In the financial crisis period that is

analyzed herein, many rates that were hitherto considered to be “risk-free” (like the U.S. bond

yield or the tri-party repo rate) experienced fluctuations. CDS bond premia represent a simple

measure through which the credit risk of the sovereign can be estimated without the use of a

“risk-free” rate, representing a “pure” measure of sovereign risk. Having said that, there is

significant recent research that shows that only 1/3rd of sovereign risk premia is determined by                                                                                                                9 Note here that the CDS of 5-year maturity remains, by volume, the most traded CDS contract for sovereigns as well as other issuers. See Pan and Singleton, 2008 for more on the term structures of sovereign CDS. 10 Complete restructuring (CR) and Modified Modified (MM) are ISDA documentation clauses governing different types of CDS contracts. CR terms treat almost any restructuring as a credit event and allow bonds of up to 30-year maturity to be “delivered” in the auction for settling and claiming compensation. This option is obviously popular in the case of sovereigns given the relatively high amount of long-term debt and propensity to “restructure” debt rather than default outright. MM terms on the other hand are more restrictive. See Parker and Zhu, 2005 for a detailed explanation.

Page 23: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    23  

sovereign-specific factors (see Longstaff, Pan, Pederson and Singleton, 2011 for details). On its

face, this represents a case against using CDS to measure sovereign risk. However, if appropriate

controls for liquidity, stock market and volatility conditions can be included in a regression

explaining sovereign risk, then the sovereign’s credit risk may be well reflected by the measure.

This presents the second reason for using CDS for measuring sovereign risk—using well-known

control variable can lead to a “pure” measure of sovereign credit risk11. Note that, Euro traded

CDS premia are selected to avoid the data being confused by currency fluctuations and other

exogenous factors. It should also be pointed out here that 5-year CR sovereign CDS are used

because they are the most frequently traded maturity and type in this market. Figure 2 shows the

co-movement of CDS on sovereigns in the EMU in the 219 weeks ranging from 10/13/2008 to

1/1/13.  

For similar reasons as

described above, banking

sector risk is measured by

CDS premia mid points for 5-

year bonds of the banks. The

particular type of CDS

premium used here is the 5-

year Euro traded Modified

                                                                                                               11 Note that the intention here is not to claim that CDS risk premia are completely unbiased. On the contrary, CDS settlement procedures have been shown to be biased (see Du and Zhu, 2012 and Duffie and Thukral, 2012 among others). The intention here is to say that given problems in measuring spreads because of problems revolving around the “risk-free” interest rate, CDS risk premia provide a good measure of market perceptions of sovereign risk.

010

020

030

040

050

0So

vere

ign

CD

S in

bas

is p

oint

s

0 50 100 150 200Week number starting 2008

France GermanyItaly Spain

Figure 2: Co-movement of sovereign CDS

Page 24: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    24  

Modified (MM) which is again chosen since it is one of the most widely traded and quoted CDS

premia in this market. Note that the disparity between CR and MM CDS included does not

matter as long as the types of CDS quotes used for each category are consistent.

Individual banks’ CDS rates are, in themselves, insufficient as a measure of banking sector

risk since they are vulnerable to internal bank factors and may not represent the extent of

banking sector risk in the country. The next step was to set up a comprehensive measure for

banking risk in each country

included in the sample. For this

purpose, a comprehensive

Banking Risk Index was

developed using an asset-

weighted average of the CDS

premia of the top 5 banks for

which CDS quotes were

available. The BRI is better

than using individual bank

CDS since it weighs various

banks by using their assets—

banks with higher assets

represent larger contingent

liabilities for the underlying

sovereign. Moreover, the BRI

is consistent across all

020

040

060

0C

DS

risk

prem

ia in

bas

is p

oint

s

France Germany Italy Spain Country

CDS_sov CDS_bank

Figure 3a: Range and distribution of risk premia observations

010

020

030

040

050

0Ba

nkin

g R

isk

Inde

x in

bas

is p

oint

s

0 50 100 150 200Week number starting 2008

France GermanyItaly Spain

Figure 3: Co-movement of Banking Risk Indices in Europe

Page 25: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    25  

countries, if Spain’s BRI value is higher than that of Germany’s then that implies that the

banking sector (or at least of the top 5 banks) of Spain is considered to be riskier than that of

Germany. The same is not necessarily true for individual level CDS comparisons. Figure 3

shows the levels of the BRI for various countries in the 219 weeks ranging from 10/13/2008 to

1/1/13.  

After establishing the measures used for sovereign and banking risk, the next step lies in

establishing whether initial evidence suggests any correlation between the measures of sovereign

010

020

030

040

050

0C

DS

Ris

k pr

emiu

m

0 50 100 150 200Week number starting 2008

Sovereign CDS, Spain Banking Risk Index, Spain

Figure 5: Co-movement of Spanish sov and banking risk

010

020

030

040

0C

DS

Ris

k pr

emiu

m

0 50 100 150 200Week number starting 2008

Sovereign CDS, France Banking Risk Index, France

Figure 7: Co-movement of French sov. and banking risk

050

100

150

200

250

CD

S R

isk

prem

ium

0 50 100 150 200Week number starting 2008

Sovereign CDS, Germany Banking Risk Index, Germany

Figure 6: Co-movement of German sov. and banking risk

010

020

030

040

050

0C

DS

Ris

k pr

emiu

m

0 50 100 150 200Week number starting 2008

Sovereign CDS, Italy Banking Risk Index, Italy

Figure 4: Co-movement of Italian sov. and banking risk

Page 26: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    26  

and banking risk selected here. Given that the initial correlation (before introducing controls) is

well documented in the literature, this exercise can also be taken as a test of the validity of the

BRI for measuring banking risk in a given country (see Figures 4-7 below).

The above graphs provide preliminary evidence of the correlation between sovereign and

banking risk in the Eurozone crisis. Note that Figure 6 focusing on German sovereign and bank

CDS may not look like a strong correlation, but a simple correlation levels analysis of the 2

variables suggests that there is a 53% sample correlation between banking and sovereign risk

without including any control variables.

Apart from the variables for measuring banking and sovereign risk, various control

variables were also added to the constructed dataset to make sure that factors apart from banking

risk that affect sovereign risk premia are included in the analyses. A brief list of the control

variables with short descriptions for each is provided hereunder:

a. VIX: An index maintained by the Chicago Board Options Exchange that represents the

implied volatility of the S&P500 index. Previous research (see Fontana and Scheicher,

2010) has suggested that VIX is a significant determinant of sovereign risk premia. The

implied thought process here is that a higher volatility for the S&P 500 represents an

“event risk” or the risk of an exogenous event affecting the price of various financial

instruments including CDS. A positive coefficient for VIX is expected given that an

increase in volatility implies an increase in risk through the event risk channel.

b. Stock exchange indices for the country: Changes in the levels of the major equity indices

which represent the country’s “blue-chip” firms represents a proxy for external

conditions. The stock market levels of each country are normalized to 100 at the

beginning of the period since the absolute levels of stock market indices by country may

Page 27: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    27  

bias the results of the regression. Note that the CAC 40 (France), DAX 30 (Germany),

FTSE MIB (Italy) and the IBEX 35 (Spain) are used as control variables in the sample.

See Figure 8 for a graphic

representation of the

normalized levels of these

stock indices.

c. U.S. 5-year CDS: The variable

in question here is the CDS

premia on the 5-year CR Euro

traded bond. Two effects might be at play here: first, sovereign CDS premia may move

together and in that sense a worsening of creditworthiness of the U.S. may be correlated

with a worsening of the creditworthiness of other sovereigns. However, a “flight to

quality” effect may also be observed in which an increase in the sovereign CDS premia

for a less creditworthy sovereign may increase demand for the bonds of a borrower that is

more creditworthy by comparison, thereby lowering its risk premia. It is unclear whether

the coefficient of this variable should be positive or negative.

d. Current account balance: The current account balance of the sovereign in question

represents the fiscal conditions of the sovereign. It is expected that sovereigns with

positive and higher respective current account balances will have lower CDS premia

associated with them. Note here that the current account data used herein is avaliable by

month and not by week.

e. Debt securities-to-GDP ratio: Data on the debt-to-GDP ratios of constituent countries

would also be an important addition to the control variables used in these regressions.

6080

100

120

140

160

Nor

mal

ized

stoc

k ex

chan

ge le

vels

(sta

rts a

t 100

)

0 50 100 150 200 Week number starting October 2008

France Italy Germany Spain

Figure 8: Normalized stock exchange levels

Page 28: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    28  

Unfortunately, such data is not available on a weekly or monthly level and yearly

regressions would lose the granularity that is sought in this paper. A ratio of outstanding

debt securities12 of that sovereign to the level of debt for the sovereign (from Fotana and

Scheicher, 2011) is here used as a proxy for other fiscal conditions especially the debt-to-

GDP ratio (see Section IV.C.1 for more details). This variable, which closely tracks the

debt-to-GDP ratio, presents an estimate of the outstanding debt securities of a country on

a monthly basis. The granularity provided by this measure helps us determine the

predictors of sovereign CDS rates in the relatively short period of the European financial

crisis.

Note that the variables outlined above vary with regards to their frequency of observation

(monthly or weekly) as well as with regards to differentiation across countries. First differences

and lagged values of the above variables are used at various points in the estimation.

The third hypothesis of the paper, revolving around the “joined-at-the-hip” assumption also

requires a calculation of the correlation between sovereign CDS and the CDS premia of

individual banks. Weekly values of the trailing 6 months are used to calculate this correlation

with an overlap of 5 weeks and 3 months between subsequent values of the correlation

coefficient. The joined-at-the-hip hypothesis states that the correlation between banking and

sovereign risk is a directly proportionate function of the level of banking risk—that is, in periods

of heightened banking and sovereign risk, the correlation between banks and sovereigns is also

higher. The detailed methodology to test this hypothesis is provided in Section IV.C.1. below.

                                                                                                               12  Note here that this measure is derived by adding values of short-term and long-term securities outstanding for the central government of each of the countries. While this ratio does not match the yearly outstanding debt-to-GDP data exactly, it tracks very closely to the same. It should be added here that this measure, like the debt-to-GDP ratio, represents the fiscal liabilities of only the central government.  

Page 29: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    29  

Section IV: Methodology and Results

The econometric estimation of the relationship between sovereign and banking risk is

carried out in 3 steps in order to determine the validity of the three hypotheses outlined above.

This section explains each step of the econometric methodology of this paper and discusses ways

in which the methodology helps us determine the answers to each one of the hypotheses

developed earlier. The three specifications to determine the validity of these hypotheses are:

IV.A.1. Tests for the co-movement hypothesis

The co-movement hypothesis states that sovereign risk premia and banking risk index

move together. This hypothesis is focused on finding out the predictors for the changes in

sovereign risk premia and not the outright levels and thus first differences in the variables are

used here. The first step in establishing this estimation technique is to examine if there is indeed

any relationship between changes in sovereign and banking risk. Figures 9-12 provide a

graphical analysis of the link between changes in banking and sovereign risk premia:

Page 30: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    30  

Note also that as seen in Graph 1 above, sovereign risk premia are also highly correlated

with the risk premia of other countries as well as the previous risk premia for the sovereign itself.

Lags of all variables are included in the regression so as to ensure that mean-reversion effects

and ripple effects of changes in one market on the other market are captured by the regression.

Note that first differences are denoted by D and lags are denoted by Ln where n represents the

-20

-10

010

20 C

hang

e in

Sov

erei

gn C

DS

-40 -20 0 20 40Change in Banking Risk Index

D_cds_sov 95% CIFitted values

Figure 10: Correlation between changes in German Sov. and bank CDS

-100

-50

050

100

Cha

nge

in S

over

eign

CD

S

-50 0 50Change in Banking Risk Index

D_cds_sov 95% CIFitted values

Figure 12: Correlation between changes in Spanish Sov. and bank CDS

-40

-20

020

40 C

hang

e in

Sov

erei

gn C

DS

-40 -20 0 20 40 60Change in Banking Risk Index

D_cds_sov 95% CIFitted values

Figure 9: Correlation between changes in French Sov. and bank CDS-1

00-5

00

5010

0

Cha

nge

in S

over

eign

CD

S

-50 0 50Change in Banking Risk Index

D_cds_sov 95% CIFitted values

Figure 11: Correlation between changes in Italian Sov. and bank CDS

Page 31: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    31  

degree of the lag. Using common practice, third degree lags for each variable are included in the

regression where they are significant. Let i represent sovereigns and t time. The equation is now:

where:

• D represents the first difference or change in the value of the variable

• Ln represents the nth lag of the variable

• t represents the time, measured in weeks for this equation

• D_CDS_sov represents the first difference or the change in the CDS premium of the

sovereign between 2 subsequent weeks

• Ln D_CDS_sov represents lagged values of the change in sovereign CDS

• LnD_CDS_bank represents the banking risk index (BRI) for country i at a given time t

• LnD_US_CDS represents the CDS risk premia for the US. This serves as a flight to quality control and a control for factors affecting health of the global financial system.

• LnD_Stock_ex represents lagged values of the changes in the normalized value of country

i’s equity index

• ε is residual

The above econometric estimation strategy will help us determine if there is a statistically

significant correlation between changes in the levels of sovereign and banking risk in a particular

country. Note that several problems are expected to arise in the estimation strategy outlined

above. Firstly, in the time period selected, 219 weeks from 10/13/2008 to 1/1/2013, country-by-

country analyses may be possible but would represent a very limited sample size. However, daily

observations in CDS markets are rather untrustworthy given that the volume of CDS contracts

traded on each of the reference entities in a given day may have been zero. In fact, especially at

D_CDS _ sovi,t =α + βLnD_CDS _ sovi,t + χLnD_CDS _banki,t + φLns=0

p

∑s=0

m

∑s=1

l

∑ D_US _CDSt−s +

ϕLnD_Stock _Exi,t−ss=0

p

∑ +Country_Di + ε i,t

Page 32: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    32  

the beginning of the crisis, the CDS quotes for single bank reference entities varied due to lack of

sufficient volume or depth in the reporting markets (See Fontana and Scheicher, 2011).

Therefore, weekly averages need to be used so as to make sure that the values used represent

significant changes over time. However, this necessitates the use of panel data techniques

throughout the estimation, as country-by-country weekly analysis is not deep enough. The use of

panel data, which “extends” the dataset in a sense, precludes or at least ameliorates, the second

concern, that of endogeneity arising from the use of lagged variables to measure second order

effects. The panel data techniques that are used allow for first-order autocorrelation coefficients

across countries and also allows for errors to vary by country (heteroskedasticity).

Thirdly, some global factors are captured by the use of US CDS premia and other

country-specific factors (not related to banking risk) are captured by the use of country level

stock exchange indices. However, there are concerns of simultaneous determination of stock

exchange indices and banking risk since the banking sector stocks form a significant portion of

the respective countries’ stock indices. Each of variables above were excluded in some

regressions to provide a preliminary robustness check for the variables. Note here that factors

reflecting the fiscal situations of each of the countries are not included in the regression here due

to this regression using weekly data whereas changes in the fiscal situation are not available at

that level. In such a case, the effect of fiscal factors may be captured by country fixed effects

variables. Fiscal factors are, however, considered again in the next step of this econometric

strategy.

Page 33: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    33  

IV.A.2. Results of the co-

movement hypothesis

Given the panel data

techniques outlined above, the

first question is if the changes

in sovereign risk and banking

risk are still relevant in the

panel data framework. The

concern here is that the

correlation and co-movement

between these two variables, when measured by country, has been confused or confounded by

the panel data framework employed to “extend” the dataset. Figure 13 provides a scatterplot

graph showing that the correlation between these 2 variables is maintained within this panel data

framework. The simple correlation coefficient for changes in the levels of banking and sovereign

risk is 49%. This presents preliminary evidence around the correlation between these two

variables in the panel framework. Table 1 presents the results for the regressions that were run to

test the co-movement hypothesis outlined above.

Table 1: Regressions for weekly changes in Sovereign CDS

Change in sovereign CDS rate Coefficient Coefficient Regression # Regression (1) Regression (2) Lag 1 of first difference in Sovereign CDS

0.050119 (0.0572)

0.0625671 (0.057)

Lag 2 of first difference in Sovereign CDS

-0.129495*** (0.050)

-0.1231938** (0.05)

Lag 3 of first difference in Sovereign CDS

-0.0384183 (0.053)

--

First difference BRI 0.6682809*** (0.048)

0.6566695*** (0.048)

-100

-50

050

100

Cha

nge

in S

over

eign

CD

S

-50 0 50Change in Banking Risk Index

D_cds_sov 95% CIFitted values

Figure 13: Correlation between changes in Sov. and bank CDS

Page 34: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    34  

Lag 1 of first difference BRI -0.2173102*** (0.060)

-0.2172261*** (0.06)

Lag 2 of first difference BRI 0.0895494* (0.05)

0.0815867 (0.056)

Lag 3 of first difference BRI 0.032178 (0.054)

--

First difference US CDS rate 0.51816*** (0.119)

0.5345184*** (0.117)

Lag 1 of first difference US CDS rate

0.2322222* (0.14)

0.2002049 (0.13)

Lag 2 of first difference US CDS rate

-0.1521693 (0.137)

-0.223013* (0.121)

Lag 3 of first difference US CDS rate

-0.0562626 (0.0933)

--

First difference in equity index -0.2983663* (0.174)

-0.639091*** (0.243)

Lag 1 of first difference in equity index

-0.1344464 (0.152)

0.0467372 (0.183)

Lag 2 of first difference in equity index

-0.0737607 (0.152)

-0.1551473 (0.201)

Lag 3 of first difference in equity index

0.2030588 (0.1522)

--

First difference in VIX -- 0.4856596**

(0.226)

Lag 1 of first difference in VIX -- 0.0840999

(0.1922)

Lag 2 of first difference in VIX -- -0.0608827

(0.148)

France dummy -- -0.289788

(0.658)

Italy dummy -- -0.1380676

(1.083)

Spain dummy -- 0.1024529

(0.935) Constant 0.0515427

(0.356) 0.0905341

(0.442) R2 0.5423 0.5515

Note here that both of the regressions included in the table above have high explanatory

power of about 55% each. The second order lags of sovereign CDS rates are statistically

significant and negative implying that the variable in question is mean-reverting—an increase in

period 0 will lead to a decrease in period 2 although of smaller magnitude than the initial

Page 35: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    35  

increase itself. Augmented Dickey-Fuller tests are conducted (by country) to further confirm if

the variable in question is mean-reverting.

In terms of the co-movement between sovereign and banking risk, weekly changes in the

banking risk index are found to be significant (with the p-value lower than 0.001) and positive

(with a coefficient of 0.66). Note here that the interpretation of this coefficient is important:

every one basis point change in the Bank Risk Index is correlated with a 0.66 basis point change

in the sovereign CDS. The contemporaneous nature of the correlation between banking and

sovereign risks implies that in the period of this dataset, risks in one sector were quickly

transmitted to the other. The first lag of the BRI, however, is negative and significant (albeit with

a much lower coefficient) suggesting an explanation of “pullback”—the market over-reacts at

t=0 and the increase in sovereign risk, correlated with an increase in the BRI, is over-estimated.

In the first lagged period, some of the overreaction is corrected. However, this course correction

described above does not mean that changes in the BRI have no net effect on market perceptions

of sovereign risk. The lincom (linear combination) function is used to determine if the net effect

of the opposite values of the first difference and the lagged first difference is, in fact, zero. The

hypothesis that the net effect of the lagged coefficients of the BRI is zero is roundly rejected by

the statistical test (for both regressions). This result holds even when other values of the lags of

BRI are included in the lincom test described above. Therefore, it can safely be said that there is

statistically significant and positive correlation and co-movement between changes in the

sovereign risk premia and the BRI. Results of the various tests from this section can be found in

Appendix B.

The coefficients for changes in US sovereign CDS are also positive and significant. The

significance of the variable implies that the effect of negative external shocks that worsen all

Page 36: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    36  

sovereign risk premia outweigh the effects of the “flight to quality” phenomena. If the flight to

quality effect were to be dominant, then a decrease in the US sovereign CDS would have

coincided with an increase in the CDS premia for other countries. So, even though it might be

true that U.S. bonds and instruments are affected by the investors’ flight to quality, the

correlation across various sovereigns seem to over-ride that effect. As above, the lincom test for

determining if the positive and negative values of various lags cancel each other out, soundly

reject the null hypothesis. Therefore, the net effect of increase in the US sovereign CDS

premium leads to an increase in sovereign CDS premia in the Euro area countries.

Furthermore, statistically significant coefficients are found for the value of the respective

countries’ normalized equity index (negative) and for the implied volatility index (VIX; positive)

of the S&P500. Note here that the stock exchange indices of the various countries are normalized

to 100; an increase of 1% (or 1 point) in the normalized stock market index, leads to a 30 basis

points (Regression 1) or a 60 basis points (Regression 2) decrease in the value of the sovereign

CDS, ceteris paribus. On the other hand, an increase in the volatility of the S&P500 is correlated

with an increase in the sovereign risk premia in the Eurozone.

In conclusion, it is safe to say that the data supports the co-movement hypothesis; there is

a statistically significant relationship between changes in sovereign risk and the banking risk

index.

IV.B.1. Methodology for the “bank dominance” hypothesis

The bank dominance hypothesis states that measures of banking risk are statistically

significant and dominant in explaining sovereign credit risk premia even when controlling for

country-specific fiscal factors. This hypothesis is directly in contrast to the strand of literature

that talks about fiscal dominance or the importance of fiscal factors in explaining sovereign risk

Page 37: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    37  

premia (See Greenlaw et al. 2013). On the other hand, the “bank dominance” hypothesis argues

for the primacy of financial sector risk in the determination of sovereign credit spreads. The

econometric strategy is to explain the levels of CDS rates on sovereigns rather than changes in

these levels with the idea being that fiscal factors are more likely to determine the levels and the

long-term aspects of sovereign risk premia rather than short-term fluctuations. The “risk premia”

are the monthly averages of the CDS rates on banks and sovereigns, respectively, included in the

sample. There are 51 months of observations for each country. The measures to estimate

sovereign and banking risk remain the same as above, just not in terms of first differences.

We turn to consideration of which measures to consider as the fiscal characteristics of

each of the countries in the sample. The ratio of gross debt-to-GDP of the sovereign is a common

measure of the indebtedness of the country. Various studies (see Reinhardt and Rogoff, 2012,

and Greenlaw et al., 2013) have determined that debt-to-GDP ratios are important determinants

of growth as well as of sovereign risk premia. However, in the limited time period of interest to

us here, yearly variations of debt-to-GDP ratios do not provide enough granularity for our

purposes. However, the

outstanding quantities and

secondary market pricing of short-

term and long-term debt securities

of the central government is

available on a monthly basis. The

total outstanding quantity of debt

securities of all maturities for a

given country closely tracks the total debt. Linear interpolation within each individual year is

4060

8010

012

0O

utst

andi

ng s

ecur

ities

to G

DP ra

tio

0 10 20 30 40 50Month number since October 2008

Italy Spain France Germany

Figure 14: Outstanding public securities to GDP ratio

Page 38: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    38  

then used to obtain a proxy, illustrated in Figure 14, for the country’s public debt-to-GDP ratio

for each month (similar to Fontana and Scheicher, 2011). The expectation here is that a higher

value of the ratio of outstanding securities to GDP is correlated with higher CDS rates for the

country.

Monthly estimates of the countries’ current account balances are also included in the

regressions as a measure of the country’s fiscal health. The current account balance, a measure of

the strength and extent of a nation’s foreign trade, comprises the country’s balance of trade

(exports minus imports), factor income (earnings on investments abroad minus payments made

to foreign-domiciled investors) and cash transfers. A country with higher levels of exports is

more likely to remain able to pay back its debts. The regression to test the “bank dominance”

hypothesis is now:

Wherein:

• i represents the country

• t represents time measured in months

• CDS_sov is the average risk premia on sovereign CDS for country i during month t

• CDS_bank, the asset weighted average of within country bank CDS rates, is the Banking

Risk Index

• US_CDS is CDS rate on 5-year CR CDS

• N_stock_ex is the country’s primary stock exchange levels normalized to 100 at the

beginning of the period

• L1_Curr_Acc is the first lag current account balance of the country

CDS _ sovi,t =α + βCDS _banki,t + χUS _CDSt +φN _ stock _ exi,t +ϕCurr _Acci,t +κCurr _Acc_Squarei,t + λSec_ to_GDPi,t + ε i,t

Page 39: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    39  

• L1_Curr_Acc_Square is the square of the first lag of the current account balance for the

country

• Sec_to_GDP is the first lag of the ratio of outstanding sovereign debt securities to GDP

• εi,t is residual

Unlike the econometric strategy associated with the “co-movement hypothesis”, this

regression model is designed to identify potential determinants of the levels of sovereign risk

premia, rather than to causes for changes in the same. There are two reasons for this shift in

perspective: first, while short-term changes in banking risk are shown to be significantly related

to short-term changes in sovereign risk, determinants of the levels of sovereign risk remain

unexplained. To put it simply, we have already learned that in the short-term, banking and

sovereign risk move together. However, we do not yet know if the levels of sovereign risk

premia are also determined by the health of the banking sector. Second, the “bank dominance”

hypothesis suggests that in the long-run, financial sector variables are important in determining

sovereign risk spreads. This hypothesis can only be tested using long-term (here monthly

averages) data on the levels of sovereign risk and not just changes in the same.

As with the previous specification, sovereign and banking risks are respectively measured

using average CDS premia (monthly in this case) and the Banking Risk Index, both measured in

basis points. Controls for both international financial shocks (US CDS premium) as well as

country-level financial shocks are included in the equation. The levels of the primary stock

exchange of the country are normalized to 100 at the beginning of the period (for October, 2008).

As discussed above, the square of the current account balance for the country and the value of

the public securities outstanding to GDP are used as fiscal variables that may also play a role in

determining the long-term levels of sovereign risk premia. The square of the current account

Page 40: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    40  

balance of each country is also included in the regression, given the results of Greenlaw,

Hamilton, Hooper and Mishkin (2013). Dummy variables to capture country fixed effects are not

included as they may confound the estimated effect of the economic explanatory variables

through over fitting.

IV.B.2. Results for the banking prominence hypothesis

The correlation

between the levels of

banking sector and

sovereign credit risk is

also high (see Figure 15),

with a sample correlation

level approaching 77%.

This level of sample

correlation exceeds even

that seen between the

changes in measures of sovereign and banking risk and implies that one should be a good

predictor of the other. Table 2 below provides the results for the regression described above:

Table 2: Regression of levels of Sov. risk premia on levels of BRI and fiscal variables Regression (3) (4) (5) Sovereign CDS rate

Coefficient Coefficient Coefficient

Banking risk Index

1.223824*** (0.506)

1.207269*** (0.117)

1.277831*** (0.0603)

Lag 1 of BRI -- 0.0848133 (0.19)

--

Lag 2 of BRI -- 0.149274 (0.172)

--

Lag 3 of BRI -- -0.1737436 --

010

020

030

040

050

0 S

over

eign

CD

S le

vels

100 200 300 400 500 Banking risk index

Figure 15: Correlation between levels of sov. and banking risk

Page 41: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    41  

(0.106) VIX 4.051508***

(0.494) 4.829157*** (0.954)

5.267176*** (0.831)

US CDS rate 1.144361*** (0.26)

0.8926596*** (0.2213)

0.855634*** (0.185)

Normalized equity index levels

-0.4667448* (0.233)

-0.3232873 (0.2663)

-0.0775059 (0.244)

Lag 1 of Current Account Balance

-0.0017978*** (0.000631)

-0.0016854** (0.000656)

--

Lag 2 of Current Account balance

-- -0.0018056*** (0.00054)

Lag 1 of outstanding securities-to-GDP ratio

0.9316905*** (0.166)

0.9484141*** (0.1852)

--

Lag 2 of outstanding securities-to-GDP ratio

-- 1.035341*** (0.1808)

Lag 1 of current account squared

8.78E-08* (4.67E-08)

9.41E-08** (3.89E-08)

--

Lag 2 of current account squared

-- 6.05E-08* (3.36E-08)

Constant -61.18769 (42.913)

-59.68267 (53.1953)

-109.6169** (51.24)

R2 0.9077 0.9207 0.9159 Table 2 shows that even when fiscal factors for each country are included, banking risk is

an important predictor of sovereign risk levels. The fraction of sample variation in sovereign

CDS rates that is explained by each of these regressions now rises to 90%. From regression (3),

we see that a 1 basis point increase in the BRI predicts a 1.22 basis point increase in the

sovereign risk premia indicating that there is a multiplier effect at play here. One hypothesis for

the “multiplier” effect observed above is that the markets initially “over-react” to high levels of

banking risk and then undergo a course-correction in subsequent periods. Regression (4) above

presents the results of a test for this hypothesis—using lags of the BRI to examine whether there

is a course correction in the evaluation of sovereign risk. Overall, suffice to say here that the

“over-reaction” hypothesis is not supported by the data and also seems much less likely when

Page 42: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    42  

considering monthly averages; the speed of modern-day trading seems unlikely to allow for

corrections over lags of a few months.

Although the fiscal explanatory variables are shown to be statistically significant (see

table 3), the economic magnitudes of the measured effects are negligible. For example, a €1

billion increase in the current account balance predicts a 0.0018 basis points decrease in the

country’s CDS rate (in regression (3)). On the other hand, each basis point increase in the

banking risk index predicts at least an equivalent decrease in sovereign CDS premia. Having

emphasized the statistical significance of the current account deficit, it should be noted that its

magnitudes are diminutive when compared to the effect of banking risk in the same regression.

Similarly, the outstanding securities-to-debt ratio is also statistically significant but

insignificant in magnitude—it would require a swing of 10% in the ratio for there to be a shift of

10 basis points in the sovereign risk premium. The evidence provided by the regressions above

seems to confirm the bank dominance hypothesis given that variables measuring the banking risk

index have a nearly 1:1 effect on the sovereign risk premium but fiscal factors have a negligible

effect. However, it is possible that the fiscal variables are extremely important determinants of

sovereign CDS rates, but due to measurement error and misspecification, their role in the

estimated linear model could be subsumed by other explanatory variables.

For illustrative purposes of comparison, let us assume two possible test cases: first, if the

current account balance changes by the mean change in current account balances in the sample

and the second if the BRI changes by the mean amount of change in the BRI for the observations

in the sample. In the first case, the current account balance increases by €125 billion which is

correlated with a decrease of 0.22 basis points in the sovereign risk premia. Given that the mean

of the CDS premia across all countries and observations in the sample is 125 basis points, this

Page 43: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    43  

marks an infinitesimal effect of a substantial increase in the current account balance. In the

second case, the BRI rises by a mere 2.2 basis points causing an increase of 2.6 basis points in

the sovereign risk premia. Clearly, banking sector risk (measured through the banking risk index)

is the dominant statistically significant predictor of sovereign CDS rates13.

The control variables for global and country-level external shocks have expected

coefficients with both VIX and US CDS risk premia carrying a positive and statistically

significant coefficient. The normalized level of the stock market also has a statistically

significant coefficient which is negative in magnitude: when the value of the stock market goes

up by 1%, the value of sovereign CDS decreases by 0.47 basis points. However, the stock market

variable is also insignificant when economic magnitudes are accounted for. For example, a 1%

increase in the normalized value of Italy’s FTSE MIB (which implies an increase of about 200

points on the actual index) corresponds to only a 0.005% or 0.5 basis points decrease in Italy’s

sovereign CDS risk premia. This 0.5 basis points decrease in the risk premia seems even smaller

when it is considered in light of Italy’s average risk premia over the period- close to 200 basis

points. Having said that, note that the coefficient remains statistically significant and country

specific financial shocks that may be reflected in the stock market are statistically significantly

correlated with the values of the sovereign risk premia.

Lagged values of the current account balance and for the ratio of outstanding securities to

GDP are statistically significant for all regressions shown above. Lagged values are used since it

is unlikely that these fiscal variables will have a contemporaneous effect on sovereign risk.

These fiscal variables are neither known to market participants contemporaneously nor are their

                                                                                                               13 Note that the example is for illustration only. It has been pointed out that since changes in both the cases mentioned may be positive or negative, the mean may not provide a good choice for comparison across variables. Similar results hold for comparisons across other measures including but not limited to standard deviation and mean +standard deviation. The mean is used above for the sake of simplicity.

Page 44: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    44  

effects in the economy simultaneous. Regression (3) used the first order lags while regression (4)

uses the second order lags.

Returning to the issue of the extent of the effect of fiscal factors, robustness analyses

support the hypothesis of the primacy of banking sector factors and the relatively negligible

effect of fiscal variables in predicting sovereign CDS rates. To begin with, the measures used for

these fiscal factors can be tweaked. Regressions 6 and 7 below use the trailing 6 month average

of the current account balance and the squares of the 6-month average to determine the

magnitude of the fiscal effects on sovereign risk premia.

Table 3: Regression of levels of Sov. risk on levels of BRI and trailing average fiscal variables Regression (6) (7) (8) Sovereign CDS rate Coefficient Coefficient Coefficient Banking risk Index 1.207344***

(0.0507) 1.184464*** (0.120)

1.26*** (0.061)

Lag 1 of BRI -- 0.0983429 (0.194)

--

Lag 2 of BRI -- 0.1522654 (0.1732)

--

Lag 3 of BRI -- -0.1887197* (.1043)

--

VIX 3.822551*** (0.5046)

4.515825*** (0.963)

5.15*** (0.83)

US CDS rate 1.053591*** (0.2667)

0.821673*** (0.217)

0.89*** (0.183)

Normalized equity index levels

-0.4767804* (0.276)

-0.4687714 (0.2902)

-0.044 (0.25)

6-month trailing current account balance

-0.00276*** (0.000931)

-0.002538*** (0.000906)

-0.00131*** (0.0004)

Lag 1 of outstanding securities-to-GDP ratio

0.795872*** (0.1792)

0.8431258*** (0.1946)

--

Lag 2 of outstanding securities-to-GDP ratio

-- -- 1.01*** (0.181)

Page 45: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    45  

Square of 6-month trailing current account balance

2.02E-07 (2.11E-07)

2.12E-07 (9.96E-07)

--

Constant -35.70054 (48.529)

-30.2423 (55.33377)

-108.52** (48.17)

R2 0.9126 0.9212 0.9146 The results of the regressions above support the earlier hypothesis that the magnitudes of

the coefficients of the fiscal variables render them a comparatively small determinant of

sovereign risk premia. In regression (6) above, a €1 billion increase in the current account

balance would only cause a 0.00276 basis points decrease in the associated sovereign risk

premia. The multiplier effect of changes in the BRI (coefficient greater than 1) on sovereign

CDS rates is observed again and regression (7) again tests the over-reaction explanation for this.

Linear combinations test results rejected the hypothesis that the net effect of an increase in the

BRI on sovereign risk premia was zero. Given the insignificance of the square of the trailing

current account deficit, regression (8) excludes this term and finds the same results. Overall, this

supports the view that when banking risk factors are included, fiscal factors play only a small

role as determinants of sovereign risk spreads.

Overall, the results in this section can be summarized as follows:

a. The Banking Risk Index is the dominant statistically significant predictor of the levels of

sovereign risk premia.

b. After controlling for the Banking Risk Index, fiscal factors play a negligible role in

explaining sovereign CDS rates.

c. The effect of BRI increases on sovereign risk premia may be amplified by a multiplier

effect that leads to a 1 basis point increase in the BRI being correlated with more than a 1

basis point contemporaneous increase in the sovereign risk premia.

Page 46: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    46  

d. Country level and global control variables for financial market conditions (and especially

external shocks) are also statistically significant.

IV.C.1. “Joined-at-the-hip” hypothesis

The “joined-at-the-hip” hypothesis is that as the default risk of the banks and their

sovereign rises, their respective CDS rates become even more highly correlated with each other.

This hypothesis is consistent with the existence of a feedback loop between sovereign and

banking risk, by which the credit strength of sovereign depends significantly on the sovereign’s

ability to rely on its banks as a source of financing, and vice versa. This feedback effect would

heighten the correlation of their credit quality as either the banks deteriorate, or their sovereign

deteriorates, or both. To be more specific, there are two distinct sub-hypotheses within this larger

hypothesis.

IV.C.1.a Methodology for sub-hypothesis 1: The first sub-hypothesis states that the correlation

between banking and sovereign risk is increases in cases where banking risk in a given country

increases. Since we have already proved that banking risk is a major correlate of sovereign risk

premia, the transitive property implies that the levels of banking risk, and through it sovereign

risk, is a determinant of the correlation between sovereign and banking risk. This hypothesis

therefore takes banking risk premia as a proxy for the levels of sovereign and banking risk and

regresses the correlation between these two measurements on this proxy. This correlation seeks

to explain the determinants of the correlation between sovereign and banking risk. That is, are

banks and sovereigns more closely tied (correlated) in times of greater financial sector risk?

The correlation between banking and sovereign risk is measured as the correlation

between sovereign CDS premia and the banking risk index. Using correlations between the

banking sector and the sovereign rather than individual banks and sovereigns allows us to ensure

Page 47: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    47  

that the results are not confounded by events affecting the credit quality of individual banks that

may not necessarily reflect the health of the financial system as a whole. Figures 16 and 17

provide a graphical representation of this correlation coefficient across the various countries

included in the sample.

The associated model specification is:

Wherein:

• i represents the sovereign

• t represents time t in weeks and t’ represents the trailing 6-month average

• Corr_bank_sov is the sample correlation between BRI and sovereign CDS premia over

the six-month period before time t.

• log_CDS_Bank represents the natural logarithm of the BRI

• US_CDS, VIX, N_Stock_ex and iTraxx_Europe are defined as in earlier specifications.

-­‐0.4  

-­‐0.2  

0  

0.2  

0.4  

0.6  

0.8  

1  

1   18  

35  

52  

69  

86  

103  

120  

137  

154  

171  

188  Correlation  coef-icient  

Week  number  

Figure  16:  Correlation  between  BRI  and  sov  CDS  over  time  

France  and  French  banks  Germany  and  German  Banks   0  

0.2  

0.4  

0.6  

0.8  

1  

1   16  

31  

46  

61  

76  

91  

106  

121  

136  

151  

166  

181  

Correlation  coef-icient  

Week  Number  

Figure  17:  Correlation  between  BRI  and  sov  CDS  over  time  

Italy  and  italian  Banks  

Spain  and  Spanish  Banks  

Corr _bank _ sovi,t ' =α + β log_CDS _Banki,t + χUS _CDSt +φVIXt

+ϕN _ stock _ exi,t + ιiTraxx _Europet + ε i,t

Page 48: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    48  

The above specification may lead to various problems in the estimation: first, given that

subsequent values of the dependent variable are correlated14, the standard errors are likely to be

biased with possible clustering. Newey-West standard errors are used to improve the robustness

if the standard error estimates. Secondly, the values of the dependent variable are bound between

-1 and 1, which makes the interpretation of effects more difficult. Therefore, the unit of

correlation is changed to percentage in this estimation. Moreover, some control variables like US

CDS premium may be highly correlated with sovereign risk in the Eurozone and banking risk

premia and therefore, also with the correlation coefficient estimated above. Regressions are also

carried out without the inclusion of such controls to ensure the robustness of the results.

The hypothesis suggests a non-linear relationship: at very low levels of banking risk, the

hypothesis suggests that the correlation between the two terms is still expected to be positive. On

the other hand, the sample correlation for very high levels of banking risk is expected to be much

higher. There is an upper bound to the correlation as it cannot exceed 1 while it is still expected

to be close to 1 at very high values of banking risk, so the increase in correlation is expected to

taper off after a certain point. Figures 18-21 below graphically show the relationship between the

two measures for all countries in the dataset. These graphs are also similar to the shape of linear-

log relationship and provide another reason to test the above specification.

                                                                                                               14 This is because the 6-month trailing averages on a weekly basis have an overlap of 5 months and 3 weeks in the sample for which they estimate the correlation.

Page 49: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    49  

IV.C.1.b Results for sub-hypothesis 1: Given that we are estimating a linear-log regression for

this specification, the coefficient for the term with the natural log (the CDS_bank) has to be

interpreted differently. A 1% increase in independent variable (that is the log term) or the risk

premium for banks leads to an increase of β/100 units in the correlation coefficient, where β is

the coefficient for the bank CDS term. The log function is confined to only one independent

variable and the coefficients of the other right hand side variable can be interpreted in the usual

-­‐0.4  -­‐0.3  -­‐0.2  -­‐0.1  0  

0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   100   200   300   400  

Trailing  correlation  coef-icient  

Banking  Risk  Index  (bps)  

Figure  18:  Relationship  between  trailing  correlation  and  BRI  for  France  

-­‐0.2  -­‐0.1  0  

0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   50   100   150   200   250   300  

Trailing  correlation  coef-icient  

Banking  Risk  Index  (bps)  

Figure  19:  Relationship  between  trailing  correlation  and  BRI  for  Germany  

0  0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   100   200   300   400   500   600  

Trailing  correlation  coef-icient  

Banking  Risk  Index  (bps)  

Figure  20:  Relationship  between  trailing  correlation  and  BRI  for  Italy  

0  0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   100   200   300   400   500   600  

Trailing  correlation  coef-icient  

Banking  Risk  Index  (bps)  

Figure  21:  Relationship  between  trailing  correlation  and  BRI  for  Spain  

Page 50: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    50  

way similar to OLS. Remember that the correlation term here is expected to take values between

-100 and 100. The Table 4 below provides estimates of the regression as detailed above:

Table 4: Regressions of trailing correlations on BRI

Regression (9) (10) (11) Correlation between sovereign CDS and BRI in %

Coefficient Coefficient Coefficient

Log of BRI 28.87497*** (2.912)

24.96431*** (2.893)

15.94*** (1.33)

Normalized equity index levels

0.062841 (0.046)

0.0072216 (0.046)

--

US CDS rate -0.5724186*** (0.098)

-- --

VIX 0.5314094*** (0.1528)

0.365623** (0.1532)

--

iTraxx_Europe -0.178319*** (0.044)

-0.195337*** (0.0446)

--

Constant -44.86393*** (16.27322)

-36.73865** (16.553)

-5.38 (6.81)

R2 0.2077 0.1741 0.1551 In regression (9) above, the coefficient for the log of the Banking Risk Index is

approximately 28.9, implying that a 1% change in the BRI leads to a 0.289 units change in the

correlation. Given that the correlation coefficient is measured in percentage points here, this

means that a 1% increase in the BRI corresponds to a 0.289% increase in the correlation

coefficient. Therefore, the coefficient for the banking risk index is not only statistically

significant in the determination of the correlation coefficient but its magnitude also implies that it

forms a significant part of the determination.

As mentioned above, there is some concern that the high magnitude for the banking risk

index may come as a result of the inclusion of certain controls like the US CDS risk premium,

which may have confounded the regression. Regression (10) above estimates the same

specification but after excluding the US CDS variable. The results of regression (10) confirm

Page 51: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    51  

that the Banking Risk Index plays a statistically significant in the determination of the correlation

between banking and sovereign risk. Regression (11) above removes all control variables to

present a simple linear correlation between the correlation coefficient and the BRI. This

regression is provided only as an illustration.

Note here that the joined-at-the-hip hypothesis wherein the correlation coefficient is

determined by the level of banking risk is not rejected here. In fact, the Banking Risk Index has a

statistically significant and large magnitude in the determination of the correlation coefficient.

Having said that, the explanatory power of these regressions is low suggesting that there remains

further work to be done in explaining the correlation between sovereign and banking risk.

IV.C.2.a. Methodology for sub-hypothesis 2: The second sub-hypothesis states that the

correlation depends on the interaction between sovereign and banking risk. To put it simply, the

correlation between banking and sovereign risk premia is correlated with the levels of both.

Therefore, if either the levels of sovereign or banking risk are heightened (they usually happen

but not always) then the correlation between these measures will also be higher. Under this

hypothesis, the product of sovereign and banking CDS premia are used to measure the jointed

levels of sovereign and banking risk. The correlation coefficient between these two variables

increases with an increase in their joined levels. Note here that the correlation coefficient is

between individual bank CDS levels and sovereign CDS levels, unlike the previous specification

wherein the correlation was between the Banking Risk Index and the Figures 22-26 below

provide a graphical representation of this relationship for each of the Italian banks included in

the sample. Similar figures (Figures C.1- C.15) for other countries in the sample can be found in

Appendix C below.

Page 52: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    52  

0  0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   5   10   15   20   25   30  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  22:  Correlation-­‐levels  graph  for  Italy  and  Bank  1  

-­‐0.2  -­‐0.1  0  

0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   5   10   15   20  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  23:  Correlation-­‐levels  graph  for  Italy  and  Italian  Bank  4  

0  0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   10   20   30   40  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  24:  Correlation-­‐levels  graph  for  Italy  and  Italian  Bank  5  

-­‐0.6  

-­‐0.4  

-­‐0.2  

0  

0.2  

0.4  

0.6  

0.8  

1  

0   10   20   30   40   50  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  25:  Correlation-­‐levels  graph  for  Italy  and  Italian  Bank  6  

0  0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   5   10   15   20   25   30  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  26:  Correlation-­‐levels  graph  for  Italy  and  Italian  Bank  6    

Page 53: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    53  

Again, it is clear from the above graphs that while a relationship between the two

variables does exist, it cannot be described with the simple linear regression. While a linear trend

line is included in the figures above for illustrative purposes, the linear trend does not seem to

describe the relationship satisfactorily. Again, the shape of the distribution appears somewhat

similar to that of the linear-log model that was used in the estimation of the previous sub-

hypothesis. Using the linear-log model again, the specification for this hypothesis becomes:

Wherein

• i represents the sovereign

• j represents the bank

• t represents the time in weeks and t’ represents the 6-month trailing average

• Corr_bank_sov represents the correlation between the CDS premium for the sovereign i and the individual bank j.

• Log(CDS_Bank*CDS_Sov) represents the log of the product of the CDS premia of

sovereign i and bank j.

• Control variables remain the same as specified earlier.

Similar to the previous specification, the linear-log model specified above means that for

every 1% change in the log variable (product of sovereign and bank risk premium), the

dependent variable changes by β/100 units. The results of the regressions are presented in Table

5 below:

Corr _bank _ sovi, j ,t ' =α + β log(CDS _Bank *CDS _ sov)i, j ,t + χUS _CDSt+φVIXt +ϕN _ stock _ exi,t + ιiTraxx _Europet + ε i, j ,t

Page 54: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    54  

Table 5: Regressions of trailing correlations on product of bank and sov. CDS

Regression (12) (13) (14) Correlation between sovereign CDS and individual bank CDS in %

Coefficient Coefficient Coefficient

Log(CDS_bank* CDS Sov.)

7.677842*** (0.555)

5.297189*** (0.54)

6.170036*** (0.367)

Normalized equity index

0.1171575*** (0.032)

-0.0090796 (0.03)

--

US CDS rate -0.7690482*** (0.0586)

-- -0.7065022*** (0.056)

VIX 0.2279812*** (0.078)

-0.0043267 (0.077)

0.1067689 (0.07)

iTraxx_Europe 0.0837815*** (0.0177)

0.0359987** (0.078)

0.089802*** (0.018)

Constant -0.6685156 (8.015)

15.04357* (8.1)

25.53288*** (3.4)

R2 0.1301 0.092 0.1275 In regression (12) above, each 1% change in the product of the bank and sovereign’s

CDS increases the correlation coefficient by 0.077%. The results are similar in specifications

when certain control variables are excluded. However, it should be noted here that the predictive

power of the regressions reflected in the R2 coefficients is lower and the magnitude of the effect

of the product of banking and sovereign risk is also lower as compared to the Banking Risk

Index.

Overall, the following conclusions can be drawn with regards to the third hypothesis:

a. The Banking Risk Index is a statistically significant determinant of the correlation

between banking and sovereign risk.

b. There is a linear-log relationship between the correlation coefficient and the BRI.

c. The BRI has greater magnitude as a determinant of the correlation that the product of

banking and sovereign risk.

Page 55: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    55  

d. Overall, the joined-at-the-hip hypothesis is not rejected: the correlation between banking

and sovereign risk is higher in times of greater banking risk.

Section V: Conclusions

On December 3, 2012, the Kingdom of Spain officially requested an approximately €40

billion bailout for its banking sector. Subsequently, CDS risk premia on the 2 largest Spanish

banks, which received a majority of the bailout package decreased by more than 10%. Countries

ranging from Ireland to the United Kingdom, from the United States to Cyprus have sought to

support their domestic financial sectors. In some cases, the nationalization of a bank such as

Anglo-Irish acted as a signal to the markets that the respective financial sectors of the countries

represent a potential contingent liability for the sovereign. In other cases, support for banks has

led to improvement in the credit conditions for both banks and sovereigns (as with the case of

Spain). Changes in the perceived creditworthiness of sovereigns have also adversely affected

bank holdings (through their holdings of sovereign debt) and bank funding conditions. The exact

nature of the link between sovereign and banking sector risk, therefore, is an important question

not only in analyzing the crisis but also in attempts to solve it.

This paper makes two unique contributions in this regard: first, this paper shows that

during the recent European financial crisis, financial sector risk has been the primary statistically

significant predictor of sovereign risk premia. This finding stands in stark contrast to a part of the

literature in this field that claims that country level fiscal factors are the primary determinant of

sovereign risk (Greenlaw, Hamilton, Hooper and Mishkin, 2013). While country-level fiscal

factors are also statistically significant, their economic magnitudes render them a negligible part

of the determination of sovereign risk premia. An increase in the Banking Risk Index of the

Page 56: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    56  

country is found to be associated with a 1:1 increase in the sovereign risk premia, highlighting

the primacy of banking risk in the determination of sovereign risk premia. This phenomena is

here termed “bank dominance” but can more fully be understood as the financial sector

dominance in the determination of sovereign risk premia.

Second, this paper finds that the correlation between banking and sovereign risk is

dependent on the level of banking risk. There is a linear-log relationship between the correlation

between banking and sovereign risk and the level of banking risk. This linear-log relationship is

expected: the correlation between banking and sovereign risk is expected to be positive and high

regardless of the levels of banking risk but is expected to be especially high in times of financial

distress. However, the correlation coefficient can, of course, not exceed 1 and thus the

“heightened” correlation tapers off after a point. It is found that the level of banking risk and the

joint level of banking and sovereign risk are both statistically significant predictors of the

correlation.

The results of this paper show that the financial health of European sovereigns in credit

markets is determined by the perceived health of their respective banking sectors. This joint

determination of banking and sovereign credit risk is especially strong in times of financial

sector distress. These results point to the importance of measures seeking to ensure the health of

the banking sector. A decrease in the creditworthiness of the banks in a given country not only

threatens the economy of the country but also threatens the sovereign’s standing in global credit

markets—it not only causes a financial crisis but also impacts the ability of the sovereign to

come out of it by weakening its creditworthiness. Further work on how the creditworthiness of

the banking sector can be maintained is needed but lies outside of the scope of this paper. Suffice

to say here, that the primacy of bank risk in the determination of sovereign risk premia should

Page 57: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    57  

put financial sector regulation at the top of any economic agenda. To conclude, in the continuing

state of financial peril across the world and especially in the Eurozone, a necessary condition for

financial stability is a sound banking system.

Page 58: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    58  

Appendix A: List of banks included in the dataset

Country Bank Name Bank asset size (in million E, 2008)

France BNP Paribas 2,075,551 France Credit Agricole 1,653,220 France Credit Lyonnais 98,437 France Societe Generale 1,130,003 France Natixis 555,760 Germany Deutsche Bank AG   2,202,423 Germany Commerzbank AG                    625,196 Germany Landesbank Baden-Württemberg                  447,932 Germany Bayerische Landesbank                  421,666 Germany Norddeutsche Landesbank Giro -GZ                  244,265 Italy Intesa SanPaolo SpA 636,133 Italy Unicredit SpA 1,045,611 Italy Banca Monte Dei Paschi Di Siena SpA 213,796

Italy Banco Popolare SC 121,375 Italy Unione di Banche Italiane (UBI Banca) 121,955 Spain Banco Santander S.A. 1,049,632 Spain Banco Bilbao Vizcaya Argentaria S.A.

(BBVA) 542,650

Spain Caixa Pensiones De Barcelona 260,827 Spain Banco Popular Espanol S.A. 110,376 Spain Banco De Sabadell S.A. 80,378

Page 59: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    59  

Appendix B: Test results for “Co-movement hypothesis” Note that these results correspond to the results of regression (1).

Table B.2- Linear combination of first difference and all lags of BRI . lincom D_cds_bank+ L1_d_cds_bank+ L2_d_cds_bank+ L3_d_cds_bank ( 1) D_cds_bank + L1_d_cds_bank + L2_d_cds_bank + L3_d_cds_bank = 0 D_cds_sov Coef. Std. Err. t P>t [95% Conf. Interval]

(1) .5731173 .0703336 8.15 0.000 .4350681 .7111664

Table B.3- Linear combination of first difference and all lags of US CDS premia . lincom D_US_CDS+ L1_d_US_CDS+ L2_d_US_CDS+ L3_d_US_CDS ( 1) D_US_CDS + L1_d_US_CDS + L2_d_US_CDS + L3_d_US_CDS

D_cds_sov Coef. Std. Err. t P>t [95%

Conf. Interval]

(1) .5419504 .1773974 3.06 0.002

0.1937584 0.8901423

Table B.1- Linear combination of first difference and first lag of BRI . lincom D_cds_bank+ L1_d_cds_bank ( 1) D_cds_bank + L1_d_cds_bank = 0 D_cds_sov Coef. Std.Err t P>t [95% Conf. Interval] (1) .4504264 .0494033 9.12 0 0.3534588 0.5473939

Page 60: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    60  

Appendix C: Graphs for the “joined-at-the-hip” hypothesis

0  

0.2  

0.4  

0.6  

0.8  

1  

1.2  

0   5   10   15   20   25  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.1:  Correlation-­‐levels  graph  for  Spain  and  Bank  1  

-­‐0.4  

-­‐0.2  

0  

0.2  

0.4  

0.6  

0.8  

1  

0   10   20   30   40   50  Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.2:  Correlation-­‐levels  graph  for  Spain  and  Bank  2  

0  0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   5   10   15   20   25  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.3:  Correlation-­‐levels  graph  for  Spain  and  Bank  3  

-­‐0.6  

-­‐0.4  

-­‐0.2  

0  

0.2  

0.4  

0.6  

0.8  

1  

0   10   20   30   40  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.4:  Correlation-­‐levels  graph  for  Spain  and  Bank  4  

-­‐0.4  

-­‐0.2  

0  

0.2  

0.4  

0.6  

0.8  

1  

0   5   10   15   20   25  Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.5:  Correlation  levels  graph  for  Spain  and  Bank  6  

Page 61: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    61  

Germany

-­‐0.3  -­‐0.2  -­‐0.1  0  

0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   0.5   1   1.5   2  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.6:  Correlation-­‐levels  graph  for  Germany  and  German  Bank  1  

-­‐0.4  -­‐0.3  -­‐0.2  -­‐0.1  0  

0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   0.5   1   1.5   2   2.5  Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.7:  Correlation-­‐levels  graph  for  Germany  and  Germany  Bank  2  

-­‐0.1  0  

0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   0.5   1   1.5   2   2.5  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.8:  Correlation-­‐levels  graph  for  Germany  and  German  Bank  3  

-­‐0.2  -­‐0.1  0  

0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   0.5   1   1.5   2   2.5  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.9:  Correlation-­‐levels  graph  for  Germany  and  German  Bank  4  

-­‐0.2  -­‐0.1  0  

0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   0.5   1   1.5   2   2.5  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.10:  Correlation-­‐levels  graph  for  Germany  and  German  Bank  5  

Page 62: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    62  

France

0  0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   1   2   3   4   5  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.11:  Correlation-­‐levels  graph  for  France  and  French  Bank  1  

-­‐0.2  -­‐0.1  0  

0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

-­‐1.5   0.5   2.5   4.5   6.5  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.12:  Correlation-­‐levels  graph  for  France  and  French  Bank  2  

0  0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   2   4   6   8  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.13:  Correlation-­‐levels  graph  for  France  and  French  Bank  3  

-­‐0.2  -­‐0.1  0  

0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   1   2   3   4   5   6  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.14:  Correlation-­‐levels  graph  for  France  and  French  Bank  4  

-­‐0.8  -­‐0.7  -­‐0.6  -­‐0.5  -­‐0.4  -­‐0.3  -­‐0.2  -­‐0.1  0  

0.1  0.2  0.3  0.4  0.5  0.6  0.7  0.8  0.9  1  

0   2   4   6   8  

Correlation  coef-icient  

Product  of  bank  (%)  and  sovereign  (%)  risk  premium  

Figure  C.15:  Correlation-­‐levels  graph  for  France  and  French  Bank  5  

Page 63: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    63  

Bibliography Angeloni and Wolff. (2012). Are banks affected by their holdings of Government Debt? Bruegel Working Papers, no. 2012/07, Bruegel. Arezki, Candelon and Sy. (2011). Sovereign rating news and financial market spillovers: evidence from the European debt crisis. IMF Working Papers, no 11/68.

Attinasi, Checherita and Nickel. (2009). What explains the surge in Euro area sovereign spreads during the Financial crisis of 2007-09? European Central Bank, working paper no. 1131.

Bank of International Settlements. (2011). The Impact of sovereign credit risk on bank funding conditions. CGFS Paper, No. 43, BIS. Bank of International Settelements. (2013). Financial crises and bank funding: recent experience in the Euro area. BIS Working Paper No. 406. Beck and Katz. (1995). What to do (and not to do) with Time-Series Cross-Section Data, The American Political Science Review , Vol. 89, No. 3 (1995), pp. 634-647.

Blundell-Wignall, A. and P. Slovik. (2010). The EU Stress Test and Sovereign Debt Exposures, OECD Working Papers on Finance, Insurance and Private Pensions, No. 4, OECD Financial Affairs Division. Brunnermeier and Pedersen. (2009). Market liquidity and funding liquidity, Review of Financial Studies 22(6): 2201–38. Campbell and Ammer. (1993). What Moves the Stock and Bond Markets? A Variance Decomposition for Long-Term Asset Returns, The Journal of Finance, Vol. 48, No. 1 (Mar., 1993), pp. 3-37. Committee on the Global Financial System (2010): “The role of margin requirements and haircuts in procyclicality”, CGFS Papers, no 36, Basel. Copeland and Walker. 2010. The Tri-Party Repo Market before the 2010 Reforms. Federal Reserve Bank of NewYork, Staff Report No. 477. Correa, Lee, Sapriza and G Suarez. (2011). Sovereign credit risk, banks’ government support, and bank stock returns around the world, Federal Reserve System, IFDP 1069.

Demirgüç-Kunt and Huizinga. (2010). Are banks too big to fail or too big to save? International evidence from equity prices and CDS spreads CentER Discussion Paper Series, No. 2010-59. Du, Songzi and Zhu, Haoxiang. (2012). Are CDS Auctions Biased? (August 22, 2012). doi: http://dx.doi.org/10.2139/ssrn.1804610

Page 64: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    64  

Duffie,D. (1999). Credit Swap Valuation Financial Analysts Journal 1999, January-February, Pages 73-87.

Duffie, D. (2010). How Big Banks Fail: and What to Do about It, Princeton University Press. Duffie and Thukral. (2012). Fixing the Flaw in Sovereign CDSs, Risk Magazine, July, 2012. Ejsing and Lemke. (2009). The Janus-headed salvation: sovereign and bank credit risk premia during 2008−09, ECB Working Papers, No. 1127.

European Central Bank. (2012) Impact of the Two Three-Year Longer-Term Refinancing Operations. Monthly Bulletin, March (Frankfurt am Main: European Central Bank).

Financial Times. (2009). Dublin Nationalizes Anglo-Irish Bank, January 16, 2009.

Fontana and Scheicher. (2010). An Analysis of Euro Area Sovereign CDS and Their Relation with Government Bonds, ECB Working Paper, No. 1271. Gerlach, Schulz and Wolff. (2010). Banking and Sovereign Risk in the Euro Area, CEPR Discussion Paper, No. 7833. Gray. (2009). Modeling Financial Crises and Sovereign Risk. Annual Review of Financial Economics 1. Greenlaw, Hamilton, Hooper and Mishkin. (2013). Fiscal Crises and the Role of Monetary Policy, U.S. Monetary Policy Forum, New York City, February 22, 2013. Hodson, Dermon. (2011). The Eurozone in 2011, Journal of Common Market Studies, vol. 50(s2), pages 178-194. Longstaff, Pan, Pedersen & Singleton. (2011). How Sovereign Is Sovereign Credit Risk?, American Economic Journal: Macroeconomics, American Economic Association, vol. 3(2), pages 75-103.

Mody, A. (2009). From Bear Sterns to Anglo Irish: How Eurozone sovereign spreads related to financial sector vulnerability, IMF Working Papers, 09/108. Mody and Sandri. (2011). The Eurozone crisis: How banks and sovereigns came to be joined at the hip, IMF Working Papers, 11/269. Monfort, A. & Renne, J-P. 2011. Credit and liquidity risks in Euro-area sovereign yield spreads, Working Paper 352, Banque de France. Moody’s Analytics. (2011). The Paradox of Europe: Many solid banks, but even more weak credit market trading levels, Capital Markets Research Group, January 26, 2011. Pan and Singleton. (2008). Default and Recovery Implicit in the Term Structure of Sovereign CDS Spreads. Journal of Finance, volume 63, pages 2345-2384.

Page 65: Written under the direction of Darrell Duffie Abstract · Wright, Chiara Angeloni, Peter Blaustein, Valentin Bolotnyy, Chris Seewald, Vivek Viswanathan and Martin Schneider for comments

May 13, 2013 Bank dominance: Financial sector determinants of sovereign risk premia Thukral, Mohit

    65  

Panetta,  F.  Faeh,  T.  Grande,  G.  Ho,  C.,  King,  M.  Levy,  A.,  Signoretti,  F.,  Taboga,  M  and  A.  Zaghini  (2009)  An  assessment  of  financial  sector  rescue  programmes.  BIS  Paper  48. Pagano and Von Thadden. (2004). The European Bond Markets under the EMU, Oxford Review of Economic Policy, Volume 20(4). Packer, F., and Zhu, H. (2005). Contractual terms and CSD pricing, BIS Quarterly Review, March 2005, pp. 89-100. Reinhardt, C and K. Rogoff. (2012). This time is different: Eight centuries of financial folly, Princeton University Press.

Reuters. (2012). “Banks gorge on 530 billion Euros of ECB Funds” Feb 29, 2012. Rime, B. (2005). Do Too-Big-To-Fail Expectations Boost Large Banks Issuer Ratings. Swiss National Banks, Systemic Stability Section: Working Paper (May 9).

Sgherri and Zoli. (2009).Euro area sovereign risk during the crisis, IMF Working Paper, 09/222