How the Subprime Crisis Went Global: Evidence from Bank Credit Default Swap Spreads

TL;DR: This article used principal component analysis to identify common factors in the movement of banks' credit default swap spreads and found that the importance of common factors rose steadily to exceptional levels from the outbreak of the Subprime crisis to past the rescue of Bear Stearns, reflecting a diffuse sense that funding and credit risk was increasing.

Abstract: How did the Subprime Crisis, a problem in a small corner of U.S. financial markets, affect the entire global banking system? To shed light on this question we use principal component analysis to identify common factors in the movement of banks’ credit default swap spreads. We find that fortunes of international banks rise and fall together even in normal times along with short-term global economic prospects. But the importance of common factors rose steadily to exceptional levels from the outbreak of the Subprime Crisis to past the rescue of Bear Stearns, reflecting a diffuse sense that funding and credit risk was increasing. Following the failure of Lehman Brothers, the interdependencies briefly increased to a new high, before they fell back to the pre-Lehman elevated levels - but now they more clearly reflected heightened funding and counterparty risk. After Lehman’s failure, the prospect of global recession became imminent, auguring the further deterioration of banks’ loan portfolios. At this point the entire global financial system had become infected.

Summary

1 Introduction

• One enduring question about the financial turbulence that engulfed the world starting in the summer of 2007 is how problems in a small corner of U.S. financial markets– securities backed by subprime mortgages accounting for only some 3 per cent of U.S. financial assets– could infect the entire U.S. and global banking systems.
• More importantly, the common component of CDS spreads became more highly related with measures of funding and credit risk as measured by spreads in the assetbacked commercial paper market and LIBOR minus the overnight index swap.
• These characterizations are likely to be the basis for defining and probing more subtle hypotheses.
• In an unfocused sense, Lehman’s failure caused that common risk to be more concretely identified with both developments in the real economy and specific problems in the financial system.
• In Section 3, the authors consider the possibility of additional spillovers from inter-bank exposures that go beyond the common movements identified by the latent factors.

2 Common Factors in CDS Spreads

• The authors start by decomposing the change in CDS spreads of N=45 global banks into common and idiosyncratic components.
• The term “banks”is used throughout in this paper, although some insurance companies are also included in the sample.
• The data are 5-year CDS spreads, as the five-year maturity is the most widely traded.
• The authors use end-of-day quotes from the New York market for payment in U.S. dollars based on U.S. dollar-denominated notional amounts.

2.1 Preliminary Data Analysis

• Average spreads over the period vary significantly across banks (from a low of 17 for Rabobank to a high of 101 basis points for AIG).
• The authors interest is not so much in the cross-sectional variation at this stage, however, as in the variation over time, which has been substantial.
• In contrast, the subsequent rise in spreads was dramatic with twin peaks corresponding to the Bear Stearns rescue and the Lehman Brothers failure.
• For U.S. banks, a high of 417 basis points was reached following the severe stress after the Lehman failure during the week of October 1, 2008; the median spread then moderated to 268 basis points in the last week of November 2008.
• 6Some, evidently, knew about the extent of its leverage.

2.2 A Dynamic Factor Model of CDS Spreads

• The first question the authors ask is whether the movements in spreads reflected common drivers.
• The estimation procedure allows for εi,t to be cross-sectionally and time correlated and heteroskedastic.
• As Bai and Ng (2002, 2008) and Stock and Watson (2002) show, the principal component (PC) estimator enables us to identify factors up to a change of sign and consistently estimate the factors space up to an orthonormal transformation.
• At each recursion an AR(p) model is applied to each series, where the order, p is determined using the individual partial autocorrelation function (PACF) and residuals from the AR(p) model are used as the filtered series.
• In general, a richer dynamic factor model of CDS spreads would allow explicitly for time-varying, stochastic volatility and correlations, and could be estimated by Markov Chain Monte Carlo (MCMC) methods.9.

2.3 Estimation Results and Discussion

• Figure 2 shows changes over time in the contributions of the common factors to the total variation in the CDS data, obtained from the estimated factors.
• The importance of the common factors continued to increase following the Bear Stearns rescue, reaching a new high in May 2008, at which point the first common factor explained almost 60 percent of the variance of CDS spreads.
• Then, the period between May and September 2008 was one of general weakness of financial-market indicators.
• ”This is supported by the results obtained by running the Onatski’s (2010) criterion for determination of the number of factors in the data through a grid of parameter values.
• To get a sense of whether the degree of commonality the authors observe for international banks is high or low, they can compare these results with those of Longstaff et al. (2010) for sovereign CDS spreads.

4 Correlating Latent Factors with Observed Financial Variables

• The next step is to examine the relation between the latent factors identified in Section 2 and the observed financial variables.
• While the exact association of a financial variable with any one of the estimated factors is hard to define due to non-uniqueness of the factor estimates, the authors can measure the association of financial variables with the entire set of estimated factors and investigate under which conditions correlations with individual factors are still informative.

4.1 Some Statistical Considerations

• Bai and Ng (2006b) develop statistical criteria which can be used to investigate whether any of the candidate series yields the same information that is contained in the factors.
• The criteria resemble the well-known likelihoodbased selection criteria BIC and HCC, using the GMM J-statistic for testing the over-identifying restrictions.
• Once this number of factors is determined, the individual correlations between factors and the observed series can be examined.
• In particular, the R2 estimates the authors highlight below are meaningful measures of the relationships of interest under moderate levels of noise.

4.2 Correlates

• The authors limit their attention to U.S. variables, since the corresponding European variables are highly correlated with U.S. series.
• The test is performed for the full sample and two subsamples (up to July 2007 and up to May 2008) in order to examine whether the most recent period (with possible outliers) influences the results.
• In the first column of Table 2 the authors can see that none of the proposed series is correlated with the idiosyncratic part of CDS spreads since the frequency of rejections of the null among all randomizations of the data is very small for all samples.
• This implies that the authors can use the moment selection criteria to investigate the relationship between the observed series and the factors.

4.3 The Real Economy Prior to the Subprime Crisis

• In the “real economy” group, the authors consider three correlates.
• Prior to the start of the Subprime Crisis, the HYS and the VIX trended down along with the median CDS spreads.
• 22For the full sample, the various criteria (namely, BIC1, BIC2, BIC3 and HQQ) suggest the existence of a relationship between the series and four factors.
• In contrast, stock returns include both upside and downside movements: while high stock returns presumably lower risk to a degree, banks’risks are apparently more clearly defined by downside risks as reflected in HYS.
• Hence returns had a much weaker association with spreads’ movements.

4.4 The Emergence of Financial Factors

• Thus, prior to the Subprime Crisis, global economic factors as summarized in HYS were the main drivers of the commonality in CDS spreads of international banks.
• Following the onset of the crisis and through the Bear Stearns bailout, however, the association with the HYS declined .
• Thus, the TED spread can be decomposed into the banking sector credit risk premium (LIBOR-OIS) and liquidity or flight-toquality premium (OIS-T-bill).
• Note also the spike after the start of the Subprime Crisis in the spread on ABCP.
• Thus, perceived bank risk, which had previously stemmed mainly from the development of the real economy, now stemmed more from banks’own internal credit and funding risks.

4.6 Sensitivity Analysis

• The consistency of the PC factor space estimates which is established in a series of papers by Bai and Ng constitutes the basis for their empirical analysis.
• To assess the seriousness of these limitations the authors use three additional methods of estimation.
• Makarov and Papanikolaou (2009) recently proposed an extension of specification (1) that ex- plicitly allows for time-varying factor volatility such that:.
• In the second step, the first-step estimates are corrected using the estimated rotation matrix such that the computed factors are also conditionally uncorrelated.
• The dynamics of correlations computed from robust PC estimation also remain unchanged, the only difference being in the level of correlations, which is slightly lower when computed using the robust factor estimates.

5 Conclusions

• The authors have analyzed common factors in bank credit default swaps both before and during the credit crisis that broke out in July 2007 in order to better understand how this crisis spread from the subprime segment of the U.S. financial market to the entire U.S. and global financial system.
• In other words, in this abnormal period investors were not yet concerned so much with the prospect of a global recession that would impact the banks’loan books as with other credit risks affecting the banks —connected, presumably, with their investments in subprime related securities.
• With benefit of hindsight (which is what a retrospective statistical analysis permits), the authors can see a substantial common factor in banks’CDS spreads that could have alerted the authorities to the risks of allowing a major financial institution to fail.
• The further increase in that common factor in the period between the outbreak of the Subprime Crisis and the critical decision concerning Lehman Brothers should have implied further caution in this regard.
• As those variables deteriorated, the result was a perfect storm.

Figures

Table 1. The Sample of Financial Institutions: Descriptive Statistics

Figure 3. Additional Spillovers - (a) Fraction of Pairs of Banks Incurring "Additional" Spillovers

Figure 2: Share of CDS Spreads’Variation Explained by the Four Factors

Figure 7. Association of Common Components with the Costs of Funding

Figure 6. CDS Spread and Costs of Funding (basis points)

Figure 1: Evolution of Spreads on Credit Default Swaps (median, in basis points)

Figure 5. Association of Common Components with the "Real Economy"

Table 2. The Validity of the Correlates

Figure 4. CDS Spread and the "Real Economy"

Full text

City, University of London Institutional Repository
Citation: Eichengreen, B., Mody, A., Nedeljkovic, M. and Sarno, L. (2012). How the
Subprime Crisis went global: Evidence from bank credit default swap spreads. Journal of
International Money and Finance, 31(5), pp. 1299-1318. doi: 10.1016/j.jimonfin.2012.02.002
This is the accepted version of the paper.
This version of the publication may differ from the final published
version.
Permanent repository link: https://openaccess.city.ac.uk/id/eprint/13161/
Link to published version: http://dx.doi.org/10.1016/j.jimonfin.2012.02.002
Copyright: City Research Online aims to make research outputs of City,
University of London available to a wider audience. Copyright and Moral
Rights remain with the author(s) and/or copyright holders. URLs from
City Research Online may be freely distributed and linked to.
Reuse: Copies of full items can be used for personal research or study,
educational, or not-for-profit purposes without prior permission or
charge. Provided that the authors, title and full bibliographic details are
credited, a hyperlink and/or URL is given for the original metadata page
and the content is not changed in any way.
City Research Online: http://openaccess.city.ac.uk/ publications@city.ac.uk
City Research Online

How the Subprime Crisis Went Global:
Evidence from Bank Credit Default Swap Spreads
Barry Eichengreen, Ashoka Mody, Milan Nedeljkovic and Lucio Sarno

First version: August 2009 - This version: February 2012
Abstract
How did the Subprime Crisis, a problem in a small corner of U.S. …nancial markets, a¤ect the
entire global banking system? To shed light on this question we use principal comp onents analysis
to identify common factors in the movement of banks’credit default swap spreads. We …nd that
fortunes of international banks rise and fall together even in normal times along with short-term
global economic prospects. But the importance of common factors rose steadily to exceptional
levels from the outbreak of the Subprime Crisis to past the rescue of Bear Stearns, re‡ecting a
di¤use sense that funding and credit risk was increasing. Following the failure of Lehman Brothers,
the interdependencies brie‡y increased to a new high, before they fell back to the pre-Lehman
elevated levels –but now they more clearly re‡ected heightened funding and counterparty risk.
After Lehman’s failure, the prospect of global recession became imminent, auguring the further
deterioration of banks’loan portfolios. At this point the entire global …nancial system had become
infected.
JEL classi…cation: G10; F30.
Keywords: subprime crisis; credit default swap; common factors.

Eichengreen, Barry. eichengr @econ.berkeley.edu, Department of Economics, University of California, Berkeley,
53 0 E vans Hall, Berkeley, CA 94720-3880, USA and National Bure au for Economic Research (NBER), Cambridg e;
Mody, Ashoka. amody@imf.org, International Monetary Fund, 700 19th S treet, N.W., Washington, D.C., 2043, USA;
Nedeljkovic, Milan. National Bank of S erbia, 12 Kralja Pe tra, Belgrade, 11000, Serbia, milan.nedeljkovic@nbs.rs; and
Sarno, Lucio (corresponding author). Lucio.Sarno@city.ac.uk, Cass Business School, City University, 106 Bunhil l Row,
London, EC1Y 8TZ, UK, +44 (0) 20 7040 8772, and Centre for Economic Policy Research (CEPR), London.
1

1 Introduction
One enduring question about the …nancial turbulence that engulfed the world starting in the
summer of 2007 is how problems in a small corner of U.S. …nancial markets— securities backed by
subprime mortgages accounting for only some 3 per cent of U.S. …nancial assets— could infect the
entire U.S. and global banking systems. Moreover, while the banking system became a¤ected in a
generalized fashion by the crisis, the fortunes of banks di¤ered substantially in terms of the market
assessment (e.g. di¤erentials in the impact on their share prices) and on the scale of government
intervention received. In particular, whether the decision to let Lehman Brothers fail was a critical
mistake that unleashed a global economic and …nancial tsunami will be debated for years. Some
say that the authorities should have known that investors perceived banks’fortunes as intertwined,
so that letting one fail was bound to undermine con…dence in the others. Others say that Lehman
Brothers was unique and everyone knew it.
1
The crisis that a¤ected the global …nancial system, in
this view, did not re‡ect the decision to let this one institution fail. Rather it re‡ected deteriorating
global economic and …nancial conditions that undermined the position of banks as a class.
This paper seeks to shed further light on these issues. We analyze the risk premium on debt owed
by individual banks as measured by banks’credit default swap (CDS) spreads, focusing on the CDS
spreads of the 45 largest …nancial institutions in the U.S., the U.K., Germany, Switzerland, France,
Italy, Netherlands, Spain and Portugal.
2
We use principal components analysis (PCA) to extract the common factors underlying weekly
variations in the CDS spreads of individual banks. If the spreads for di¤erent banks move inde-
pendently, then we can infer that the risk of bank failure is driven by bank-speci…c factors. If they
move together, then we infer that banks are perceived as subject to common risks. This provides us
with the …rst bit of evidence on how the crisis spread. In addition to estimating the importance of
common factors, we attempt to ascertain what they re‡ect. We examine the association between the
1
Amon g o ther things, whereas other institutions could be saved because they had adequate collateral against which
the U.S. Treasury and Federal Reserve could lend, Le hman did not.
2
These swaps are insurance contracts. The buyer of the CDS makes payments to the sel ler in order to receive a
payment if a credit instrument (e.g. a bond or a loan) goes into default or in the e vent of a speci…ed credit event,
such as bankruptcy. The spreads are, in e¤ect, a measure of the credi t risk or the insurance premium charged. This
measure has sever al advantages over the traditional measures which are based on banks’ balance sheet information.
First, the CDS spreads are forward lookin g since they encompass available inf ormation wit h respect to expected default
risk. Balance s heet data only re‡ects e x-post informa tion on the institutions’performance. Second, CDS spreads are
timely updated without t he need to rely on (subjective) i nterpolation te chniques, whereas balance shee t data ar e only
availab le at q uarterly frequency. The CDS spreads a lso o¤er advantages over other market measures of risk based on,
e.g. bond spreads and stock returns. They are the most actively traded derivatives and lead b ond (Blanco, Brennan
and Marsh, 2005) and stock (Acharya and Johnson, 2007) markets i n price discovery. Also, bond spreads may re‡ect
factors other than the ones related to default risk (due to, for example, di¤erent tax treatments) and are sensitive to
the choice of the benchmark risk -free rate (Jorion and Zhang, 2007). However, there has been a recent concern that
specula tive pr essure within the CDS market sometimes causes the swaps to become delinked from their function of
he dging against defau lt (Soros, 2009). See also Longsta¤ et al. (2010 ), who analyze sprea ds on sovereign CDS, and
Zhang, Zhou, and Zhu (2009), who examine th e d eter minants of spreads o n corporate CDS spreads.
2

common factors on the one hand and real-economy in‡uences outside the …nancial system, transac-
tional relationships among banks, and transactional in‡uences between banks and other parts of the
…nancial system on the other hand.
3
We reach the following conclusions. The share of common factors was already quite high, at
62 percent, prior to the outbreak of the Subprime Crisis in July 2007. Banks’ fortunes rose and
fell together to a considerable extent, in other words, even before the crisis. These common factors
were associated with U.S. high-yield spreads— the premium paid relative to Treasury bonds by U.S.
corporations that had less than investment grade credit ratings— which we take as an indicator of the
perceived probability of default by less creditworthy U.S. corporations, and in turn re‡ects economic
growth prospects.
4
For obvious reasons, those defaults and the growth performance that drives them
have major implications for the condition of the banking system even in normal times.
The share of the variance accounted for by common factors then rose to 77 percent in the period
between the July 2007 eruption of the Subprime Crisis and Lehman’s failure in September 2008. This
is indicative of a perception that banks as a class faced higher common risks than before. At the same
time, the measured association between the common factors and U.S. high-yield spreads declined,
while the association with measures of banks’own credit risk and of generalized risk aversion increased
(Brunnermeier, 2009; Dwyer and Tkac, 2009). An interpretation is that the Subprime Crisis made
investors more wary of the risks in bank portfolios for reasons largely independent of the evolution
of the real economy but that lack of detailed information on those risks led them to treat all banks
as riskier rather than discriminating among them.
Following Lehman’s failure, there was a further brief increase in the share of the variance ac-
counted for by the common components. Then, although the level of CDS spreads remained high,
the share of their variance accounted for by the common component fell back relatively quickly to
levels below those that prevailed just before the Lehman episode. In other words, the common
movements declined from their peaks but remained at the post-Bear Stearns elevated levels. Thus,
the perception persisted that the banks’fortunes were linked. The association between the common
factors and high-yield corporate spreads also reemerged, evidently re‡ecting the perception that a
global recession was now in train. More importantly, the common component of CDS spreads became
more highly related with measures of funding and credit risk as measured by spreads in the asset-
backed commercial paper market and LIBOR minus the overnight index swap. An interpretation is
that whereas in the July 2007-September 2008 period investors became more aware of systemic risk
3
To be clear, we do no t attempt to identify causality. However, the association m easures o¤er a r ich set of stylized
characte riza tions. Th ese characteri zations are likel y to be the basis for de…ning and probing more subtle hypotheses.
4
These high-yield s preads have been found to be good predictors of U.S. GDP growth at h orizons of about a year,
re‡ecting a …nancial-accelerator interact ion between credit markets and the real economy (Mody and Taylor, 2003;
Mody, Sarno and Taylor, 2007). Because E uropean high -yield spreads are closely correlated with US spreads and, as
such, o¤er no additional information, U.S. high-yield spreads are a lso a measure of global prospects.
3

in an unfoc used sense, Lehman’s failure caused that common risk to be more concretely identi…ed
with both developments in the real economy and speci…c problems in the …nancial system.
In sum, then, our answer to the question posed in the title is as follows. Banks fortunes rise
and fall together even in normal times. But the importance of common factors rose to exceptional
levels between the outbreak of the Subprime Crisis and the rescue of Bear Stearns, re‡ecting increased
di¤use sense that credit risk was increasing. The period following the failure of Lehman Brothers then
saw a further increase in those interdependencies, re‡ecting heightened funding and counterparty risk.
In addition there were direct spillovers, as opposed to common movements, from the CDS spreads
of U.S. banks to those of European banks. After Lehman’s failure the prospect of global recession
became imminent, auguring the further deterioration of banks’ loan portfolios. At this point the
entire global …nancial system had become infected.
It is helpful to be clear about what this paper does not do. It does not pinpoint any one bank
or set of banks as systemically important. Rather, the extent of comovement in spreads points to
tendencies of the degree to which the system is perceived to be tied to common factors. An individual
bank within the set examined may be more or less tied to the common factors to the extent that it
has a larger or smaller extent of idiosyncratic risk. Ultimately, then, the methodology outlined here
is a guide for policy only to the extent that it highlights overall trends. The task of determining the
systemic importance of an individual bank requires examining the data in the books of the banks— or
worse, data that should be on the books but is not.
The rest of the paper is organized as follows. Section 2 speci…es a dynamic factor model in which
common latent factors explain the movement of the CDS spreads of the 45 banks in our sample. Th e
model is estimated using PCA in recursive fashion, allowing the contributions of the components to
change over time. In Section 3, we consider the possibility of additional spillovers from inter-bank
exposures that go beyond the common movements identi…ed by the latent factors. Then in Section
4, we describe the changing relations between these latent factors and a number of high frequency
…nancial series. We also prov ide a sensitivity analysis to check the robustness of our results. A …nal
section concludes.
2 Common Factors in CDS Spreads
We start by decomposing the change in CDS spreads of N=45 global banks into common and idio-
syncratic components. The term “banks”is used throughout in this paper, although some insurance
companies are also included in the sample. The sample runs from July 29, 2002 to November 28,
2008. Thereafter, the intense involvement of the U.S. authorities in managing the short-term vul-
nerability of the …nancial sector increasingly re‡ects the o¢ cial interventions which, because they
4

Frequently asked questions

Q1. What are the contributions mentioned in the paper "How the subprime crisis went global: evidence from bank credit default swap spreads" ?

In this paper, the authors use principal components analysis to identify common factors in the movement of banks ' credit default swap spreads. 

Q2. What future works have the authors mentioned in the paper "How the subprime crisis went global: evidence from bank credit default swap spreads" ?

The heightened comovement at least in part reflected incomplete knowledge about the magnitude of toxic asset positions in this relatively early stage of the crisis and, hence, raised the possibility that instability could spread more quickly and widely than assumed in the consensus view. 

Q3. What other concerns may have acquired greater prominence in assessing bank risks?

In addition other concerns, such as lack of transparency of the complex asset holdings, may have also acquired greater prominence in assessing bank risks. 

Q4. What do some say about the decision to let Lehman Brothers fail?

Some say that the authorities should have known that investors perceived banks’fortunes as intertwined, so that letting one fail was bound to undermine confidence in the others. 

Q5. What are the banks for which additional spillovers matter?

The banks for which additional spillovers matter tend to be well-known names: they include ING, Royal Bank of Scotland and UBS in Europe, and Bank of America, J.P. Morgan and Morgan Stanley in the U.S.[Insert Figure 3 about here] 

Q6. What is the implication of the spreads of the banks?

Another implication is that for sovereign spreads, the second component and beyond have a more substantial contribution than is the case for banks, implying greater variety of global common influences on sovereign spreads.