the term structure of cds spreads and sovereign credit risk
TRANSCRIPT
The Term Structure of CDS Spreads and Sovereign Credit Risk1
Patrick Augustina∗
a McGill University - Desautels Faculty of Management.
April 3, 2018
2
Abstract3
The shape of the term structure of credit default swap spreads is an informative signal about4
the importance of global and domestic risk factors to the time variation of sovereign credit spreads.5
Exploiting cross-country heterogeneity among 44 countries, I document that the importance of6
global and country-specific risk in explaining sovereign credit risk varies with the sign of the slope7
of the term structure and the duration of its inversion. A model is used to show that global8
uncertainty shocks determine spread changes when the slope is positive, and that domestic shocks9
are more important when the slope is negative.10
Keywords: Credit Default Swaps, Default Risk, Sovereign Debt, Term Structure11
JEL classification: C1, E43, E44, G12, G1512
∗Correspondence to: McGill University, Desautels Faculty of Management, 1001 Sherbrooke Street West,Montreal, Quebec H3A 1G5, Canada; Tel.: +1 514 398 4726; E-mail address: [email protected].
The Term Structure of CDS Spreads and Sovereign Credit Risk 2
1. Introduction1
The defaults of several emerging market economies over the last two decades and the2
multiple recent European government bailouts have fueled the interest in understanding3
the pricing of sovereign credit risk.1 Yet the literature is inconclusive as to the relative4
importance of global and country-specific risk factors to the time variation in sovereign5
credit spreads. Until the end of the 2007-09 financial crisis, there appeared to be some6
consensus that sovereign credit risk is driven by global factors (Pan and Singleton, 2008;7
Borri and Verdelhan, 2016; Ang and Longstaff, 2013) and that it is better explained by U.S.8
financial market factors than by country-specific fundamentals (Longstaff et al., 2011). Since9
the start of the sovereign debt crisis in Europe in 2009, Gennaioli et al. (2012) and Acharya10
et al. (2014), among others, establish a tight link between sovereign risk and the performance11
of the domestic financial sector.12
Such differences in findings are intriguing and beg the question of what could explain13
the divergence in emphasis among these results, and, perhaps, reconcile both parties? One14
possible explanation could be the focus of the above studies on the level of credit spreads.15
For many asset classes, the entire term structure has been shown to convey valuable economic16
information on the pricing of risk, including U.S. government bonds (Cochrane and Piazzesi,17
2005), equity and dividend derivatives (Binsbergen et al., 2012, 2013), currencies (Lustig18
et al., 2017; Zviadadze, 2017), inflation (Fleckenstein et al., 2017), and volatility (Gruber19
et al., 2017), among many others. Thus, incorporating the information from the slope of20
the term structure (henceforth the “slope”) may help us understand why researchers reach21
different conclusions in the sovereign credit risk literature. The examination of the term22
structure of credit default swap (CDS) spreads of different countries across time suggests,23
indeed, that the slope conveys differences in information that cannot be distinguished by24
the level of spreads. Russia, for example, exhibits an identical 5-year CDS spread in three25
1For example, Greece was bailed out three times (2010, 2011, and 2015) and officially defaulted in 2012.Explicit or implicit bailouts were implemented for Ireland (2010), Portugal (2011), Spain (2012), and Cyprus(2013).
The Term Structure of CDS Spreads and Sovereign Credit Risk 3
separate months, despite significant differences in the slope. It is 418 and 270 basis points1
(bps) in January and June 2002, respectively, and -106 bps (inverted) in March 2009, when2
many countries had upward sloping term structures.2 This suggests that dependence on a3
common (level) factor alone is insufficient to explain country-specific heterogeneity in the4
term structure of CDS spreads, and that the shape of the slope may be informative about5
the underlying sources of risk.6
I show that the shape of the term structure of sovereign CDS spreads conveys useful in-7
formation on the importance of global and domestic risk factors for the dynamics of sovereign8
credit risk. In particular, global shocks are the primary source of time variation for spreads9
when the term structure is upward-sloping. A negative slope, in contrast, indicates that10
local shocks dominate. Importantly, for each country, the relative influence of global and11
country-specific risk factors can be inferred in real time, as CDS spreads are observable at12
a daily frequency. These empirical findings are supported by a general equilibrium asset13
pricing model for CDS spreads with recursive preferences and long-run risk. This model can14
be used to explore the time-varying dynamics and cross-country heterogeneity of the term15
structure of sovereign credit risk in relation to common and country-specific shocks.16
The informational power of the term structure of sovereign CDS spreads is documented17
in three ways, using a novel data set on six maturities of sovereign CDS spreads for 4418
countries from January 2001 to February 2012. The level of CDS spreads is defined as the19
5-year spread, and the slope as the difference between the 10-year and the 1-year spreads.20
The focus is on the sign of the slope and the number of months that the slope is negative (the21
duration of inversion). First, using a simple cross-sectional analysis, it is shown that country-22
specific fundamentals explain a significantly greater fraction of the variation in monthly CDS23
spread changes for countries with term structure inversions, compared to countries that have24
never had a negative slope. More importantly, the fraction of variation in spreads due to25
domestic risk increases monotonically with the duration of the inversion.26
2These empirical observations are illustrated graphically in the Online Appendix.
The Term Structure of CDS Spreads and Sovereign Credit Risk 4
Second, the slope is explicitly used as an interaction term in panel regressions to show its1
power in identifying the importance of global and domestic risk factors in driving the time2
variation of sovereign credit risk. Country stock market returns proxy for domestic risk, as3
they are the most significant domestic determinant of CDS spreads in the empirical analy-4
sis. The magnitude of the impact of domestic stock market returns on CDS spread changes5
increases fifteenfold when the slope is negative. It is also greater for larger spread changes,6
and these impacts are statistically significant primarily for spread innovations that are asso-7
ciated with a flattening of the slope. Furthermore, the slope is shown to significantly explain8
quarterly real GDP growth, a country-specific measure, only when the slope is negative.9
An examination of the factor structure of changes in the slope of spreads across countries10
suggests that the slope exhibits little commonality, in contrast to the strong factor structure11
documented for the levels (Pan and Singleton, 2008; Longstaff et al., 2011). One factor12
influences only about 22% of changes in the slope of spreads, while it influences about 57%13
of changes in the level of spreads, i.e., almost three times as much. Overall, these findings14
support the view that the slope contains country-specific information not accounted for by15
the level of CDS spreads, and that this information is useful for understanding cross-country16
heterogeneity related to the dynamics of sovereign credit spreads.17
I next demonstrate that a model with recursive preferences and long-run risk has im-18
plications for time variation in the term structure of CDS spreads that are consistent with19
the observed dynamics of the term structure of spreads. Given the well-documented role of20
time-varying macroeconomic uncertainty (Jurado et al., 2015) and its impact on asset prices21
(Lettau et al., 2008), global macroeconomic uncertainty is the common risk factor in each22
country’s default process, which depends both on global macroeconomic uncertainty and on23
country-specific shocks. Both types of shocks impact the CDS term structure all the time.24
Common shocks work in two counteracting directions, but the dominating effect is to steepen25
the slope. Country-specific shocks work in the opposite direction to the global shock. The26
differential impact on the term structure is due to a differential impact of the shocks on de-27
The Term Structure of CDS Spreads and Sovereign Credit Risk 5
fault probabilities and risk premia. Domestic shocks only affect default probabilities because1
they are unpriced. A negative country-specific shock increases default probabilities more2
for short maturities and less for long maturities, as conditions are expected to improve over3
time. As a result, the term structure inverts. Priced uncertainty shocks also command a risk4
premium, which increases more for longer maturities, due to preference for early resolution of5
uncertainty. Thus, the term structure steepens, as the increase in the term structure of risk6
premia outweighs the decrease in the term structure of default probabilities. The mechanism7
underlying the term structure inversions is consistent with my findings that country-specific8
shocks have greater explanatory power when the slope is negative.9
The model is calibrated to the unconditional moments of all 44 countries in sample.10
Countries, which, on average, have upward-sloping term structures, load heavily on aggregate11
risk. For countries that, on average, have downward-sloping term structures, the leverage12
factor on global risk is small, and the default intensity depends more on idiosyncratic shocks.13
Simulations suggest that the model describes the data well, as the 5-year spread level always14
lies within the 5th and the 95th percentiles of the small sample distribution. The model15
qualitatively fits the slope patterns and, importantly, the frequency of the term structure16
inversion. The simulated panel also closely matches the factor structure of high commonality17
in the levels, despite little commonality in the slopes. Model-implied risk premia rise with18
maturity, and represent a smaller proportion of the level of the spreads when the term19
structure is inverted.20
The model has two testable predictions, which bear out in the data. First, the ratio of21
risk premia relative to expected default losses is lower for countries when the term structure22
is inverted. Second, expected excess returns are greater when global economic uncertainty23
is high. The confirmation of both predictions validates the model and supports the findings24
that the shape of the CDS term structure conveys useful information on the importance of25
global and domestic risk for the dynamics of sovereign credit risk.26
My findings provide support for several conclusions in the literature, despite an apparent27
The Term Structure of CDS Spreads and Sovereign Credit Risk 6
disagreement about the importance of global and country-specific risk. As the influence of1
global and domestic risks on sovereign credit risk is varying over time, the results may be2
linked to the evidence on a tight relation between sovereign default risk and the domestic3
financial sector observed during times of distress (Gennaioli et al., 2012; Acharya et al.,4
2014). My findings may also be linked to the evidence on the relation between sovereign CDS5
spreads and U.S. financial factors during benign times (Pan and Singleton, 2008; Longstaff6
et al., 2011; Ang and Longstaff, 2013). Remolona et al. (2008) provide evidence that global7
risk aversion is unconditionally the dominant determinant of sovereign risk premia, while8
country-specific fundamentals and market liquidity matter more for expected credit losses.9
In contrast to these papers, I examine the interactions between global and country-specific10
risk factors and study their implications for the dynamics of the CDS term structure.311
Duffie et al. (2003) and Zhang (2008) use an exogenous default intensity to study the12
credit risk of Russia and Argentina. The presented framework embeds such a reduced-form13
default intensity into a recursive utility framework, similar to Augustin and Tedongap (2016)14
and Chernov et al. (2017), although the latter endogenize sovereign default to examine the15
size of U.S. CDS premia. While these two papers focus on aggregate moments of the CDS16
term structure, this paper focuses on the dynamics of the CDS term structure. Hence, cross-17
sectional differences in the term structure of CDS spreads may be related to the importance18
of global and local risk. Contagion is not modeled. This distinguishes the paper from Benzoni19
et al. (2015), who use the fragile beliefs framework to study how a hidden factor generates20
contagion in the cross-section of sovereign default probabilities.21
Importantly, among the above papers, almost none use the information embedded in22
the CDS term structure. Notable exceptions are Pan and Singleton (2008) and Longstaff23
et al. (2011), who use the cross-sectional information in the term structure to estimate the24
risk-neutral parameters of the default process. Arellano and Ramanarayanan (2012), on the25
3Augustin (2014) surveys the role of global and local risk factors for sovereign CDS spreads. Other relatedreferences are Mauro et al. (2002), Geyer et al. (2004), Uribe and Yue (2006), Reinhart and Rogoff (2008),Obstfeld and Rogoff (2009), Hilscher and Nosbusch (2010), Dieckmann and Plank (2011), Jeanneret (2015),and Kallestrup et al. (2016).
The Term Structure of CDS Spreads and Sovereign Credit Risk 7
other hand, illustrate how the endogenous choice of debt maturity may lead to an inverted1
sovereign yield curve. Jotikasthira et al. (2015) show that world inflation and the U.S. yield2
curve are a source of co-movement of the yield curves of the U.S., the U.K., and Germany.3
My focus is on the empirical findings that the slope of the CDS term structure is a useful4
real-time market indicator of the country-specific sources of sovereign credit risk, which is5
explained using a general equilibrium asset pricing model.6
The rest of this paper proceeds as follows. In Section 2., the data is described and stylized7
facts about the term structure of sovereign CDS spreads are presented. The development of8
the model and a discussion of the asset pricing implications follows in Section 3. The model9
is validated in Section 4. The conclusion is presented in Section 5.10
2. The Term Structure of Sovereign CDS Spreads11
In this section, I present stylized facts about the term structure of sovereign CDS spreads.12
After an overview of the data, I provide cross-sectional and time series evidence on how the13
slope relates to global and domestic risks, and country-specific economic growth. The section14
ends with a statistical analysis of the factor structure of the level and the term structure of15
CDS spreads.16
2.1. Data17
The study of the term structure of sovereign CDS spreads is based on daily spreads from18
Markit for maturities of 1, 3, 5, 7, and 10 years. All swaps are denominated in USD, apply19
to senior foreign debt, and contain the full restructuring credit event clause. Monthly CDS20
spreads are derived from the last available observation in each month. The slope of the CDS21
term structure is defined as the difference between the 10- and 1-year spreads. With 4422
countries from Europe/Eastern Europe, Asia, Latin America, and the Middle East/Africa,23
the panel spans a broad geographical area and exhibits significant time series and cross-24
sectional heterogeneity in both the level and the slope.25
The Term Structure of CDS Spreads and Sovereign Credit Risk 8
Table 1 presents summary statistics. The earliest starting period is January 2001, and1
all observations end in February 2012. For any country, the market typically charges higher2
spreads over longer horizons to compensate for greater uncertainty about future default risk.3
Plotted against progressively longer maturities, this produces a smooth, upward-sloping4
curve. In contrast to the findings in Pan and Singleton (2008), I find that the average5
term structure in column (7) is not always upward-sloping; it is negative for four countries:6
Greece, Ireland, Portugal, and Uruguay. The mean slope is as negative as -382 bps in the7
case of Greece, and Colombia has the largest positive slope at 234 bps. The average 5-year8
spread, reported in column (5), ranges from 13 bps for Finland to 868 bps for Venezuela.9
Columns (8) and (9) display the number and frequency of months during which the term10
structure was inverted. Twenty countries had a negative slope for at least one month, while11
six countries had an inverted term structure for more than twelve months, with the frequency12
of inversion ranging up to 20% of the sample period. There are 198 (monthly) inversions,13
staggered across countries and across time.4 The term structure of individual countries14
inverts (and consequently reverts) up to four times during the sample period. The fact that15
inversions (and reversals) occur for different countries at different points of time creates rich16
panel dynamics that are key to identifying the stylized facts highlighted in the following17
subsections.18
2.2. Cross-sectional explanatory power of domestic and global risk19
To understand the relative role of domestic and global risk factors, I first examine whether
the sign of the slope and the duration of inversion can be related to cross-sectional differences
in the explanatory power of common risk factors and country-specific fundamentals. To this
end, for each country i, the changes in the monthly 5-year sovereign CDS spreads (∆CDSit)
are projected on the changes of three local factors and changes of three groups of global
4Due to space limitations, additional summary statistics are in the Online Appendix. The Appendix alsoprovides examples of countries with low CDS spread levels and an inverted CDS term structure that arecompared to countries which contemporaneously have high CDS spread levels and a positive term structureof spreads.
The Term Structure of CDS Spreads and Sovereign Credit Risk 9
factors. The regression output is used to construct “local ratios” that proxy for the amount
of the explanatory power of the country-specific risks. Specifically, the following regression
model is estimated:
∆CDSit = αi + β>i ∆Lit + γ>i ∆Gt + εit, εit ∼ N (0, 1) , (1)
where Lit denotes a vector of domestic factors and Gt refers to the vector of global factors.1
The three local factors are the domestic stock market return in local currency, the ex-2
change rate relative to the USD, and the USD-denominated foreign currency reserves.5 The3
first set of global factors are financial market indicators from the U.S.: the U.S. excess stock4
market return, changes in the 5-year constant maturity Treasury yield, and changes in the5
spreads of U.S. investment-grade and high-yield bond indices. The second set of global fac-6
tors comprises proxies for international risk premia, based on the intuition that risk premia7
should correlate across asset classes. The equity risk premium is proxied by changes in the8
earnings-price ratio of the S&P 500 Index. Changes in the spread between the implied and9
realized volatility of index options are used for the volatility premium, and changes in the10
expected excess returns on five-year Treasury bonds approximate the term premium. Val-11
uation effects based on international capital flows are captured using net global flows into12
equity and bond mutual funds. Finally, to account for any residual economic sources of risk,13
the predictor variables also include a regional and a global sovereign spread. The regional14
spread is computed as the mean spread of all other countries in the same region, whereas the15
global spread is measured as the mean spread of the countries in all other regions, except16
the region being analyzed. Only the residual part orthogonal to all other regressors is used17
in the regressions. Data sources and details about variable constructions are provided in the18
Online Appendix.19
The local ratio (LR) statistic captures cross-sectional differences in the explanatory power20
5While exchange rates may contain a systematic component (Verdelhan, 2018), they are classified as localfollowing Longstaff et al. (2011). All conclusions remain unchanged if the local variation in spread changesis associated with the component unexplained by all global risk factors.
The Term Structure of CDS Spreads and Sovereign Credit Risk 10
of local risk. The LR is computed as the ratio of the adjusted R2 from a restricted regression,1
which includes domestic variables only, to that of an unrestricted regression, which includes2
all variables. The metric is thus increasing in the importance of country-specific factors.3
Longstaff et al. (2011) show, using a sample of 26 countries, that spread changes are related4
primarily to global determinants, in particular U.S. equity returns, volatility, and bond5
market risk premia. I document cross-sectional differences in the explanatory power of local6
and global risk factors, which I relate to the sign of the slope and the duration of inversion.7
All 44 countries are separated into two groups: those that never had an inverted slope8
(G1) and those that had an inverted slope during at least one month (G2). Table 2 reports9
the mean and median LR for both groups, which are highly statistically different.6 The10
median (mean) LRs for the two groups are 38% (40%) and 63% (61%). The set of individual11
country LRs contains three outliers. In G1, the LR for Venezuela is roughly 5%. This12
country arguably plays a special role, given its importance for global oil production. In G2,13
Mexico and Egypt, two countries with a high LR, never have an inverted term structure14
yet experienced significant financial or political trouble. Mexico underwent an economic15
downturn in connection with the drug cartel wars, while Egypt suffered as a result of the16
Arab Spring. Their LRs are 77% and 65%, respectively. Excluding the outliers, the difference17
in the mean and median LRs between the two groups widens further, both the mean and18
median LRs being 36% for G1 and 64% for G2. The LRs for Portugal, Greece, and Ireland19
are 109%, 76% and 100%. Overall, these findings suggest that country-specific risk factors20
explain more variation in the CDS spreads of countries with inverted term structures.21
Figure 1 shows that the LR is monotonically increasing in the months when the term22
6Inference is obtained by block-bootstrapping 10,000 times a sample size of 36 months for each country.A one-sided t-test on the equality of means, assuming paired data, against the alternative that G1 has asmaller mean, is rejected at the 1% significance level. The Wilcoxon matched-pairs signed-rank test rejectsthe null hypothesis that both distributions are the same, while the one-sided sign test rejects the equality ofthe medians against the alternative that the median of G1 is lower. Note that in this inference, I assume thatthe correlation between local and global factors is the same for each country. Adjusting for this correlationwould lower the LR. If the local economies of distressed countries are more highly correlated with globalfactors than are those with non-distressed countries (the most unlikely case), the results would be evenstronger.
The Term Structure of CDS Spreads and Sovereign Credit Risk 11
structure is inverted. The relation between the duration of inversion and how much variation1
the domestic factors explain is depicted in the upper panel of Figure 1 for countries in G22
(excluding Venezuela). A linear regression fitted to the scatter plot yields an R2 statistic of3
32%, with a statistically significant t-statistic of 3.11. The regression fit further improves if4
countries from both G1 and G2 are considered, as is illustrated in the lower graph in the5
figure. The R2 increases to 45% and the t-statistic is 5.76. These results are not driven6
by the European sovereign debt crisis, nor by a clustering of events over time. Excluding7
Europe from the sample yields an R2 of 32% with a t-statistic of 3.16. The explanatory8
power decreases slightly to 23% using the full sample that includes Venezuela, and the9
relation remains significant with a t-statistic of 3.12. These findings support the view that10
the explanatory power of country-specific risk is monotonically increasing in the duration11
of term structure inversion.7 The results in Figure 1 suggest that the LR, imputed from12
CDS spread levels, yields similar cross-country rankings of the importance of local risk to13
those obtained from the term structure of spreads. The estimation of LRs requires a long14
time series for an important number of observables. The slope is model-independent and15
observable in real time.16
2.3. The role of the slope for the dynamics of sovereign credit risk17
I next use panel regressions and augment the empirical model from equation (1) with an18
indicator variable that is equal to one if the term structure is negative, and zero otherwise.19
This indicator variable is interacted with both the local and global risk factors, but the20
focus is on the domestic and U.S. stock market returns, as these are the most significant21
determinants in country and panel regressions. In other words, the information in the slope22
is used to distinguish the importance of global and local risk factors as a function of the23
sign of the slope. For the sake of brevity, Panel A of Table 3 only reports the coefficients of24
7Detailed results of individual country regressions are available upon request. The most significant globalrisk factors are the U.S. stock returns, the U.S. equity premium, and global and regional spreads, consistentwith Longstaff et al. (2011). The most significant local risk factor is the domestic stock market return.
The Term Structure of CDS Spreads and Sovereign Credit Risk 12
interest.1
All specifications indicate that an increase in both the domestic and global stock market2
return reduces sovereign CDS spreads, as demonstrated by the statistically significant nega-3
tive coefficient on both risk factors. The unconditional effect of the U.S. equity return is, on4
average, three times as large as that of the country-specific stock market return. The coeffi-5
cient on the interaction term between the negative slope indicator and the local stock market6
return is negative. It has a magnitude of 0.15, suggesting that the impact of domestic stock7
market returns on sovereign CDS spreads is fifteen times larger when the term structure8
is inverted, and five times larger than the unconditional impact of the U.S. stock market9
performance. The coefficient on the negative slope indicator multiplied by the U.S. equity10
return suggests, likewise, a stronger dependence in times of term structure inversion. The11
coefficient is, however, not robust, and mostly statistically insignificant. The specifications12
presented in columns (2) to (8) of Table 3 successively feature controls, country fixed effects,13
and different standard error corrections.8 In column (8), the full set of interaction effects14
between all risk factors and the negative slope indicator is included. Importantly, none of15
the specifications change the statistical significance or economic magnitude of the coefficient16
for the interaction term between the domestic stock market return and the negative slope17
indicator.18
The sample is next split conditional on positive and negative changes in the CDS spread19
slopes. While the relation between changes in CDS spreads and domestic stock market20
returns is consistently negative in Panel B of Table 3, it is asymmetric. The dependence is21
about two to three times greater for changes that flatten the slope (column (2)), as opposed22
to changes that steepen it (column (1)). Importantly, the correlation between domestic23
equity returns and sovereign credit risk increases in months when the slope is negative, in24
8Regression specifications are reported with and without country fixed effects, clustering by country,month, time and month, and with Driscoll-Kraay standard errors with country fixed effects, accounting forcross-sectional correlation and time series correlation of up to three lags. Results are similar for specificationsthat use the component of local shocks that is orthogonal to global shocks, as reported in the OnlineAppendix.
The Term Structure of CDS Spreads and Sovereign Credit Risk 13
an economically significant way, only when the slope is flattening, as demonstrated by the1
significant regression coefficient in column (2) of Panel B. In contrast, the impact of U.S.2
equity returns on spread changes is of equal economic magnitude for shocks that flatten the3
slope, than for shocks that steepen the slope. In both cases, the negative coefficient has a4
magnitude of 0.027. In addition, the regression that conditions on a flattening of the slope5
has an R2 of 24%, compared to a weaker fit of 18% for the regression that conditions on a6
steepening of the slope. The results in column (3) corroborate the finding that the relation7
between domestic stock market returns and changes in CDS spreads significantly increase8
when the slope is flattening, while this is not the case for the relation with U.S. equity9
returns. For the specification in column (3), in which the full panel and dynamics of the10
slope are exploited, the explanatory power increases to 39%.11
I provide even more granular results in Panel C of Table 3, which reports the impact of12
local stock market performance on CDS spread changes at different quantiles of the distri-13
bution of 5-year CDS spread changes. Given that the interaction term with the U.S. equity14
return is insignificant, only the coefficient of the domestic stock market return and its inter-15
action with the negative slope indicator are reported. The magnitude of the coefficient of the16
local stock market return increases monotonically from 0.003 (in absolute value) for the 50th17
to 0.009 for the 95th percentile of the distribution of CDS spread changes, corresponding18
to a multiplicative factor of three. Similarly, the magnitude of the beta coefficient on the19
interaction term increases monotonically from a magnitude of 0.04 to 0.27, but the increase20
is more pronounced, given that the impact at the 95th percentile is more than six times as21
large as that at the 50th percentile. These findings jointly underscore that the contempora-22
neous impact of domestic stock market returns on sovereign CDS spread changes is greater23
when the slope is negative, for shocks that flatten the slope, and for larger spread changes.24
The Term Structure of CDS Spreads and Sovereign Credit Risk 14
2.4. The slope of CDS spreads and economic growth1
In this section, I examine the relation between the term structure of sovereign CDS2
spreads and macroeconomic fundamentals. If a negative slope truly signals the impact of3
domestic risks, then we should observe a significant relationship between the slope and4
country-level economic growth (country-specific measure) only when the slope is negative.5
Quarter-on-quarter seasonally adjusted GDP growth data is sourced for 42 countries from6
the OECD, Oxford Economics, IHS Global Insight, Thomson Reuters Datastream, IMF7
International Financial Statistics, and the national statistical offices or central banks of8
Uruguay, Peru, Egypt, South Africa, Slovenia, and Cyprus.9 Merging the data results in an9
unbalanced panel of 42 countries from 2001:Q1 to 2012:Q1 with a total of 4,951 observations.10
First, all 4,951 observations are ranked in ascending order based on the size of the slope,11
and grouped into sliding windows of 100 observations. Within each window, the fraction of12
observations with negative real GDP growth relative to the previous quarter is computed,13
as well as the average slope of the term structure of CDS spreads.10 The resulting relation14
between real GDP growth and the slope of CDS spreads, plotted in Panel A of Figure 2,15
illustrates a visible discontinuity at the threshold of a flat slope. Once the slope shifts from16
being positive to negative, there is a sharp drop in the average level of growth, as well as17
a sharp increase in the number of countries with negative growth. Panel B in Figure 218
illustrates the same relation between real GDP growth and the magnitude of the slope for19
values above -250 bps.20
One valid concern is that the slope is highly correlated with the level of CDS spreads,21
which, would capture the relation between the dynamics of the CDS spreads and the ex-22
planatory power of domestic risk. Panels C and D of Figure 2 depict the proportion of23
countries with an inverted term structure and the average slope, respectively, against the24
9No reliable information was found for Panama and Russia. The information for Lebanon is availableonly at a yearly frequency.
10If the ranked CDS spreads are indexed with integer numbers starting from 1, then the first windowprovides the average slope of the term structure and real GDP growth for observations 1 to 100; the secondfor observations 2 to 101; the third for observations 3 to 102, and so forth.
The Term Structure of CDS Spreads and Sovereign Credit Risk 15
average level of spreads. The figures show, however, that such concerns are unfounded. Even1
though the level and the slope are positively correlated, almost 40% of the countries do not2
have a negative slope at spread levels as high as 1,000 bps. This confirms that the slope3
captures information different from that captured by the level of CDS spreads.4
The “slope-growth” relation is formally validated using panel regressions, whereby the5
seasonally adjusted quarter-on-quarter real GDP growth is projected on the level and the6
slope of CDS spreads, as well as the interaction between the slope and a negative slope7
indicator, which is equal to one if the slope is negative in a quarter, and zero otherwise. The8
results in Table 4 suggest that the slope is contemporaneously predictive of the quarterly real9
GDP growth only when it is negative. The level of spreads is negatively predictive of growth,10
and growth is, on average, 2.07 percentage points lower when the slope is negative, based on11
the most conservative results in column (5). A steeper slope, on the other hand, is positively12
predictive of growth, but the unconditional effect is not statistically significant. The slope13
is a significant predictor of economic growth only in those quarters when the term structure14
is inverted. Adding the interaction term to the regression increases the explanatory power15
of GDP growth from 2.49% to 3.01%, in terms of adjusted R2. Excluding Greece from the16
sample (column (6)) does not affect the results. Similar regression results at the monthly17
frequency are provided in the Online Appendix; the interaction between the slope and the18
negative slope indicator is consistently significant at either the 5% or 1% significance level,19
depending on the specification.20
2.5. Principal component analysis21
I next evaluate the statistical content of the slope by examining its factor structure. Table22
5 reports the results of a principal component analysis (PCA) on the covariance matrix of23
changes in both CDS spread levels and slopes. The results of the PCA are reported for the24
whole sample period, the pre-crisis period of 2003-2006, the financial crisis of 2007-2009, and25
The Term Structure of CDS Spreads and Sovereign Credit Risk 16
the sovereign debt crisis of 2010-2012.11 Over the whole sample period, the first principal1
component provides explanatory power, on average, for 57% of the variation in monthly2
5-year CDS spread changes (Panel A). In line with the literature, this number varies from3
43% in the pre-crisis period (with sovereign defaults) to 75% during the financial meltdown4
(with no sovereign defaults).12 This suggests that the drop in correlation is driven by the5
countries in distress. Panel B shows that similar results are found for a PCA performed on6
pooled spreads of all maturities, as in Pan and Singleton (2008).137
The results in Panels A and B contrast sharply with those in Panel C, which show8
that the first principal component has relatively little explanatory power for changes in9
the slope. On average, the first factor provides explanatory power for only about 22%10
of the variation. Conditional values change from 24% to 38%, and back to 31% over the11
three time periods. The first three common factors combined provide explanatory power12
for only about 39% of the slope variation. Thus, the high degree of commonality in spread13
levels contrasts with the low degree of commonality in the slope. This reflects a significant14
degree of heterogeneity in the behavior of the slope in the cross-section and over time. This15
heterogeneity is symptomatic of the slope’s ability to capture cross-sectional differences in16
the importance of domestic and global risk over time.17
Even though liquidity is more balanced across contract maturities for sovereign than for18
corporate CDS contracts (Pan and Singleton, 2008), idiosyncratic liquidity shocks, which19
affect separate maturities differently, could explain a lower co-movement of the term struc-20
tures. Such shocks should, however, also reduce the co-movement in the level of spreads.21
Thus, idiosyncratic liquidity shocks are unlikely to affect the difference in explanatory power22
provided by the first principal components from a PCA on the level and slope of CDS spreads.23
11To perform the PCA, balanced panel dataset is needed, which is why the starting year of the analysis is2003. The analysis is conducted for standardized spread changes.
12During the sovereign debt crisis, one might expect spreads to co-move even more. However, the fractionexplained by the first factor falls back to the long-run average of 58%. Such a result is not found if thesample is limited to the 26 countries used in Longstaff et al. (2011), who examine no country in distress,apart from Hungary and Venezuela. These results are available upon request.
13The requirement of a balanced panel for the full term structure reduces the sample to 30 countries.
The Term Structure of CDS Spreads and Sovereign Credit Risk 17
Note that 42 out of 44 countries in the sample rank among the 1,000 most liquid CDS refer-1
ence entities, as reported by the Depository Trust & Clearing Corporation (Augustin, 2014).2
This should somewhat mitigate concerns about liquidity in these contracts.3
The difference in the factor structures implied by the PCAs of CDS spread levels and4
slopes is a fourth stylized fact about the slope of the term structure of sovereign CDS5
spreads. In addition, it has been documented that in the cross-section, the explanatory6
power of country-specific risk is increasing in the duration of term structure inversion; the7
contemporaneous impact of domestic stock market returns on spread changes is greater in8
times when the slope is negative, for larger spread changes, and for changes that attenuate9
the slope; the slope contemporaneously predicts quarterly real GDP growth only when the10
slope is negative.11
3. A Preference-based Model for Credit Default Swaps12
I now turn to rationalize my findings using one of the workhorse equilibrium asset-pricing13
models. The challenge is to link the dynamics in the term structure to the dynamics of both14
global and country-specific risk factors, such that the slope is an informative signal about15
the relative importance of each source of risk. To this end, a reduced-form default process is16
embedded into a consumption-based asset pricing framework with recursive preferences and17
a long-run risk economy. The results rely on three key assumptions: persistent shocks to18
the volatility of aggregate consumption growth (Jurado et al., 2015), a preference for early19
resolution of uncertainty (a standard assumption in the literature), and integrated capital20
markets.21
3.1. Pricing credit default swaps22
The proposed CDS pricing framework is conceptually standard. The difference from
reduced-form models is that default dynamics are impacted by economic fundamentals,
which, together with investor preferences, define the pricing kernel used to discount ex-
The Term Structure of CDS Spreads and Sovereign Credit Risk 18
pected cash flows. The general equilibrium valuation of CDS spreads builds on Augustin
and Tedongap (2016). Without loss of generality, the trading frequency is monthly. Each
coupon period contains J trading months, thus a K-period swap has a time to maturity of
KJ months.14 The buyer’s leg πpbt is defined as:
πpbt = CDSt (K)
(K∑k=1
Et [Mt,t+kJI (τ > t+ kJ)] + Et
[Mt,τ
(τ − tJ−⌊τ − tJ
⌋)I (τ ≤ t+KJ)
]),
(2)
where CDSt (K) is the constant time t (annualized) premium of a K-period CDS, quoted1
as a percentage of the insured face value, and expected to be paid until the earlier of either2
maturity (month t+KJ) or a credit event occurring at a random month τ . Mt,t+j denotes3
the stochastic discount factor that values in month t any financial payoff to be claimed at4
a future month t + j. Note that b·c rounds a real number to the nearest lower integer, and5
I (·) is an indicator function taking the value 1 if the condition is met, and 0 otherwise. The6
protection leg in equation (2) is the sum of two parts. The first relates to payments made7
by the protection buyer if there is no credit event. The second defines accrual payments for8
default occurring between two payment dates.9
The seller’s leg πpst is defined as the net present value (NPV) of expected losses incurred
by the buyer when there is a credit event. πpst is described by:
πpst = Et [Mt,τ (1−R) I (τ ≤ t+KJ)] , (3)
where R represents the constant post-default recovery rate. A fairly priced CDS at date t is10
thus obtained by equating at inception the NPVs of cash flows for the buyer and the seller,11
which yields the K-period CDS spread12
14For example, an annual (bi-annual, quarterly) payment frequency for a 5-year swap implies that J =12 (6, 3) and K = 5 (10, 20) with 60 months to maturity.
The Term Structure of CDS Spreads and Sovereign Credit Risk 19
CDSt (K) =
KJ∑j=1
Et [Mt,t+j (1−R) (St+j−1 − St+j)]
K∑k=1
Et [Mt,t+kJSt+kJ ] +KJ∑j=1
(jJ− b j
Jc)Et [Mt,t+j (St+j−1 − St+j)]
, (4)
where the process St ≡ Prob (τ > t | It) denotes the survival probability conditional on the
time-t information set It. Survival probabilities depend on an hazard rate ht:
St = S0
t∏j=1
(1− hj) for t ≥ 1, (5)
which defines the instantaneous probability of default conditional on no earlier default, i.e.,1
ht ≡ Prob (τ = t | τ ≥ t; It). Equation (4) contains two main ingredients: the pricing kernel2
Mt,t+j and the survival probabilities St. These are defined in terms of investor preferences3
and the dynamics of the aggregate economy.4
Counterparty risk is not modeled, as Arora et al. (2012) and Du et al. (2017) show5
that while counterparty risk is statistically priced, its economic significance is negligible.6
The framework also does not account for the introduction of the Big Bang and Small Bang7
protocols in the U.S. and Europe, respectively, as the standardization of coupon payments8
with up-front settlements has little quantitative pricing implications (Collin-Dufresne et al.,9
2012). Finally, accounting for differential liquidity across different contract maturities would10
most likely only affect the convexity of the slope, at the expense of making the model11
substantially more cumbersome.12
3.2. Preferences13
I assume the existence of U.S.-based representative investor selling USD-denominated
insurance contracts. This view is consistent with the strong concentration of CDS trading
The Term Structure of CDS Spreads and Sovereign Credit Risk 20
in the OTC dealer market for CDS.15 The utility-maximizing marginal investor has Epstein
and Zin (1989)-Weil (1989) recursive preferences. This implies that the logarithm of the
stochastic discount factor can be written as a function of consumption growth ∆ct+1 =
ln (Ct+1/Ct), the natural logarithm of the gross return on a claim to aggregate wealth rc,t+1,
and preferences:
mt,t+1 = θ ln δ − θ
ψ∆ct+1 − (1− θ) rc,t+1, (6)
where θ = 1−γ1− 1
ψ
, with both the coefficient of relative risk aversion γ ≥ 0 and the elasticity1
of intertemporal substitution (EIS) ψ ≥ 0 being non-negative, and where δ defines the time2
preference. I assume that the agent prefers early resolution of uncertainty, i.e., γ > 1ψ
, which3
is necessary to obtain an upward-sloping term structure of CDS spreads.4
3.3. Economy5
As in Bansal and Yaron (2004), and supported by Hansen et al. (2008) and Bansal et al.
(2012), aggregate consumption growth ∆ct+1 embeds a slowly mean-reverting predictable
component xt, which determines the conditional expectation of consumption growth,
∆ct+1 = xt + σtεc,t+1
xt+1 = µx + φx (xt − µx) + νxσtεx,t+1,
(7)
where the short- and long-run consumption shocks εc,t+1 and εx,t+1 are independent and6
identically distributed normal errors with zero mean and unit variance. The parameter φx7
modulates the persistence of expected growth, whose long-run mean is defined by µx, and8
whose sensitivity to long-run shocks is guided by νx. While persistent shocks to expected9
growth help matching the moments of other asset classes (risk-free bonds, equity, etc.), they10
are not necessary for generating any of the implications for the CDS term structure, which11
depends crucially on the persistence of shocks to the volatility of aggregate consumption12
15See Giglio (2017) and Augustin (2014), among others. Sovereign CDS are priced in USD. Thus, resultswould carry through for a foreign investor as long as markets are complete and all shocks are spanned byexchange rates (Borri and Verdelhan, 2016).
The Term Structure of CDS Spreads and Sovereign Credit Risk 21
growth.1
Consumption growth and its conditional mean inherit the same stochastic volatility pro-
cess σ2t , modeled as an autoregressive gamma process (Gourieroux and Jasiak, 2006; Le et al.,
2010),
σ2t+1 ∼ ARG
(νσ, φσσ
2t , cσ
), (8)
where the parameter φσ modulates the persistence of volatility. The parameters νσ > 0 and
cσ > 0 define the shape and scale of the distribution, and are related to the unconditional
mean µσ and variance ωσ of the volatility process by µσ = (νσcσ) / (1− φσ) and ωσ =
(νσc2σ) / (1− φσ)2. Thus, consumption growth has conditionally time-varying first and second
moments, given by:
Et[σ2t+1
]= φσσ
2t + νσcσ and Vt
[σ2t+1
]= 2cσφσσ
2t + νσc
2σ, (9)
consistent with empirical evidence of conditional volatility in consumption growth (Kandel2
and Stambaugh, 1990; Stock and Watson, 2002), and the co-movement between macroe-3
conomic volatility and asset prices (Bansal et al., 2005; Lettau et al., 2008). With this4
specification, the model remains nested in the class of general affine equilibrium models de-5
scribed by Eraker (2008) and it ensures that consumption volatility is a positive process.6
More importantly, it allows me to easily introduce aggregate macroeconomic uncertainty as7
a proxy for global risk into the underlying default process to generate dynamics in the term8
structure of CDS spreads.9
The primary modeling contribution of this paper is the specification of the dynamics for
sovereign default risk. In Equation (5), I describe how survival probabilities depend on the
hazard rate ht, which is driven by the country-specific default intensity λit+1 through the
following relation:
hit+1 = 1− exp(−λit+1
), (10)
The Term Structure of CDS Spreads and Sovereign Credit Risk 22
where the superscript i will subsequently be dropped for ease of exposition. To ensure
that the hazard rate is bounded between zero and one, the default process must be positive.
Furthermore, it should inherit both global and country-specific shocks. To attain these goals,
the default process λt+1 is modeled as a bivariate autoregressive gamma process:
λt+1 ∼ ARG(νλ, φλσσ
2t + φλλt, cλ
), (11)
where the parameters φλσ > 0 and 0 < φλ < 1 modulate the sensitivity of the default
process to the global factor and its own past, respectively, while the νλ > 0 and cλ > 0
define the shape and scale parameters of the distribution. Thus, macroeconomic uncertainty
σ2t may feed directly into expectations about future defaults. The default process features
time-varying conditional first and second moments, given by:
Et [λt+1] = φλσσ2t + φλλt + νλcλ and Vt [λt+1] = 2cλ
(φλσσ
2t + φλλt
)+ νλc
2λ, (12)
where time-varying second moments generate time-varying correlations in CDS spreads
across countries. These expressions reflect that the conditional first and second moments are
high when macroeconomic uncertainty σ2t is high and the exposure φλσ is large, or when the
default process is very persistent, that is φλ is close to one.16 It is useful to illustrate the
autoregressive form of the default intensity:
λt+1 = νλcλ + φλσσ2t + φλλt + ηλ,t+1, (14)
which highlights the unpriced country-specific innovations ηλ,t+1 (zero mean shocks). The1
mean-reverting common factor σ2t introduces co-movement into spreads across countries.2
16This is also well illustrated based on the unconditional first and second moments of the default process,defined as:
E [λ] =φλσµσ + νλcλ
1− φλand V [λ] =
2cλ (φλσµσ + φλµλ) + φ2λσωσ + νλc2λ
1− φ2λ. (13)
The Term Structure of CDS Spreads and Sovereign Credit Risk 23
Each country’s default process is driven by shocks to macroeconomic uncertainty through1
its dependence on σ2t . However, it also inherits idiosyncratic shocks ηλ,t+1, which are uncor-2
related across countries. Sensitivity to global shocks is modulated through φλσ. If φλσ is3
zero, the default process becomes purely idiosyncratic.174
The process driving default risk differs here from the literature in four respects. First,5
the default intensity is specified under the objective probability measure and describes the6
actual default process underlying sovereign default risk. This contrasts with the common7
reduced-form pricing framework of Duffie (1999), in which default intensities are specified8
under the risk-neutral probability measure. Second, the default process describes the risk9
attributed to unpredictable variation in the probabilities of triggering CDS credit events,10
which does not include default or bankruptcy.18 The default process therefore appeals to11
distress risk, which influences the market’s perception of sovereign default risk. This is12
relevant from an investment perspective for marking-to-market portfolios of government debt13
and CDS positions. It does not capture jump-at-default risk, in line with Pan and Singleton14
(2008) and Longstaff et al. (2011). Third, the functional form of the default process, i.e.,15
autoregressive gamma, converges in the limit to a square-root process, which is used to16
study the default risk of Argentina in Zhang (2008) or corporate default risk in Longstaff17
et al. (2005). Berndt et al. (2008) and Longstaff et al. (2011) settle on a lognormal process.18
Fourth, while the previous papers use a one-factor model, which can be sufficient to capture19
the strong commonality in spreads, I specify a two-factor process, which incorporates both20
global and idiosyncratic shocks. Two factors are necessary to capture differences in the shape21
of the slope across countries and over time.22
17Default risk does not feed back into the pricing kernel, nor into the consumption process of the marginalinvestor, consistent with recent multi-country no-arbitrage approaches as in Jotikasthira et al. (2015). Suchalternative specifications would be interesting avenues for future research.
18Standard ISDA documentation for sovereign CDS contracts list four different credit events: obligationacceleration, restructuring, failure to pay, and repudiation/moratorium, but not default, due to the inexis-tence of a formal international bankruptcy court for sovereign issuers (Pan and Singleton, 2008).
The Term Structure of CDS Spreads and Sovereign Credit Risk 24
3.4. Calibration of preferences and of the endowment economy1
The parameters for preferences and aggregate consumption growth, based on a monthly2
decision interval, are summarized in Table 6. The subjective discount factor δ is set to 0.9987,3
while the EIS ψ and the coefficient of relative risk aversion γ are 1.7 and 10 respectively. In4
line with Bansal et al. (2012), consumption dynamics are calibrated to have an annualized5
growth rate of 1.8% and a volatility of 2.5%. The mean of expected consumption growth µx6
is 0.0015. The process has a persistence φx of 0.975 and a volatility leverage coefficient νx7
equal to 0.034. The level of stochastic volatiliy√µσ is calibrated to 0.00725, with the uncon-8
ditional volatility of volatility√ωσ given by 2.8035e-005. Shocks to consumption volatility9
are persistent, with the value of φσ set at 0.9945. Table 6 shows that the dynamics reproduce10
the moments in the data well, both in-population and out-of-sample. The calibrated values11
are fixed for the subsequent analysis. To support the calibration, it is shown in the Online12
Appendix that the model provides reasonable results for the first and second moments of the13
equity premium, the risk-free rate, the real and nominal term structures of interest rates,14
the variance risk premium, and the wealth-consumption and price-dividend ratios.1915
3.5. Time variation in the term structure of CDS spreads16
I next discuss how the tension between global and country-specific shocks generates time17
variation in the slope of the term structure of CDS spreads in the model. Such dynamics18
arise through the joint evolution of risk premia and expected losses. In normal times, the19
term structure of expected losses is flat or slightly decreasing. Risk aversion introduces20
a risk premium and raises the level of CDS spreads. However, the increase is higher for21
longer maturities, provided the Epstein-Zin agent prefers early resolution of uncertainty.22
Following a series of negative shocks, expectations about future default rates become more23
uncertain and more volatile. As a consequence, expected losses increase dramatically around24
19Processes for dividends and inflation are specified in the Online Appendix to examine price-dividendratios and the nominal term structure of interest rates.
The Term Structure of CDS Spreads and Sovereign Credit Risk 25
short maturities. Because of mean reversion in prices, the term structure of expected losses1
becomes steeply downward-sloping. Yet, a flat or increasing term structure of risk premia2
is insufficient to offset the strongly inverted shape of expected losses. The net outcome is3
a negative term structure of spreads. Thus, the joint dynamics of global and local shocks4
together with investor preferences are responsible for time variation in the term structure.5
Global shocks are the dominant force underlying spread variation when the slope is positive.6
An inverted term structure nevertheless indicates that domestic shocks are more important.7
These dynamics are in addition modulated by the sensitivity of countries to each risk factor.8
Figure 3 illustrates the mechanism graphically for a simulated sample path of 600 months9
for a hypothetical country.20 In Panel 3.A, I plot the evolution of the 1-year and 10-year10
spreads. While on average the long-maturity spread is higher, it occasionally falls below11
the short spread. Panel 3.B depicts the slope and the volatility of the default process. It12
is straightforward to see that the term structure inverts when uncertainty about default13
rises sharply. In these situations, expected losses rise more quickly than risk premia, as is14
shown for the 5-year spread in Panel 3.C. Finally, Panel 3.D illustrates that risk premia are15
strongly correlated with global macroeconomic uncertainty. Also note that risk premia are16
only weakly correlated with the country-specific default process.17
The comparative statics depicted in Figure 4 provide deeper insights into the mechanics of18
the model. I plot the difference between the model-implied 10-year and 1-year CDS spreads19
for perturbed values of φλσ and φλ, keeping the mean (µλ) and volatility (ωλ) of default risk20
constant at 0.005 and 5e-04, respectively.21 Panel 4.A shows that, for small values of φλσ,21
the slope becomes more negative as the default process becomes more persistent. Panel 4.B,22
in contrast, shows that, for high values of φλσ, raising the persistence has a positive effect on23
the slope. Thus, all else equal, the slope tends to be more negative for low loadings on the24
global factor and a high persistence of past idiosyncratic shocks. Panel 4.C depicts the effect25
20The default parameters for the simulated path are φλσ = 1.84, φλ = 0.9871, νλ = 1.60e-03, andcλ = 1.59e-04.
21Keeping the mean and volatility constant requires an adjustment to νλ > 0 and cλ > 0. The lines onlydepict the values remaining in their domains.
The Term Structure of CDS Spreads and Sovereign Credit Risk 26
of default volatility on the slope for a constant mean default rate of 0.0039. The outcome is1
reproduced for various combinations of φλσ and φλ. Overall, volatility decreases the slope of2
the term structure.223
One feature of the model is that macroeconomic uncertainty positively contributes to the4
volatility of the default process, which itself impacts the inversion. The global shock thus5
works in two counteracting directions, but the impact on the risk premium channel dominates6
for realistic levels of CDS spreads. Local shocks impact the term structure in the opposite7
direction to the global shock. Depending on the cross-sectional variation in the exposures8
to these two types of shocks, and the intensity of their realizations, one or the other could9
dominate. The Online Appendix contains additional empirical evidence, which suggests10
that both global and local shocks impact the term structure all the time, which generates11
heterogeneity in the cross-section and time series. The observed dynamics are consistent12
with the theoretical properties of the model, which in turn improve the interpretation of the13
evidence.14
3.6. Calibration of the default process15
The default process is calibrated to fit the unconditional moments of the term struc-16
ture for all 44 countries in the sample. The vector of the default parameters is defined by17
Θ = [φλσ, φλ, νλ, cλ]> and note that the CDS spread is a function of expected consump-18
tion growth xt, macroeconomic uncertainty σ2t , and the latent default process λt, that is19
CDS = f (xt, σ2t , λt (Θ)). Like Longstaff et al. (2011), the 5-year CDS spread is assumed to20
be perfectly priced. Assuming that all state variables are at their steady state, and condi-21
tional on a set of starting values for Θ, one can back out λ, which becomes a function of22
the observed mean of the 5-year CDS spread CDS (5) and the steady state values µx and23
µσ, i.e., λ = f(CDS (5) , µx, µσ; Θ
). The implied mean hazard rate λ can be injected back24
22Note that if the slope is very negative, raising volatility increases the slope (it becomes less negative).This reflects an option-type feature in the term structure of CDS spreads. A severely distressed country hasa strongly downward-sloping curve. If the country is close to default, the seller’s position behaves like a deepout-of-the money put. Raising volatility increases the likelihood of a positive payoff and increases the slope.
The Term Structure of CDS Spreads and Sovereign Credit Risk 27
into the pricing equation (4) to generate the term structure of credit spreads. This two-step1
iterative procedure is repeated until the distance between the implied CDS spreads of dif-2
ferent maturities and the observed sample moments is minimized. The recovery rate is kept3
constant at 25%, consistent with industry practice and the findings by Pan and Singleton4
(2008). Due to the structural break in sovereign spreads in the summer of 2007, with spreads5
even for “safe” countries like Finland jumping from an average 5-year CDS spread of 3 to 326
bps, i.e., an increase of 967%, the calibration is separated for the periods before and after7
July 2007.23 The results for the pre-crisis period are not reported, but are available upon8
request.9
The calibration outcomes for the period after July 2007 are reported in Table 7 (columns10
(2) to (5)). The parameter φλσ modulates the sensitivity of the default process to global11
shocks, while φλ determines the persistence of the process. When the global leverage factor12
φλσ is small and the persistence φλ is large, the process is mainly driven by contemporaneous13
and past country-specific shocks, as illustrated in equation (14). If φλσ is large, global shocks14
dominate the behavior of the default process. In addition, the higher its value, the higher15
is the correlation with the stochastic discount factor. Common dependence on the global16
factor also introduces co-movement of spreads across countries. The results indicate that17
there is a systematic difference in the parameter values of countries that exhibit positive or18
negative slopes, on average. For countries with a positive slope, the leverage coefficient on19
the global factor φλσ is large and above 1, ranging between 3.63 for Finland and 91.16 for20
Venezuela. At the same time, the persistence of the default process φλ tends to be smaller21
and below 1, with a maximum of 0.5343 for France. For countries with a negative slope,22
the global leverage factor is below one, while the persistence is larger, ranging from 0.449023
for Cyprus to 0.9797 for Greece. The shape and scale parameters, νλ and cλ, determine24
the density of the default process. The results suggest that countries that have a negative25
slope, on average, exhibit a lower sensitivity to the common risk factor, as well as a greater26
23As another example, spreads jumped from 10 to 300 bps in the case of Malaysia, an increase of 2,900%.
The Term Structure of CDS Spreads and Sovereign Credit Risk 28
sensitivity to lagged idiosyncratic shocks.1
3.7. Asset pricing implications2
For each country, a time series of 120,000 months is simulated to examine the asset3
pricing implications of the general equilibrium model. The CDS pricing model is evaluated4
along three dimensions: (i) the ability to match the factor structure in the levels and slopes5
of CDS spreads; (ii) the ability to fit the moments of the level and slope of CDS spreads;6
and (iii) the frequency of term structure inversion.7
To examine the first dimension, I report in Panel D of Table 5 the results from a PCA8
conducted on the simulated panel data. These results are compared to the factor structure9
of the observed data, which are reported in Panels A, B, and C of Table 5. While the first10
principal component provides explanatory power for about 81% of the variation in the level11
of spreads, it provides explanatory power for only 25% in the slope. This closely resembles12
the factor structure observed in the data, although, unconditionally, the variation in the13
level of spreads identified by the first principal component is a bit lower, with a value of14
56%. Overall, this exercise confirms that the model is able to reconcile the high degree of15
commonality in the level of spreads, despite little commonality in the slope.16
Columns (6) to (8) in Table 7 report the population values of the average 5-year spread17
and the average slope (in bps), along with the fraction of simulated observations with an18
inverted term structure of CDS spreads, respectively. Columns (12) to (13) report the small19
sample moments (5th and 95th percentiles) of the 5-year CDS spread and the slope of spreads,20
based on 1,000 simulations of 100 years (1,200 months). Five-year CDS spreads are close to21
the data (column (9)). Other than Greece, the average difference between the simulated and22
observed spread is 15 bps, or 9% of the observed level of spreads. Across countries, these23
differences range from 0 bps to 59 bps, the latter corresponding to 18% of the 5-year CDS24
spread. Only the fit for Greece is less satisfactory, as the simulated spread of 1,644 bps is25
too high, compared to the observed average of 1,214 bps. Importantly, for all 44 countries,26
The Term Structure of CDS Spreads and Sovereign Credit Risk 29
the sample moment always lies within the 5th and the 95th percentiles of the small sample1
distribution. Thus, the model is able to quantitatively fit the level of the CDS spreads of all2
44 countries in the sample.3
In Table 7, the slope of the term structure tends to be underestimated, with simulated4
averages (column (7)) that are mostly below the observed sample moments reported in col-5
umn (10). For five countries (Lithuania, Venezuela, Cyprus, Greece, and Portugal), the6
average observed slope lies within the 5th and the 95th percentiles of the small sample dis-7
tributions (column (13)), but, in general, the simulated slope of the term structure is too8
flat. The challenge to quantitatively fit the magnitude of the slope of CDS spreads is pri-9
marily attributable to the difficulty in matching short-term spreads, a challenge that is also10
faced by reduced-form credit spread models (Pan and Singleton, 2008). Even though the11
model underestimates the magnitude of the slope, it qualitatively fits the difference between12
10-year and 1-year spreads well, despite the restriction of a unique stochastic discount factor13
imposed for all countries. Reduced-form specifications typically link the risk-adjusted de-14
fault dynamic under the risk-neutral and historical pricing measures through an exogenously15
defined price of risk, whose coefficients are estimated country by country, using both time16
series and term structure data. This is equivalent to assuming a different functional form for17
the stochastic discount factor of each country. Note that the stochastic discount factor is an-18
imated by fundamental macroeconomic shocks and the marginal investor’s attitude towards19
these risks. It provides an economic interpretation of how aggregate shocks feed into the20
term structure of CDS spreads and it enables a study of comparative statics. Although the21
former specification is statistically more flexible, it provides less insight into the economic22
mechanism.23
The third dimension along which I examine the asset pricing implications is with respect24
to the frequency of term structure inversion. The inversion of the term structure of CDS25
spreads is a rare occurrence. Therefore, for some countries in the sample, it could be true26
that we do not have a time series long enough to observe it. On the other hand, for other27
The Term Structure of CDS Spreads and Sovereign Credit Risk 30
countries, we may have too many observations with an inverted curve with respect to long1
run averages. Given that the simulation for some countries yields, on average, an inverted2
term structure, one may raise the concern that the model-implied inversion is, in fact, not3
a rare event. This is, however, not the case. The results in column (8) show that safe4
countries, like Germany, exhibit term structure inversions for 0.10% of all observations in5
the simulated sample. Two countries, Ireland and Portugal, which have a negative mean6
slope in the sample, exhibit an inversion frequency of the term structure of only 3.34% and7
3.24%, respectively (column (11)). The frequency of inversion for Venezuela, on the other8
hand, is 6.33%, higher than for Ireland and Portugal, despite a mean positive slope. The9
highest inversion frequency after Venezuela is accounted for by Greece, as 35.72% of the10
observations in the simulated data have a negative slope. This inversion frequency in the11
simulated data compares to 46.43% in the observed data. The model-implied outcomes12
align reasonably with the ranking of the frequencies of term structure inversions, although13
the ranking is not absolutely perfect. Overall, there is a greater frequency of inversion for14
countries that have a greater sensitivity of the default intensity to local shocks. The model15
qualitatively captures the dynamics of the slope in the data well.16
Finally, column (14) in Table 7 reports model-implied risk premia as a percentage of the 1-17
year and 10-year level of spreads. The term structure of risk premia is upward-sloping. Their18
magnitude hovers around 3% for short maturities up to approximately 10% for 5-year spreads19
(unreported), and 18% at the 10-year maturity.24 Another pattern is that risk premia are20
proportionally smaller for countries that load weakly on the global factor, even though their21
absolute risk premium may be larger. This is due to two reasons: lower values for φλσ imply22
lower correlations with the stochastic discount factor. In addition, countries with a negative23
slope are marked by higher expected losses, which rise more quickly than risk premia. While24
24Longstaff et al. (2011) report 5-year CDS risk premia of 30%, on average. While they allow for a differentstochastic discount factor for each country, I impose a unique pricing kernel for all countries. Note also thatthe autoregressive gamma specification for consumption volatility implies smaller risk premia compared togaussian dynamics, as the practice of assigning positive outcomes to negative realizations of the varianceincreases the correlation with the pricing kernel.
The Term Structure of CDS Spreads and Sovereign Credit Risk 31
it may seem counterintuitive that riskier countries command lower risk premia, it is rational1
from the perspective of a global investor, who, given integrated markets, can diversify away2
idiosyncratic country risk. Only aggregate risk is priced. When the dependence on the global3
factor is low, default risk becomes mainly country-specific. At those times, speculators who4
bet on default are only compensated for expected losses.5
3.8. Fundamental versus uncertainty shocks6
Many countries experienced term structure inversions during the global financial crisis.7
This raises the question whether a global fundamental shock could also lead to an inversion8
of the term structure, challenging the unique association of term structure inversions with9
local fundamental shocks. I therefore explore the distinct impact of global fundamental and10
uncertainty shocks on the term structure of sovereign CDS spreads. Details of this analysis11
are reported in the Online Appendix.12
The data suggest that term structure inversions and reversals occur almost always in13
isolated fashion, staggered across countries and over time. Such patterns are difficult to14
reconcile with the interpretation that common fundamental shocks are responsible for an15
inversion of the term structure, as such shocks would necessarily lead to contemporaneous16
inversions and reversals all the time.17
As a formal test, I examine the cross-sectional relation between the slope and both global18
fundamental and uncertainty shocks using Fama-MacBeth regressions. Global fundamental19
risk is proxied by U.S. consumption growth and global uncertainty risk by the Baker et al.20
(2016) U.S. economic policy uncertainty index. In robustness tests, filtered U.S. expected21
growth and consumption growth volatility serve as additional proxies for global fundamental22
and uncertainty shocks. Unconditional risk loadings (cash flow and uncertainty betas) from23
first-stage regressions of changes in 5-year CDS spreads on fundamental (uncertainty) shocks24
are negative (positive). Coefficients from the projection of the slope on fundamental cash-flow25
betas (uncertainty betas) are negative (positive). Thus, a more negative (positive) exposure26
The Term Structure of CDS Spreads and Sovereign Credit Risk 32
to fundamental shocks (uncertainty shocks) is associated with a steeper slope. This suggests1
that negative fundamental and positive uncertainty shocks have a similar positive impact on2
the term structure. These results hold both at the country level and for 10 portfolios sorted3
monthly on the 5-year CDS spread.4
Economic uncertainty is the only source of global risk affecting default in the baseline
model. I therefore examine an extension that incorporates common fundamental shocks into
the default process. Specifically, the economy in equation (7) is augmented with large drops
in consumption growth (disasters) as proxy for global fundamental shocks:
∆ct+1 = xt + σtεc,t+1 − zt+1, (15)
where the jumps zt+1 are drawn from an independent Poisson-Gamma mixture distribution5
and arrive with time-varying probability πt+1.25 Thus, disasters (or disaster probabilities)6
are allowed to impact each country’s default process, possibly with heterogeneity in the7
sensitivities to common disaster shocks. Asset pricing implications from the augmented8
model suggest that disasters or their probabilities have an even stronger positive impact on9
the term structure than uncertainty shocks.2610
In sum, the additional analysis supports the interpretation that term structure inver-11
sions are more likely driven by country-specific shocks, rather than by common fundamental12
shocks. A negative shock to a (mean-reverting) country-specific source of risk will only im-13
pact default probabilities, which increase more for short-dated tenors and less for long-dated14
tenors, as conditions are expected to normalize over time. This leads to an inversion of15
the term structure of CDS spreads. A large drop in consumption growth, in contrast, will16
also command a risk premium, which is increasing more for longer maturities, because of17
25Specifically, the jump arrival rate jt+1 follows a Poisson distribution with disaster intensity πt+1, jt+1 ∼P(πt+1), and zt+1|jt+1 ∼ Gamma(jt+1, α). The disaster probability follows autoregressive gamma dynamics,similar to those of consumption growth volatility: πt+1 = νπcπ + φππt + ηπ,t+1.
26An alternative source of fundamental shocks could be short-run or long-run consumption shocks. Suchan extension of the model does not allow for term structure inversions either, unless I allow for negative CDSspread levels, a violation of no-arbitrage conditions.
The Term Structure of CDS Spreads and Sovereign Credit Risk 33
preference for early resolution of uncertainty. Global fundamental shocks have thus opposite1
effects on the term structure, similar to uncertainty shocks. The risk premium effect domi-2
nates as the increase in the term structure of risk premia outweighs the decrease in the term3
structure of default probabilities, leading to a steepening of the slope.4
4. Model Validation5
The results of the CDS pricing model suggest two testable predictions for a validation of6
the model. First, risk premia represent a lower fraction of spreads when the slope is negative.7
Second, excess returns are greater when U.S. economic uncertainty is high. The validity of8
these predictions is tested in the following subsections.9
4.1. Risk premia and the slope10
The first asset pricing implication of the model is that, while the relative risk premia11
(rel.RPt), as a fraction of spreads, should be lower, the absolute risk premia (RPt) should12
be higher when the slope is negative. To test this prediction, I estimate the country-specific13
measures of credit risk premia as a linear combination of forward CDS spreads following14
Friewald et al. (2014), who rely on the intuition of Cochrane and Piazzesi (2005) that forward15
spreads contain information about future excess returns.2716
Panel A in Figure 5 depicts the relation between the slope of the term structure of CDS17
27The computation of relative and absolute risk premia involves several steps. First, for a given predictionhorizon τ and a T -year CDS contract STt+τ , I compute forward CDS spreads F τ×Tt , which at time t containinformation about expected future T -year CDS spreads starting at time t+τ . Second, for each CDS contractmaturity Tk ∈ T = {1, 3, 5, 7}, I compute CDS excess changes, defined as RXTk
t+τ = STkt+τ −Fτ×Tkt , and CDS
excess returns, i.e., rel.RXTkt+τ = logSTkt+τ − logF
τ×Tkt . Third, for each country, I construct the cross-maturity
CDS excess change, RXt+τ = 14
∑Tk∈T
RXTkt+τ , and excess return rel.RXt+τ = 1
4
∑Tk∈T
rel.RXTkt+τ . Fourth, I
regress the average excess return and excess change, respectively, on a constant and the term structure of CDSspreads, defined by the current one-year CDS spread and forward CDS spreads of contracts starting in one,three, five, and seven years, and effective for one year. Define the vector Ft =
(1, S1
t , F1×1t , F 3×1
t , F 5×1t , F 7×1
t
)and the corresponding regression coefficients γ = (γ0, γ1, γ2, γ3, γ4, γ5). The estimated absolute and relativerisk premia are then obtained using the relation RPt = −(γRX)>Ft and rel.RPt = −(γrel.RX)>Ft. I maprelative risk premia from percent into a ratio using the transformation erel.RPt , such that I can interpretrelative risk premia as the ratio of CDS spreads under the risk-neutral Q measure and the physical P measure.Results are robust when we use a simpler calculation of excess returns from holding long over short CDS.
The Term Structure of CDS Spreads and Sovereign Credit Risk 34
spreads and relative CDS risk premia in the upper panel; Panel B depicts the absolute1
level of the risk premia. A regression line separately fitted to the right side of the figure2
(corresponding to data points when the slope is positive), and to the left side of the figure3
(corresponding to data points when the slope is negative), highlights a significant drop in4
relative risk premia when the slope becomes inverted. Panel B shows, however, that the level5
of risk premia is increasing in the inversion of the term structure, despite a lower relative risk6
premium. This confirms the model’s implications that expected losses (risk premia) represent7
a greater (lower) fraction of the level of spreads when the slope of the term structure of CDS8
spreads is negative.9
4.2. Excess returns and economic uncertainty10
The CDS pricing model uses U.S. economic uncertainty as a global risk factor. Moreover,11
the findings in column (14) in Table 7 suggest an upward-sloping term structure of risk12
premia. The second testable prediction is, thus, that excess returns should be greater when13
U.S. economic uncertainty is high. To test this conjecture, it is preferable to use a longer14
time series of excess returns, computed using bond prices from an unbalanced panel of 2515
countries starting in December 1985. Specifically, one-year holding period excess returns are16
projected on the Baker et al. (2016) U.S. economic policy uncertainty index, which serves as17
a proxy for U.S. economic uncertainty.18
I collect data on the monthly Citigroup world government bond total return indices in19
USD (former Salomon Smith Barney Indices) from Datastream for five different maturity20
brackets (1-3, 3-5, 5-7, 7-10, 10+ years), and compute yearly holding period returns in excess21
of the return on the JP Morgan one-month total return cash index. For each country c, the22
annual return on an n-period bond in excess of the risk-free yield is defined as rxnc,t+12 =23
pn−1c,t+12 − pnc,t − y1
t , where p denotes the log bond price and y denotes the log yield defined24
as ync,t = −pnc,t/n. Then, for each country and maturity bracket, the regression rxnc,t+12 =25
ac + bcEPUt + εnc,t+12 is evaluated, where rxnc,t+12 denotes the annual excess return for an26
The Term Structure of CDS Spreads and Sovereign Credit Risk 35
n-period bond of country c and EPUt is the economic policy uncertainty index.1
The results in Table 8, by and large, confirm that (i) excess returns are greater in mag-2
nitude for larger contract maturities, and (ii) that excess returns are positively predicted by3
the economic policy uncertainty index. For almost all countries, the average excess return is4
greater for the maturity bucket that contains longer-term bonds. For a few select countries,5
the returns decrease for very long-term bonds (above ten years), which is primarily related to6
significant illiquidity in the long-term maturity brackets (Chaieb et al., 2014). The average7
annual excess bond return across countries ranges from 2.94% for bonds of one to three years,8
to 4.47% for bonds with maturities above ten years. The average coefficient is approximately9
0.06 to 0.07 in panel regressions, and ranges from 0.01 to 0.19 in country regressions. It is10
almost always positive and significant for more than half of all coefficients. Only Poland and11
Singapore load negatively on economic uncertainty, but the coefficients are not statistically12
significant. In addition, the explanatory power is meaningful, on average around 6% to 7%13
percent, ranging up to 28% for New Zealand. In the Online Appendix, it is shown that14
results are robust across a large range of different measures of economic uncertainty taken15
from Gilchrist and Zakrajsek (2012), Bekaert et al. (2013), Jurado et al. (2015), Baker et al.16
(2016). While the results are also robust against measures of financial uncertainty, such as17
the S&P500 option-implied volatility index VIX and the 1-year swaption-implied volatility18
for 10-year USD interest rate swaps, they tend to be weaker. Overall, these results also19
provide validation of the model, and support for the informativeness of the slope about the20
relative importance of global and local risk.21
5. Conclusion22
A country’s term structure of CDS spreads conveys useful information to market partici-23
pants with which to pinpoint the importance of global and domestic risk. Global shocks are24
the primary source of a country’s CDS spread variation when its term structure is upward-25
sloping. In contrast, a country’s spreads are primarily influenced by domestic shocks when26
The Term Structure of CDS Spreads and Sovereign Credit Risk 36
its term structure is downward-sloping. Thus, both global and country-specific risk factors1
impact the dynamics of sovereign credit risk. As the slope of the CDS term structure is2
available at high frequencies, it improves the identification of the importance of local and3
global risk across countries in real time.4
The findings in this paper document several new stylized facts about the slope of the5
term structure of CDS spreads, based on a panel of 44 countries from January 2001 to6
February 2012. Country-specific risk factors explain cross-sectionally a greater fraction of7
the variation in spreads for countries that have had inverted slopes. More importantly, there8
exists a monotonic relation between the duration of term structure inversion and how much9
CDS spread variation is due to domestic risk factors. In addition, the contemporaneous10
impact of domestic stock market returns on spread changes is greater for larger spread11
changes, stronger for shocks that flatten the slope, and fifteen times greater when the slope12
is negative. Finally, the slope is shown to be contemporaneously predictive of quarterly real13
GDP growth, but only when the slope is negative. Consistent with the findings that the slope14
contains information about the cross-sectional differences in the importance of domestic risk15
over time, the dynamics of the slope are found to exhibit a weak factor structure, in contrast16
to the strong commonality documented for the dynamics of the level of spreads.17
I develop a recursive preference-based model with long-run risk for CDS spreads, where18
the underlying default process depends both on aggregate macroeconomic uncertainty and19
domestic risk. The joint dynamics of global uncertainty and country-specific shocks, com-20
bined with investor preferences, economically explain time variation in the slope of the term21
structure. Two empirical tests support the model’s implications. While absolute risk premia22
are greater when the term structure is inverted, they represent a lower fraction of the level23
of CDS spreads compared to when the slope is positive. In addition, excess bond returns24
are positively predicted by U.S. economic uncertainty. Overall, my analysis reconciles differ-25
ences in conclusions in the literature favoring either local or global risk as an unconditionally26
dominant source of sovereign credit risk.27
The Term Structure of CDS Spreads and Sovereign Credit Risk 37
A Acknowledgements1
I am grateful for constructive comments and suggestions from the editor Urban Jermann,2
an anonymous referee, Adrian Alter, David Backus, Laurent Barras, Nicola Borri, Nina3
Boyarchenko, Ines Chaieb, Mike Chernov, Magnus Dahlquist, Itamar Drechsler, Wenxin4
Du, Mathieu Fournier, Nils Friewald, Jean Helwege, Alexander Herbertsson, Jens Hilscher,5
Alexandre Jeanneret, Bige Kahraman, Leonid Kogan, Rajnish Mehra, Sergei Sarkissian,6
Olivier Scaillet, Paolo Sodini, Valeri Sokolovski, Marti G. Subrahmanyam, Saskia ter Ellen,7
Romeo Tedongap, Stijn van Nieuwerburgh, Michael Weber, Fan Yu, Stanley Zin, Irina Zvi-8
adadze; from seminar participants at the University of Melbourne, the University of Sydney,9
the University of New South Wales, the University of Geneva, McGill (Economics), HEC10
Liege, the Amsterdam Business School, HEC Paris, the University of Hong Kong, the Univer-11
sity of Toronto, the Federal Reserve Board of Governors, the University of Virginia McIntire12
School of Commerce, McGill (Finance), Bocconi University, the European Central Bank, the13
Bank of France, Nova Lisbon, Warwick Business School, LUISS Guido Carli, the Stockholm14
School of Economics, the Luxembourg School of Finance; from participants at the 2015 Fixed15
Income Conference, the 2014 NFA, the IFSID 3rd Conference on Derivatives, the 2014 IFM216
Mathematical Finance Days, the 2013 WFA, the 2013 FMA Europe, the Marie Curie ITN17
Conference on Financial Risk Management & Risk Reporting. Feng Jiao provided excellent18
research assistance. I acknowledge outstanding hospitality offered by the Swedish House of19
Finance, and financial support from the Jan Wallander and Tom Hedelius Foundation, the20
Nasdaq-OMX Nordic Foundation, the Bank Research Institute in Stockholm, the Institute of21
Financial Mathematics of Montreal, the Fonds de Recherche du Quebec - Societe et Culture22
grant 2016-NP-191430, the Canada Social Sciences and Humanities Research Council grant23
435-2016-1504.24
The Term Structure of CDS Spreads and Sovereign Credit Risk 38
B Supplementary Material1
Supplementary material can be found in the Online Appendix at http://dx.doi.org/10.1016/j.jmoneco.2
xxxx.xx.xxx.3
References4
Acharya, V. V., Drechsler, I., Schnabl, P., June 2014. A pyrrhic victory? - Bank bailouts5
and sovereign credit risk. The Journal of Finance 69 (6), 2689–2739.6
Ang, A., Longstaff, F. A., 2013. Systemic sovereign credit risk: Lessons from the U.S. and7
Europe. Journal of Monetary Economics 60 (5), 493–510.8
Arellano, C., Ramanarayanan, A., 2012. Default and the maturity structure in sovereign9
bonds. Journal of Political Economy 120 (2), 187–232.10
Arora, N., Gandhi, P., Longstaff, F. A., 2012. Counterparty credit risk and the credit default11
swap market. Journal of Financial Economics 103 (2), 280–293.12
Augustin, P., 2014. Sovereign credit default swap premia. Journal of Investment Management13
12 (2), 65–102.14
Augustin, P., Tedongap, R., April 2016. Real economic shocks and sovereign credit risk.15
Journal of Financial and Quantitative Analysis 51 (2), 541–587.16
Baker, S. R., Bloom, N., Davis, S. J., 2016. Measuring economic policy uncertainty. Quarterly17
Journal of Economics 131 (4), 1593–1636.18
Bansal, R., Khatchatrian, V., Yaron, A., 2005. Interpretable asset markets? European19
Economic Review 49 (3), 531–560.20
Bansal, R., Kiku, D., Yaron, A., 2012. An empirical evaluation of the long-run risks model21
for asset prices. Critical Finance Review 1 (1), 183–221.22
The Term Structure of CDS Spreads and Sovereign Credit Risk 39
Bansal, R., Yaron, A., 2004. Risks for the long run: A potential resolution of asset pricing1
puzzles. The Journal of Finance 59 (4), 1481–1509.2
Bekaert, G., Hoerova, M., Duca, M. L., 2013. Risk, uncertainty and monetary policy. Journal3
of Monetary Economics 60 (7), 771–788.4
Benzoni, L., Collin-Dufresne, P., Goldstein, R. S., Helwege, J., 2015. Modeling credit conta-5
gion via the updating of fragile beliefs. Review of Financial Studies 28 (7), 1960–2008.6
Berndt, A., Douglas, R., Duffie, D., Ferguson, M., 2018. Corporate Credit Risk Premia.7
Review of Finance 22 (2), 419–454.8
Binsbergen, J., Brandt, M., Koijen, R., 2012. On the timing and pricing of dividends. Amer-9
ican Economic Review 102 (4), 1596–1618.10
Binsbergen, J., Hueskes, W., Koijen, R., Vrugt, E., 2013. Equity yields. Journal of Financial11
Economics 110 (3), 503–519.12
Borri, N., Verdelhan, A., 2016. Sovereign risk premia. SSRN eLibrary, Working Paper13
1343746.14
Chaieb, I., Errunza, V. R., Gibson, R., 2014. Integration of sovereign bonds markets: Time15
variation and maturity effects. Swiss Finance Institute Research Paper No. 14-47.16
Chernov, M., Schmid, L., Schneider, A., 2017. A macrofinance view of U.S. sovereign cds17
premiums. Working Paper UCLA.18
Cochrane, J. H., Piazzesi, M., 2005. Bond risk premia. The American Economic Review19
95 (1), 138–160.20
Collin-Dufresne, P., Goldstein, R. S., Yang, F., 2012. On the relative pricing of long-maturity21
index options and collateralized debt obligations. The Journal of Finance 67 (6), 1983–22
2014.23
The Term Structure of CDS Spreads and Sovereign Credit Risk 40
Dieckmann, S., Plank, T., 2011. Default risk of advanced economies: An empirical analysis1
of credit default swaps during the financial crisis. Review of Finance 16 (4), 903–934.2
Du, W., Gordy, M. B., Gadgil, S., Vega, C., 2017. Counterparty risk and counterparty choice3
in the credit default swap market. Working Paper, Federal Reserve Board.4
Duffie, D., 1999. Credit swap valuation. Financial Analysts Journal 55 (1), 73–87.5
Duffie, D., Pedersen, L. H., Singleton, K. J., 2003. Modeling sovereign yield spreads: A case6
study of russian debt. The Journal of Finance 58 (1), 119–159.7
Epstein, L. G., Zin, S. E., 1989. Substitution, risk aversion, and the temporal behavior of8
consumption and asset returns: A theoretical framework. Econometrica 57 (4), 937–969.9
Eraker, B., December 2008. Affine general equilibrium models. Management Science 54 (12),10
2068–2080.11
Fleckenstein, M., Longstaff, F. A., Lustig, H., 2017. Deflation risk. The Review of Financial12
Studies 30 (8), 2719-2760.13
Friewald, N., Wagner, C., Zechner, J., 2014. The cross-section of credit risk premia and14
equity returns. The Journal of Finance 69 (6), 2419–2469.15
Gennaioli, N., Martin, A., Rossi, S., 2012. Sovereign default, domestic banks, and financial16
institutions. The Journal of Finance 69 (2), 819–866.17
Geyer, A., Kossmeier, S., Pichler, S., 2004. Measuring systematic risk in EMU government18
yield spreads. Review of Finance 8 (2), 171–197.19
Giglio, S., March 2017. Credit default swap spreads and systemic financial risk. Working20
Paper, University of Chicago.21
Gilchrist, S., Zakrajsek, E., 2012. Credit spreads and business cycle fluctuations. American22
Economic Review 102 (4), 1692–1720.23
The Term Structure of CDS Spreads and Sovereign Credit Risk 41
Gourieroux, C., Jasiak, J., 2006. Autoregressive gamma processes. Journal of Forecasting1
25 (2), 129–152.2
Gruber, P. H., Tebaldi, C., Trojani, F., 2017. The price of the smile and variance risk premia.3
Swiss Finance Institute Research Paper No. 15-36.4
Hansen, L. P., Heaton, J. C., Li, N., 2008. Consumption strikes back? Measuring long-run5
risk. The Journal of Political Economy 116 (2), 260–302.6
Hilscher, J., Nosbusch, Y., 2010. Determinants of sovereign risk: Macroeconomic fundamen-7
tals and the pricing of sovereign debt. Review of Finance 14 (2), 235–262.8
Jeanneret, A., 2015. The dynamics of sovereign credit risk. Journal of Financial and Quan-9
titative Analysis 50 (5), 963–985.10
Jotikasthira, C., Le, A., Lundblad, C., 2015. Why do term structures in different currencies11
co-move? Journal of Financial Economics 115 (1), 58–83.12
Jurado, K., Ludvigson, S. C., Ng, S., 2015. Measuring uncertainty. American Economic13
Review 105 (3), 1177–1216.14
Kallestrup, R., Lando, D., Murgoci, A., 2016. Financial sector linkages and the dynamics of15
bank and sovereign credit spreads. Journal of Empirical Finance 38 (A), 374–393.16
Kandel, S., Stambaugh, R. F., 1990. Expectations and volatility of consumption and asset17
returns. The Review of Financial Studies 3 (2), 207–232.18
Le, A., Singleton, K. J., Dai, Q., 2010. Discrete-time affineq term structure models with19
generalized market prices of risk. Review of Financial Studies 23 (5), 2184–2227.20
Lettau, M., Ludvigson, S., Wachter, J., 2006. The declining equity premium: What role does21
macroeconomic risk play? Review of Financial Studies 21 (4), 1653–1687.22
The Term Structure of CDS Spreads and Sovereign Credit Risk 42
Longstaff, F. A., Mithal, S., Neis, E., Oct. 2005. Corporate yield spreads: Default risk or1
liquidity? New evidence from the credit default swap market. The Journal of Finance2
60 (5), 2213–2253.3
Longstaff, F. A., Pan, J., Pedersen, L. H., Singleton, K. J., 2011. How sovereign is sovereign4
credit risk? American Economic Journal: Macroeconomics 3 (2), 75–103.5
Lustig, H., Stathopoulos, A., Verdelhan, A., 2017. Nominal exchange rate stationarity and6
long-term bond returns. Working Paper, Stanford University.7
Mauro, P., Sussman, N., Yafeh, Y., 2002. Emerging market spreads: Then versus now.8
Quarterly Journal of Economics 117 (2), 695–733.9
Obstfeld, M., Rogoff, K. S., October 2009. Global imbalances and the financial crisis: Prod-10
ucts of common causes. Unpublished Working Paper, Harvard Business School.11
Pan, J., Singleton, K. J., 2008. Default and recovery implicit in the term structure of12
sovereign cds spreads. The Journal of Finance 63 (5), 2345–2384.13
Reinhart, C. M., Rogoff, K. S., 2008. This time is different: A panoramic view of eight14
centuries of financial crises. NBER Working Papers 13882, National Bureau of Economic15
Research, Inc.16
Remolona, E., Scatigna, M., Wu, E., 2008. The dynamic pricing of sovereign risk in emerging17
markets: Fundamentals and risk aversion. Journal of Fixed Income 17 (4), 57–71.18
Stock, J., Watson, M., 2002. Has the business cycle changed and why? in M. Gertler and19
Kenneth Rogoff, NBER Macroeconomics Annual: 2002, MIT Press Cambridge.20
Uribe, M., Yue, V. Z., 2006. Country spreads and emerging countries: Who drives whom?21
Journal of International Economics 69 (1), 6–36.22
Verdelhan, A., 2018. The share of systematic risk in bilateral exchange rates. The Journal23
of Finance 73 (1), 375–418.24
The Term Structure of CDS Spreads and Sovereign Credit Risk 43
Weil, P., 1989. The equity premium puzzle and the risk-free rate puzzle. Journal of Monetary1
Economics 24 (3), 401–421.2
Zhang, F. X., 2008. Market expectations and default risk premium in credit default swap3
prices: A study of argentine default. Journal of Fixed Income 18 (1), 37–55.4
Zviadadze, I., 2017. Term-structure of consumption risk premia in the cross-section of cur-5
rency returns. The Journal of Finance 72 (4), 1529–1566.6
The Term Structure of CDS Spreads and Sovereign Credit Risk 44
Figure 1: The Slope of the CDS Term Structure and Local Ratios
The local ratio (LR) denotes the ratio of the adjusted R2 from the restricted regression of changes in 5-year
CDS spreads on local variables only to that from the unrestricted regression on local and global variables.
In the figure, I plot the LRs for the countries that have had inverted term structures against the number
of months for which the term structure was inverted. For illustration purposes, Venezuela is excluded from
the graphs. It has a LR of 0.05 and its term structure was inverted for 26 months. Figure 1.A (Figure 1.B)
excludes (includes) the countries that did not have an inverted slope.
(Figure 1.A)
Korea
Slovakia
Bulgaria
Brazil
Portugal
Hungary
Philippines
Ireland
Italy
Spain
Romania
Greece
Uruguay
Cyprus
Russia
Croatia
Czech Rep.
Lithuania
Turkey
.2.4
.6.8
11.
2E
xpla
ined
by
Loca
l Ris
k (L
R)
0 5 10 15 20 25# Months of Inverted Slope
linear fit 95% CIR2 = 32% t-stat = 3.11
Local Risk and Negative Slope
(Figure 1.B)
Sweden
France
PeruThailandBelgiumColombia
Morocco
Panama
Austria
Lebanon
Poland
Germany
China
JapanFinlandSlovenia
Malaysia
Qatar
Denmark
Chile
Israel
Korea
SlovakiaBulgaria
Brazil
Portugal
Hungary
Philippines
Ireland
Italy
Spain
Romania
Greece
UruguayCyprus
Russia
Croatia
Czech Rep.
Lithuania
Turkey
Sth. Africa
0.3
.6.9
1.2
Exp
lain
ed b
y Lo
cal R
isk
(LR
)
0 5 10 15 20 25# Months of Inverted Slope
linear fit 95% CIR2 = 45% t-stat = 5.76
Local Risk and Negative Slope
The Term Structure of CDS Spreads and Sovereign Credit Risk 45
Figure 2: Real GDP Growth and the Term Structure of CDS Spreads
In Figure 2.A, I plot the average real quarter-on-quarter GDP growth (left axis, dashed line) and the
proportion of observations with negative real seasonally adjusted quarter-on-quarter GDP growth (right
axis, solid line) as a function of the average slope of the CDS term structure, in sliding windows of 100
growth-slope observations. The observations are ranked according to their slope in ascending order. For
instance, the left-most point reflects the average real GDP growth (or the proportion of countries with
negative growth) against the average CDS slope among the 100 observations with the lowest (most negative)
slope. The next point reflects the averages for observations 2 to 101, and so forth. The right-most points
will be computed based on fewer than 100 observations. In Figure 2.B, I restrict the average slope to be
more than -250 bps. In Figures 2.C and 2.D, I plot the fraction of observations with a negative slope and
the average slope as a function of the average 5-year CDS spread level in sliding windows of 100 level-slope
observations, where the observations are ranked according to their 5-year CDS spread in ascending order.
(Figure 2.A)
0.2
.4.6
.8F
ract
ion
of o
bser
vatio
ns w
ith n
egat
ive
grow
th
-20
24
6A
vera
ge r
eal G
DP
gro
wth
(%
), q
oq
-1000 -500 0 500Mean 10y-1y CDS (Slope), in bps
Real GDP growth (%) Negative Growth (Fraction)
(Figure 2.B)
0.2
.4.6
.8F
ract
ion
of o
bser
vatio
ns w
ith n
egat
ive
grow
th
-20
24
6A
vera
ge r
eal G
DP
gro
wth
(%
), q
oq
-200 0 200 400 600Mean 10y-1y CDS (Slope), in bps
Real GDP growth (%) Negative Growth (Fraction)
(Figure 2.C)
0.2
.4.6
.8#
Neg
ativ
e S
lope
- F
ract
ion
(%)
0 500 1000 1500 2000Mean 5-year CDS (conditional on CDS<2,000 bps)
(Figure 2.D)
-600
-400
-200
020
0M
ean
Slo
pe (
CD
S10
y-C
DS
1y)
0 500 1000 1500 2000Mean 5-year CDS (conditional on CDS<2,000 bps)
The Term Structure of CDS Spreads and Sovereign Credit Risk 46
Figure 3: Time Variation in the Term Structure
In this figure, I plot model-implied metrics for a simulated sample path of 600 months. The default pa-
rameters used for the simulation are the calibrated values for Uruguay, that is φλσ = 1.84, φλ = 0.9871,
νλ = 1.60e-03 and cλ = 1.59e-04. In Panel 3.A, I plot the evolution of the 1-year (dash-dotted line) and
10-year (solid line) CDS spreads in bps. Panel 3.B illustrates the slope of the CDS curve in bps (solid line,
left scale) versus the conditional volatility of the default process (dash-dotted line, right scale). Panel 3.C
shows the simulated time series for the risk premium (solid line, right axis) and expected loss (dash-dotted
line, left axis) in bps. In Panel 3.D, I plot the evolution of aggregate macroeconomic uncertainty (solid
line, left axis), annualized and in %, against the risk premium in bps (dash-dotted line, right axis), and the
default process λt (solid line with dotted markers, right axis). The default process is multiplied by 1,000 for
better visualization.
(Figure 3.A)
0 100 200 300 400 500 6000
500
1000
1500
2000
2500
CDS Spread (bps)
Time Series - Hypothetical Sample Path
1-year and 10-year CDS Spread
CDS1y
CDS10y
(Figure 3.B)
0 100 200 300 400 500 600-600
-400
-200
0
200
400
600Slope 10y-1y CDS Spread (bps)
Time Series - Hypothetical Sample Path
Slope 10y-1y
0 100 200 300 400 500 600
0
1
2
3
4
5
6
7
8
x 10-6
Dynamics of the Slope
Conditional Volatility of default risk
Default Risk Volatility
(Figure 3.C)
0 100 200 300 400 500 6000
200
400
600
800
1000
1200
1400
1600
1800
5-year Expected Loss (bps)
Time Series - Hypothetical Sample Path
Expected Loss 5y
0 100 200 300 400 500 600
10
15
20
25
30
35
Expected Loss and Risk Premium
5-year Risk Premium (bps)
Risk Premium 5y
(Figure 3.D)
0 100 200 300 400 500 6000.6
0.8
1
1.2
1.4
Time Series - Hypothetical Sample Path
Global Volatility, Risk Premia and Default Risk
Glo
bal V
ola
tility
(%
)
0 100 200 300 400 500 6000
10
20
30
40
Ris
k P
rem
ium
(b
ps)
an
d D
efa
ult
Ris
k (
x 1
000)Consumption Volatility
Default Risk
Risk Premium 5y
The Term Structure of CDS Spreads and Sovereign Credit Risk 47
Figure 4: Comparative Statics of the CDS Term Structure
Panels 4.A and 4.B illustrate the sensitivity of the slope of the CDS term structure (in bps) when φλ and φλσ
are perturbed, keeping the mean µλ and volatility of default ωλ constant at 0.005 and 5e-04, respectively. In
Panel 4.A, I perturb φλ for different values of φλσ equal to 1 (dotted red line), 2.6 (solid line), 42 (dash-dotted
line), and 66 (dashed line). In Panel 4.B, I perturb φλσ for different values of φλ equal to 0.1 (dotted line),
0.25 (dash-dotted line), 0.5 (dashed line), and 0.75 (solid line). Keeping the mean and volatility constant
requires an adjustment to νλ > 0 and cλ > 0. The lines are plotted for values remaining in their respective
domains. Panel 4.C illustrates a similar analysis for the slope of the CDS term structure by perturbing the
volatility of the default process ωλ and keeping the mean default rate µλ constant at 0.0039. The outcome
is reproduced for different combinations of φλ and φλσ, that is 0.90 and 7.5 (dotted line), 0.50 and 37.5
(dashed-dotted line line), 0.20 and 60 (dashed line), 0.01 and 74.25 (solid line).
(Figure 4.A)
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9-30
-20
-10
0
10
20
30
40
50
60
CD
S10y
-CD
S1y
(bps
)
= 0.005,
= 5e-4
= 1
= 2.6
= 42
= 66
(Figure 4.B)
10 20 30 40 50 60 70 800
10
20
30
40
50
60
CD
S10y
-CD
S1y
(bps
)
= 0.005,
= 5e-4
= 0.1
= 0.25
= 0.5
= 0.75
(Figure 4.C)
0.001 0.01 0.02 0.03 0.04 0.05
-30
-20
-10
0
10
20
30
40
50
CD
S10y
-CD
S1y
(bps
)
= 0.0039
= 7.5, = 0.90
= 37.5, = 0.5
= 60, = 0.20
= 74.25, = 0.01
The Term Structure of CDS Spreads and Sovereign Credit Risk 48
Figure 5: Relation between Risk Premia and the Term Structure
In Panels 5.A and 5.B, I plot the relation between the slope of the term structure of CDS spreads (CDS10y-
CDS1y) and either the relative CDS risk premia (Panel 5.A) or the absolute level of risk premia (Panel
5.B). The computation of relative and absolute risk premia follows Friewald et al. (2014). First, for a
prediction horizon τ and a T -year CDS contract STt+τ , I compute forward CDS spreads F τ×Tt , which at
time t contain information about expected future T -year CDS spreads starting at time t + τ . Second, for
each maturity Tk ∈ T = {1, 3, 5, 7}, I compute CDS excess changes RXTkt+τ = STkt+τ − F τ×Tkt , and CDS
excess returns rel.RXTkt+τ = logSTkt+τ − logF
τ×Tkt . Third, for each country, I construct the cross-maturity
CDS excess change RXt+τ = 14
∑Tk∈T
RXTkt+τ , and excess return rel.RXt+τ = 1
4
∑Tk∈T
rel.RXTkt+τ . Fourth, I
regress the average excess return and excess change on a constant and the term structure of CDS spreads,
defined by the current one-year CDS spread and forward CDS spreads of contracts starting in one, three,
five, and seven years, and effective for one year. Define the vector Ft =(1, S1
t , F1×1t , F 3×1
t , F 5×1t , F 7×1
t
)and
the corresponding regression coefficients γ = (γ0, γ1, γ2, γ3, γ4, γ5). The estimated absolute and relative risk
premia are obtained using RPt = −(γRX)>Ft and rel.RPt = −(γrel.RX)>Ft. I map relative risk premia
from percentage into a ratio using the transformation erel.RPt , such that I can interpret relative risk premia
as the ratio of CDS spreads under the risk-neutral Q and the physical P measures.
(Figure 5.A)
.99
.995
11.
005
1.01
CD
S R
elat
ive
Ris
k P
rem
ia (
~C
DS
(Q)/
CD
S(P
))
-1000 -500 0 500 1000Slope CDS10y-CDS1y (bps)
(Figure 5.B)
-100
0-5
000
500
Leve
l CD
S R
isk
Pre
mia
-500 0 500Slope CDS10y-CDS1y (bps)
The Term Structure of CDS Spreads and Sovereign Credit Risk 49
Table 1: Summary Statistics of Sovereign CDS Spreads
This table presents summary statistics for sovereign CDS spreads. In columns (1) to (3), I report the namesof the 44 countries in the sample, the first monthly observation in the panel, and the number of observations,respectively. All series end in February 2012. In columns (4) to (6), I report the sample average (in bps)CDS spread levels, for swap maturities of 1, 5, and 10 years. The average slope of the CDS term structure,measured as the difference between 10- and 1-year CDS spreads, is documented in column (7). I report thenumber and frequency of months during which the term structure was inverted in columns (8) and (9).
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Country Information Spread Levels Slope < 0 Slope
Country Start Obs 1y 5y 10y 10y-1y N Freq
Austria 2001-10 125 22 37 41 19 0 0.00Belgium 2001-2 133 29 44 47 19 0 0.00Brazil 2001-1 134 318 498 530 212 7 0.05Bulgaria 2001-4 131 113 203 233 120 5 0.04Chile 2002-2 121 34 80 101 67 0 0.00China 2001-1 134 29 58 75 46 0 0.00Colombia 2001-4 131 140 329 374 234 0 0.00Croatia 2001-2 133 107 161 183 76 4 0.03Cyprus 2002-7 116 129 133 133 3 10 0.09Czech Rep. 2001-4 131 30 49 58 28 2 0.02Denmark 2003-1 110 14 26 30 16 0 0.00Egypt 2002-4 119 150 251 290 140 0 0.00Finland 2002-9 114 9 17 21 12 0 0.00France 2002-8 115 17 33 38 20 0 0.00Germany 2002-10 113 9 21 26 17 0 0.00Greece 2001-1 134 814 515 433 -382 26 0.19Hungary 2001-2 133 99 136 147 48 5 0.04Ireland 2003-1 110 162 155 139 -23 22 0.20Israel 2001-11 124 47 95 115 68 0 0.00Italy 2001-1 134 51 69 74 23 4 0.03Japan 2001-2 133 12 32 45 33 0 0.00Korea 2001-3 132 58 89 106 48 1 0.01Lebanon 2003-3 108 300 413 460 159 0 0.00Lithuania 2002-9 114 125 150 157 32 15 0.13Malaysia 2001-4 131 43 85 108 66 0 0.00Mexico 2001-1 134 66 152 187 122 0 0.00Morocco 2001-4 131 101 201 242 141 0 0.00Panama 2002-7 116 77 212 256 179 0 0.00Peru 2002-2 121 94 260 306 212 0 0.00Philipp. 2001-3 132 138 307 369 231 4 0.03Poland 2001-1 134 44 78 92 48 0 0.00Portugal 2002-2 121 170 155 139 -31 22 0.18Qatar 2001-9 126 44 79 102 58 0 0.00Romania 2002-7 116 139 218 239 100 6 0.05Russia 2001-10 125 137 226 261 124 7 0.06Slovakia 2001-11 124 37 63 74 37 1 0.01Slovenia 2002-2 121 36 55 62 27 0 0.00Sth. Afr. 2001-3 132 70 141 173 103 0 0.00Spain 2001-7 128 55 72 73 18 4 0.03Sweden 2001-5 130 11 20 23 13 0 0.00Thailand 2001-5 130 47 92 115 68 0 0.00Turkey 2001-1 134 251 397 440 188 8 0.06Uruguay 2002-4 119 790 731 712 -78 19 0.16Venezuela 2001-2 133 695 868 865 170 26 0.20
The Term Structure of CDS Spreads and Sovereign Credit Risk 50
Table 2: Local Ratios
The local ratio (LR) denotes the ratio of the adjusted R2 from the restricted regression of changes in 5-yearCDS spreads on local factors to that from the unrestricted regression on local and global factors. Countriesare grouped into two categories. G1 contains all countries that never had an inverted slope. G2 containsall countries with at least one month of inverted slope. The restricted sample excludes outliers Egypt andMexico in G1 and Venezuela in G2. Inferences are obtained by block-bootstrapping 10,000 times a samplesize of 36 months for each country. A one-sided t-test on the equality of means, assuming paired data,against the alternative that G1 has a smaller mean, is rejected at the 1% significance level. The Wilcoxonmatched-pairs signed-rank test rejects the null hypothesis that both distributions are the same, while theone-sided sign test rejects equality of medians against the alternative that the median of G1 is lower.
Full Sample (44 countries) Restricted Sample (41 countries)Mean LR (%) Median LR (%) Mean LR (%) Median LR (%)
Group 1: Slope was never inverted 40 38 36 36Group 2: Slope was inverted 61 63 64 64
t-test p < 0.01 p < 0.01Sign test p < 0.01 p < 0.01Wilcoxon signed-rank test p < 0.01 p < 0.01
The Term Structure of CDS Spreads and Sovereign Credit Risk 51
Table 3: Dynamics of Sovereign Credit Risk - Panel and Quantile Regressions
Panel A reports the coefficients of interest (for brevity) from equation (1), augmented with interaction termsbetween the risk factors and an indicator variable equal to one, if the term structure is negative, and zerootherwise. Column (1) reports the results of a restricted specification that uses the 5 reported variables.Column (2) includes all controls. The specifications in columns (3) to (6) contain country fixed effects, and Isuccessively introduce clustering by country, by month, and by time and month. In column (7), I use Driscoll-Kraay standard errors with country fixed effects, accounting for cross-sectional and time series correlationof three lags. The specification in column (8) includes the full set of interaction terms. Columns (1) and (2)in Panel B report results from sample splits for a specification similar to that of column (3) in Panel A. Thesample is split for CDS spread changes that steepen the slope (column (1): ∆CDS · I(∆Slope > 0)) andthat flatten the slope (column (2): ∆CDS · I(∆Slope < 0)). Column (3) in Panel B includes the full sampleand an interaction term that is equal to one if the slope flattens, and zero otherwise. Panel C reports thecoefficients for the local stock market return and its interaction with the indicator variable from quantileregressions with a specification similar to that of column (3) in Panel A. *, ** and ***, denote significanceat the 1%, 5%, and 10% levels.
Panel A: 5-year CDS Spread Changes
Variables (1) (2) (3) (4) (5) (6) (7) (8)
LocalRet -0.01*** -0.01*** -0.01*** -0.01*** -0.01*** -0.01*** -0.01*** -0.01***LocalRet× I(Slope < 0) -0.15** -0.15** -0.15** -0.15** -0.15** -0.15** -0.15** -0.16**I(Slope < 0) 0.00 0.00 0.00 0.00 0.00 0.00 0.00 -0.01USret -0.03*** -0.03*** -0.03*** -0.03*** -0.03*** -0.03*** -0.03*** -0.03***USret× I(Slope < 0) -0.11 -0.10 -0.11* -0.11* -0.11 -0.10* -0.11 -0.10
Country FE No No Yes Yes Yes Yes Yes YesFcluster No No No Yes No Yes 3lags 3lagsTcluster No No No No Yes Yes Yes YesControls No Yes Yes Yes Yes Yes Yes YesControls×I(Slope < 0) No No No No No No No YesAdj.R2 0.17 0.19 0.19 0.19 0.18 0.19 0.19 0.24N 5476 5456 5456 5456 5456 5456 5456 5456
Panel B: Conditional 5-year CDS Spread Changes
Variables (1) (2) Variables (3)
LocalRet -0.0057*** -0.0125*** LocalRet -0.0060**LocalRet× I(Slope < 0) -0.0701 -0.1785** LocalRet× I(∆Slope < 0) -0.0151**I(Slope < 0) -0.0085* 0.0078*** I(∆Slope < 0) -0.7084***USret -0.0277*** -0.0270*** Usret -0.0391***USret× I(Slope < 0) -0.0282 -0.1117* USret× I(∆Slope < 0) -0.0005
Adj.R2 0.18 0.24 Adj.R2 0.39N 2805 2651 N 5456
Panel C: Quantiles of 5-year CDS Spread Changes
Quantile 0.5 0.75 0.9 0.95
LocalRet -0.0030*** -0.0045*** -0.0064 -0.0087LocalRet× I(Slope < 0) -0.0441*** -0.0811*** -0.1536*** -0.2700***
The Term Structure of CDS Spreads and Sovereign Credit Risk 52
Table 4: Growth Regressions
This table presents the results from regressions of seasonally adjusted quarter-on-quarter real GDP growth
on the level (CDS5y) and the slope (CDS10y-CDS1y) of CDS spreads, as well as their interactions. The level
is defined as the quarterly 5y CDS spread, where the quarterly spread is defined as the average spread over
all monthly observations. The slope is defined as the difference between the 10y and the 1y CDS spreads.
I(Slope< 0) is equal to one if the slope is negative in a quarter, and zero otherwise. The frequency of the
regressions is quarterly. All regressions are clustered at the country level, and the regressions for the results
in columns (4) to (6) contain country fixed effects. In the regression for the results in column (5), I use
Driscoll-Kraay robust standard errors that account for arbitrary spatial dependence and serial correlation
up to four quarters. The results in column (6) exclude Greece. *, ** and ***, denote significance at the 1%,
5%, and 10% levels. Source: Markit, OECD, Thomson Reuters Datatsream, IHS Global Insight, Oxford
Economics.
Variables (1) (2) (3) (4) (5) (6)
Level(CDS5y) -27.46*** -31.20*** -38.83** -53.40*** -53.40*** -54.89***(7.76) (9.25) (15.06) (15.74) (16.21) (15.99)
Slope(CDS10y − CDS1y) 45.01* 61.51* 10.23 10.23 15.08(24.55) (34.02) (23.12) (19.62) (22.89)
I(Slope < 0) -2.41* -2.07* -2.07** -1.81(1.30) (1.21) (0.96) (1.17)
I(Slope < 0)× Slope -142.79** -135.51* -135.51* -155.92**(69.41) (67.24) (74.84) (71.89)
Constant 2.86*** 2.58*** 2.58*** 3.21*** 3.21*** 3.20***(0.40) (0.40) (0.41) (0.20) (0.43) (0.20)
Observations 1,672 1,672 1,672 1,672 1,672 1,632Country FE No No No Yes Yes YesCluster Country Yes Yes Yes Yes No YesDK No No No No Yes NoLags – – – – 4 –Greece Yes Yes Yes Yes Yes NoAdj.R2(%) 1.69 2.49 3.01 3.59 3.82 3.27
The Term Structure of CDS Spreads and Sovereign Credit Risk 53
Table 5: Principal Component Analysis
This table reports the variation in CDS spread changes explained by the first three principal components.Panel A presents results from a principal component analysis (PCA) conducted on 5-year CDS spreads. PanelB presents results from a PCA conducted on the level of spreads, using the maturities of 1, 2, 3, 5, 7, and 10years. Panel C presents results from a PCA conducted on the slope of the term structure of CDS spreads.The slope is defined as the difference between the 10-year and 1-year spreads. The column Full Samplerefers to the results from the entire sample period from January 2003 to February 2012. The subperiodsrefer to the time before the financial crisis (Jan2003-Dec2006), the financial crisis (Jan2007-Dec2010) andthe sovereign debt crisis (Jan2011-Feb2012). Panel D reports the results from a PCA conducted on changesin the 5-year CDS spread levels and on changes in the slope of spreads obtained from a simulated panel of44 countries over 120,000 months.
Full Sample 2003-2006 2007-2010 2011-2012
Panel A: 44 countries - 5-year spreads
% Cumulative
PC1 56.52 56.52PC2 8.14 64.66PC3 4.44 69.10
% Cumulative
42.63 42.6315.25 57.8811.95 69.83
% Cumulative
75.36 75.366.05 81.414.74 86.15
% Cumulative
57.96 57.9613.33 71.297.56 78.85
Panel B: 30 countries - Term structure of spreads
% Cumulative
PC1 54.59 54.59PC2 10.04 64.63PC3 5.03 69.66
% Cumulative
54.60 54.6012.74 67.347.43 74.77
% Cumulative
75.84 75.845.71 81.553.70 85.26
% Cumulative
62.70 62.7012.82 75.529.26 84.78
Panel C: 30 countries - Slope of spreads
% Cumulative
PC1 21.79 21.79PC2 9.26 31.06PC3 7.92 38.98
% Cumulative
24.20 24.2016.35 40.5510.38 50.93
% Cumulative
38.00 38.0012.85 50.8510.22 61.07
% Cumulative
31.30 31.3018.32 49.6212.43 62.05
Panel D: Model (120,000 months of simulated data)
44 countries - 5-year spreads 44 countries - Slope of spreads
% Cumulative
PC1 80.58 80.58PC2 2.30 82.87PC3 2.28 85.15
% Cumulative
PC1 25.06 25.06PC2 16.22 41.28PC3 2.32 43.61
The Term Structure of CDS Spreads and Sovereign Credit Risk 54
Table 6: Model Parameter Calibration
Panels A and B in this table reports model and preference parameter values, which are calibrated at amonthly decision interval. Panel C reports the endogenous coefficients of the wealth-consumption ratio.Panels D and E present moments of consumption dynamics from the data and the model. The data are real,sampled at an annual frequency, and cover the period 1929 to 2011. Standard errors are Newey-West withone lag. For the model, I report percentiles of these statistics based on 10,000 simulations of 600 months,equalling 50 years of data. The column Pop reports population statistics based on a long simulation of 1.2million months. All statistics are time-averaged. Data for consumption growth are taken from the Bureauof Economic Analysis National Income and Product Accounts Tables. The parameters νσ > 0 and cσ > 0are linked to the unconditional mean µσ and variance ωσ of the volatility process by µσ = (νσcσ) / (1− φσ)
and ωσ =(νσc
2σ
)/ (1− φσ)
2.
Panel A: Preference Parameter Values
Subjective discount factor δ 0.9987Intertemporal elasticity of substitution ψ 1.7
Coefficient of relative risk aversion γ 10
Panel B: Consumption Growth Dynamics
Mean consumption growth µx 0.0015Persistence of expected consumption growth φx 0.975
Sensitivity to long-run risk shocks νx 0.034Persistence of volatility φσ 0.9945
Volatility level√µσ 0.00725
Volatility of volatility√ωσ 2.8035e-005
Panel C: Coefficients of the wealth-consumption ratio - Model
Ac0 Ac1 Ac2
6.85 15.80 -1085.18
Panel D: Consumption - Model
Mean (%) 1% 5% 50% 95% 99% Pop
E [∆c] 1.79 0.40 0.83 1.78 2.77 3.20 1.80σ [∆c] 2.30 0.24 1.61 2.26 3.13 3.50 2.38AC1 [∆c] 0.34 -0.01 0.11 0.35 0.57 0.65 0.41
Panel E: Consumption - Data
Estimate SE
E [∆c] (%) 1.97 0.28σ [∆c] (%) 2.02 0.38AC1 [∆c] 0.48 0.12
The Term Structure of CDS Spreads and Sovereign Credit Risk 55
Tab
le7:
Def
ault
Par
amet
ers
and
CD
SP
rici
ng
Implica
tion
s(a
fter
July
2007
)
Th
ista
ble
rep
orts
the
calib
rate
dp
aram
eter
sof
the
def
ault
pro
cess
Θλ
=(φλσ,φλ,νλ,cλ)>
inco
lum
ns
(2)
to(5
)fo
rth
e44
cou
ntr
ies
inth
esa
mp
le.
Th
em
od
elis
sim
ula
ted
over
ati
me
seri
esof
120,
000
mon
ths.
Colu
mn
s(6
)to
(8)
rep
ort
the
pop
ula
tion
valu
esof
the
aver
age
5-y
ear
spre
ad
an
dth
eav
erag
esl
ope
(in
bp
s),
and
the
frac
tion
ofsi
mu
late
dob
serv
ati
on
sw
ith
an
inve
rted
term
stru
ctu
reof
CD
Ssp
read
s.C
olu
mn
s(9
)to
(11)
rep
ort
the
obse
rved
sam
ple
equiv
alen
tsfo
rth
ep
erio
dJu
ly20
07u
nti
lF
ebru
ary
2012.
Colu
mn
s(1
2)
to(1
3)
rep
ort
small
sam
ple
mom
ents
base
don
1,0
00
sim
ula
tion
sof
100
year
s(1
,200
mon
ths)
.C
olu
mn
(12)
rep
ort
sth
e5th
and
95th
per
centi
les
of
the
dis
trib
uti
on
of
the
5-y
ear
CD
Ssp
read
(in
bp
s),
wh
ile
colu
mn
(13)
rep
orts
the
5th
and
95th
per
centi
les
ofth
ed
istr
ibu
tion
of
the
10y-1
ysl
op
eof
spre
ad
s(i
nb
ps)
.C
olu
mn
(14)
rep
ort
sm
od
el-i
mp
lied
risk
pre
mia
for
1-ye
aran
d10
-yea
rC
DS
spre
ads
(in
%)
asa
fract
ion
of
tota
lsp
read
sb
ase
don
the
sim
ula
ted
valu
esof
120,0
00
month
sof
data
.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
Defa
ult
Param
eters
Sim
.in
Popula
tio
nD
ata
SS
CD
S5y
SS
Slo
pe
Risk
Prem
iaΘλ
(bps)
(%
)(bps)
(%
)(bps)
(bps)
(%
)C
ountr
yφλσ
φλ
νλ
cλ
CD
S5y
Slo
pe
Fra
c<
0C
DS5y
Slo
pe
Fra
c<
0[5
th,9
5th
][5
th,9
5th
][1
y,1
0y]
Aust
ria
Belg
ium
Bra
zil
Bulg
ari
aC
hile
Chin
aC
olo
mbia
Cro
ati
aC
ypru
sC
zech
Rep.
Denm
ark
Egypt
Fin
land
Fra
nce
Germ
any
Gre
ece
Hungary
Irela
nd
Isra
el
Italy
Japan
Kore
aL
ebanon
Lit
huania
Mala
ysi
aM
exic
oM
oro
cco
Panam
aP
eru
Philip
p.
Pola
nd
Port
ugal
Qata
rR
om
ania
Russ
iaSlo
vakia
Slo
venia
Sth
.A
fric
aSpain
Sw
eden
Thailand
Turk
ey
Uru
guay
Venezuela
10.4
211.9
226.0
530.2
411.8
113.3
320.9
821.5
90.0
115.6
24.8
339.0
43.6
34.9
111.9
30.1
936.9
90.7
216.8
422.9
112.4
814.1
735.1
738.8
614.5
023.0
123.4
725.3
823.4
527.4
017.7
00.0
414.3
235.5
629.5
127.8
433.9
926.3
446.2
04.0
521.6
722.0
234.8
891.1
6
0.2
244
0.3
025
0.0
234
0.3
367
0.2
422
0.1
346
0.2
987
0.5
220
0.4
490
0.0
278
0.4
102
0.3
050
0.3
329
0.5
343
0.0
003
0.9
794
0.2
720
0.7
787
0.1
970
0.1
740
0.0
001
0.4
119
0.5
274
0.1
598
0.2
458
0.1
192
0.1
894
0.0
968
0.1
549
0.2
351
0.2
757
0.8
325
0.2
390
0.3
207
0.2
793
0.0
083
0.0
177
0.1
352
0.0
263
0.4
173
0.0
375
0.4
605
0.2
120
0.5
196
4.1
8e-0
41.5
6e-0
35.1
1e-0
41.3
0e-0
21.4
9e-0
37.9
2e-0
42.3
6e-0
35.0
5e-0
25.3
8e-0
33.5
9e-0
23.6
0e-0
11.1
8e-0
22.4
2e-0
15.2
2e-0
13.4
6e-0
61.6
6e-0
15.4
9e-0
34.9
2e-0
31.9
9e-0
36.6
2e-0
41.2
8e-0
61.5
0e-0
31.7
4e-0
25.3
7e-0
21.2
0e-0
48.7
9e-0
42.4
9e-0
35.0
4e-0
35.2
4e-0
44.0
6e-0
33.3
6e-0
34.5
8e-0
36.6
3e-0
11.1
0e-0
21.4
6e-0
21.0
0e-0
61.0
3e-0
51.9
3e-0
33.5
3e-0
54.0
2e-0
12.0
8e-0
39.9
7e-0
38.1
4e-0
45.0
7e-0
1
2.2
1e-0
27.8
7e-0
38.0
0e-0
46.5
2e-0
43.8
1e-0
32.0
7e-0
33.5
6e-0
41.3
2e-0
44.9
6e-0
11.0
0e-0
61.0
0e-0
69.2
9e-0
61.0
0e-0
61.0
0e-0
68.9
8e-0
12.8
9e-0
31.0
5e-0
45.0
9e-0
12.7
9e-0
32.5
8e-0
22.5
3e-0
11.1
8e-0
31.1
6e-0
44.4
9e-0
35.9
8e-0
32.4
9e-0
35.7
7e-0
44.0
5e-0
42.5
5e-0
45.8
0e-0
58.2
4e-0
46.5
4e-0
11.0
0e-0
69.3
5e-0
41.0
2e-0
36.3
2e-0
19.2
0e-0
11.5
2e-0
46.8
2e-0
11.0
0e-0
61.5
5e-0
31.2
8e-0
41.0
1e-0
68.1
9e-0
4
69
89
139
238
81
80
155
235
257
84
43
292
28
55
33
1644
264
338
109
141
52
125
386
265
99
136
151
146
144
186
127
387
98
273
214
89
93
158
146
36
117
212
230
1056
12
15
24
38
14
14
26
38
-10
15 8
46 5
10 6
-493
42
-80
19
24 9
22
57
38
17
23
26
25
25
31
22
-105
17
43
35
16
16
27
25 7
20
35
38
103
2.4
75.4
20.2
31.0
64.0
80.6
50.4
81.0
43.0
00.1
40.1
00.6
00.0
90.1
10.1
035.7
20.5
63.3
41.7
83.0
30.1
14.9
91.2
61.2
34.4
80.4
70.3
70.2
40.2
70.3
30.8
93.2
40.1
51.3
10.9
50.2
70.2
90.2
70.5
00.1
00.2
10.7
50.4
06.3
3
78
100
154
256
93
90
173
253
257
92
48
315
32
63
38
1214
283
300
124
153
61
135
409
278
111
151
165
163
161
206
142
328
107
291
226
98
101
177
158
41
131
235
243
1096
38
37
104
76
67
58
109
76 -7 43
28
111
21
38
29
-933
73
-50
73
39
56
51
119
41
67
89
76
103
104
122
69
-76
56
77
61
44
43
98
36
26
76
133
125
101
0.0
00.0
00.0
08.9
30.0
00.0
00.0
07.1
417.8
63.5
70.0
00.0
00.0
00.0
00.0
046.4
38.9
339.2
90.0
07.1
40.0
01.7
90.0
026.7
90.0
00.0
00.0
00.0
00.0
07.1
40.0
039.2
90.0
010.7
112.5
01.7
90.0
07.1
40.0
00.0
00.0
08.9
30.0
025.0
0
[47,1
00]
[61,1
30]
[95,2
03]
[162,3
48]
[56,1
19]
[55,1
17]
[106,2
28]
[161,3
44]
[252,2
61]
[57,1
23]
[29,6
3]
[199,4
27]
[19,4
2]
[37,8
0]
[23,4
8]
[763,3
404]
[180,3
87]
[283,4
32]
[75,1
60]
[97,2
06]
[36,7
6]
[86,1
83]
[262,5
65]
[189,3
76]
[68,1
45]
[93,1
99]
[103,2
21]
[100,2
14]
[99,2
12]
[127,2
73]
[87,1
86]
[306,6
54]
[67,1
44]
[186,3
99]
[147,3
13]
[61,1
31]
[64,1
36]
[108,2
32]
[100,2
13]
[25,5
3]
[80,1
72]
[144,3
11]
[157,3
37]
[735,1
504]
[10,1
4]
[13,1
8]
[19,2
8]
[30,4
6]
[12,1
7]
[12,1
6]
[21,3
1]
[29,4
5]
[-30,1
][1
2,1
7]
[6,9
][3
5,5
5]
[4,6
][8
,11]
[5,7
][-
1222,-
17]
[32,5
1]
[-253,1
][1
5,2
2]
[19,2
8]
[8,1
1]
[17,2
5]
[42,7
1]
[30,4
6]
[14,2
0]
[19,2
7]
[20,3
0]
[20,2
9]
[20,2
9]
[24,3
7]
[18,2
6]
[-355,-
1]
[14,2
0]
[33,5
2]
[27,4
1]
[13,1
8]
[13,1
9]
[21,3
2]
[19,2
9]
[5,8
][1
6,2
4]
[27,4
1]
[29,4
5]
[60,1
41]
[2.7
9,1
8.1
6]
[2.7
4,1
7.9
9]
[2.9
0,1
8.1
7]
[2.7
3,1
7.5
4]
[2.7
9,1
8.2
2]
[2.8
6,1
8.3
7]
[2.7
7,1
7.9
9]
[2.5
5,1
7.4
0]
[0.7
5,0
.51]
[2.9
0,1
8.4
2]
[2.6
8,1
8.4
3]
[2.7
6,1
7.4
1]
[2.7
5,1
8.5
5]
[2.5
5,1
8.2
6]
[2.9
0,1
8.5
5]
[0.6
4,0
.98]
[2.7
8,1
7.5
4]
[0.7
6,0
.76]
[2.8
2,1
8.1
5]
[2.8
1,1
7.8
9]
[2.9
1,1
8.5
7]
[2.6
9,1
8.0
4]
[2.5
5,1
6.8
3]
[2.5
7,1
5.8
9]
[2.8
1,1
8.2
7]
[2.8
6,1
8.1
3]
[2.8
3,1
8.0
5]
[2.8
7,1
8.1
0]
[2.8
5,1
8.1
1]
[2.8
1,1
7.8
9]
[2.7
8,1
8.1
0]
[0.7
4,0
.54]
[2.8
0,1
8.2
8]
[2.7
4,1
7.3
9]
[2.7
6,1
7.6
0]
[2.9
1,1
8.4
0]
[2.8
9,1
8.2
7]
[2.8
6,1
8.0
5]
[2.8
7,1
7.9
4]
[2.6
8,1
8.4
5]
[2.8
9,1
8.2
2]
[2.6
3,1
7.6
4]
[2.8
2,1
7.7
1]
[2.3
6,1
3.2
6]
The Term Structure of CDS Spreads and Sovereign Credit Risk 56
Tab
le8:
Exce
ssB
ond
Ret
urn
san
dE
conom
icP
olic
yU
nce
rtai
nty
Col
um
ns
(2)
to(1
1)in
this
tab
lere
por
tth
eco
effici
ent
esti
mate
san
dR
2fr
om
the
pro
ject
ion
of
on
e-ye
ar
hold
ing
per
iod
exce
ssb
on
dre
turn
s(c
olu
mn
s(1
2)to
(16)
)on
the
Bak
eret
al.
(201
6)ec
onom
icp
olic
yu
nce
rtain
tyin
dex
.S
tati
stic
al
sign
ifica
nce
isb
ase
don
stan
dard
erro
rsth
at
acc
ou
nt
for
seri
al
corr
elat
ion
up
totw
elve
lags
asin
New
eyan
dW
est
(198
7).
Ies
tim
ate
the
regre
ssio
ns
for
five
matu
rity
ban
ds
an
d25
cou
ntr
ies.
Th
em
atu
rity
ban
ds
are
1-3
years
,3-
5ye
ars,
5-7
year
s,7-
10ye
ars
and
over
10ye
ars
.T
he
cou
ntr
ies
are
Au
stra
lia,
Au
stri
a,
Bel
giu
m,
Can
ad
a,
Den
mark
,F
inla
nd
,F
ran
ce,
Ger
man
y,G
reec
e,Ir
elan
d,
Ital
y,Jap
an,
Kor
ea,
Mal
aysi
a,
Mex
ico,
Net
her
lan
ds,
New
Zea
lan
d,
Norw
ay,
Pola
nd
,P
ort
ugal,
Sin
gap
ore
,S
pain
,S
wed
en,
Sw
itze
rlan
d,
and
the
U.K
.B
ond
retu
rns
are
com
pu
ted
bas
edon
the
Cit
igro
up
Worl
dG
over
nm
ent
Bon
dU
SD
Tota
lR
etu
rnIn
dic
es(f
orm
erS
alo
mon
Sm
ith
Bar
ney
Ind
ices
).T
he
retu
rns
are
inex
cess
ofth
esh
ort
-ter
mJP
Morg
an
Cash
on
e-m
onth
Tota
lR
etu
rnIn
dex
.T
he
earl
iest
start
ing
per
iod
for
the
dat
ais
Dec
emb
er19
85.
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
(13)
(14)
(15)
(16)
1-3
years
3-5
years
5-7
years
7-1
0years
10+
years
AverageExcess
Retu
rns
Cou
ntr
yβ
R2
βR
2β
R2
βR
2β
R2
1−
33−
55−
77−
10
+10
Au
stra
lia
0.1
1***
15.0
70.1
2***
16.1
70.1
2***
16.6
20.1
2***
16.5
80.1
2***
16.4
85.1
45.8
86.4
76.7
47.2
2A
ust
ria
0.0
9**
8.8
00.1
0***
9.7
70.1
0***
9.6
90.1
0***
9.1
20.1
0***
8.7
21.5
42.6
13.1
83.6
45.2
4B
elgiu
m0.1
0***
10.3
70.1
1***
10.8
10.1
1***
10.4
00.1
0***
9.4
90.1
0***
8.5
21.7
92.7
43.3
73.8
14.3
6C
an
ad
a0.0
4*
5.0
80.0
4*
5.0
00.0
4*
4.5
60.0
4*
3.7
70.0
42.2
93.1
43.9
74.6
54.9
75.9
8D
enm
ark
0.0
6*
4.3
40.0
6*
5.2
90.0
6**
5.3
00.0
6**
5.2
90.1
0***
9.2
42.9
63.8
64.5
55.1
66.6
2F
inla
nd
0.1
0***
11.1
10.1
1***
12.4
50.1
1***
11.6
80.1
***
10.7
80.0
7***
5.7
60.3
81.3
42.5
13.7
11.5
8F
ran
ce0.0
53.5
50.0
6*
4.6
90.0
6*
4.7
80.0
6**
4.7
50.0
6**
4.6
32.6
33.5
64.0
84.4
95.3
8G
erm
any
0.0
53.4
50.0
6*
3.8
60.0
6*
4.1
50.0
6*
3.6
20.1
0***
5.8
62.0
92.8
63.5
23.5
35.6
2G
reec
e0.0
5*
3.8
10.0
52.1
30.0
51.5
40.0
51.4
40.0
61.7
14.2
03.2
02.7
92.4
92.6
1Ir
elan
d0.0
7*
8.0
80.0
9***
11.3
50.0
7**
6.3
20.0
6*
3.3
20.0
7*
3.5
92.0
82.2
91.7
03.6
12.9
6It
aly
0.0
41.6
90.0
31.2
10.0
20.3
90.0
41.5
70.0
8**
3.4
33.3
03.8
02.8
72.8
56.4
6Jap
an
0.1
0***
11.0
20.1
1***
11.5
00.1
1***
10.9
50.1
0***
9.9
60.1
2**
11.2
00.7
81.7
82.5
73.1
16.2
0K
ore
a0.1
2**
12.6
80.1
3***
13.9
40.1
4***
15.0
20.1
4***
14.2
60.1
5***
14.9
10.7
10.9
91.3
81.5
41.5
8M
ala
ysi
a0.0
4*
6.1
60.0
4**
5.5
70.0
4*
4.5
80.0
32.0
20.0
10.3
64.4
85.0
95.5
25.6
06.7
9M
exic
o0.0
53.1
10.0
53.2
10.0
52.4
40.0
51.8
60.0
20.3
54.5
65.7
56.5
77.1
88.8
5N
eth
erla
nd
s0.0
63.7
70.0
6*
4.3
20.0
6*
4.5
90.0
6*
4.3
50.0
52.7
02.3
03.0
63.5
93.7
64.2
4N
ewZ
eala
nd
0.1
9***
27.8
00.1
9***
24.3
60.1
6***
22.3
70.1
8***
21.3
00.1
6***
18.8
76.1
34.2
33.2
25.4
06.1
5N
orw
ay
0.1
3***
17.8
70.1
3***
16.2
20.1
0***
11.9
60.1
2***
14.1
70.0
2***
8.5
71.5
22.6
72.9
43.9
1-2
.66
Pola
nd
-0.0
20.2
4-0
.01
0.0
6-0
.01
0.0
30.0
00.0
1-0
.05
1.5
49.6
310.4
65.6
711.0
65.5
9P
ort
ugal
0.0
55.0
70.0
53.7
30.0
52.4
90.0
52.2
30.0
74.6
54.3
34.0
43.7
53.6
50.5
7S
ingap
ore
0.0
21.8
40.0
22.0
30.0
10.5
2-0
.01
0.1
9-0
.01
0.1
72.3
44.0
35.0
55.3
54.6
6S
pain
0.0
53.2
10.0
53.3
40.0
7**
5.9
20.0
52.6
50.0
30.7
11.6
82.5
83.4
53.7
22.9
8S
wed
en0.0
63.1
60.0
63.0
30.0
52.5
70.0
51.9
00.0
41.4
41.6
32.5
93.3
74.2
25.0
1S
wit
zerl
an
d0.0
41.9
60.0
52.3
60.0
52.2
10.0
52.5
20.0
63.0
51.4
52.0
42.4
62.8
93.4
4U
.K.
0.0
20.7
10.0
21.0
00.0
31.0
80.0
31.1
20.0
30.9
02.7
73.3
73.8
44.0
64.3
3
Pan
el0.0
6***
5.1
50.0
7***
5.5
30.0
7***
5.1
50.0
7***
4.5
50.0
6***
3.8
2
Aver
age
0.0
77.1
30.0
77.1
60.0
76.6
40.0
76.0
50.0
65.5
92.9
43.5
53.7
24.4
24.4
7