11、Do Rare Events Explain CDX Tranche Spreads（2018JF）

THE JOURNAL OF FINANCE • VOL. LXXIII, NO.
5 • OCTOBER 2018
Do Rare Events Explain CDX Tranche Spreads?

SANG BYUNG SEO and JESSICA A. WACHTER∗
ABSTRACT
We investigate whether a model with time-varying probability of economic disaster
can explain prices of collateralized debt obligations. We focus on senior tranches of the
CDX, an index of credit default swaps on investment grade firms. These assets do not
incur losses until a large fraction of previously stable firms default, and thus are deep
out-of-the money put options on the overall economy. When calibrated to consumption
data and to the equity premium, the model explains the spreads on CDX tranches
prior to and during the 2008 to 2009 crisis.
The period from 2005 to September of 2008 witnessed a more than 100-fold
increase in the cost of insuring against economic catastrophe. This cost is
apparent in the pricing of derivative contracts written on the CDX, an index
of credit default swaps on investment-grade firms. During the 2005 to 2009
period, tranches on the CDX were actively traded, with investors purchasing
and selling insurance that would pay off only if a certain fraction of firms
represented by the CDX went into default. The most senior tranches were
structured to pay off only if corporate defaults became extremely widespread—
more so than during the Great Depression. These senior tranches, which can be
viewed as extremely deep out-of-the-money options, are nonredundant assets
whose prices uniquely reflect the probability that investors place on an extreme
state of the world.
While the costs of insuring senior tranches on the CDX were close to zero
through most of 2006 and 2007, fluctuations began to appear in late 2007,
culminating in sharply rising prices in the summer and fall of 2008. Ex post,
such insurance did not pay off—only a small number of firms represented by
the CDX index went into default. Yet the pricing of these securities suggests a
substantial, time-varying fear of economic catastrophe.
∗ Sang Byung Seo is at the C.T. Bauer College of Business, University of Houston. Jessica A.
Wachter is at the Wharton School, University of Pennsylvania. We thank Pierre Collin-Dufresne;

Hitesh Doshi; Kris Jacobs; Mete Kilic; Stefan Nagel; Nick Roussanov; Christian Schlag; Ivan
Shaliastovich; Pietro Veronesi; Amir Yaron; and seminar participants at AFA annual meetings,
the Midwest Finance Association, Baruch College, the University of Houston, and the Wharton
School for helpful comments. We thank the Rodney L. White Center for research support. The
authors do not have any potential conflicts of interest to disclose as identified in the Journal of
Finance’s disclosure policy.
DOI: 10.1111/jofi.12705
2343
2344 The Journal of FinanceR
The CDX and its tranches are an example of the structured finance prod-
ucts that proliferated in the period prior to the 2008–2009 financial crisis.1 In
the years since, both academic literature and the popular press have deeply
implicated structured finance in the series of events beginning with the near-
default of Bear Stearns in March 2008 and culminating in the collapse of
Lehman Brothers later that year (Reinhart and Rogoff (2009), Gorton and
Metrick (2012)).2 Yet, despite the centrality of structured products to the crisis
and its aftermath, there have been few attempts to quantitatively model these
securities in a way that connects them to the underlying economy.
In this paper, we investigate whether an equilibrium model with rare eco-
nomic disasters in the spirit of Barro (2006) and Rietz (1988) explains the
striking behavior of senior tranches on the CDX before and during the crisis.
We start with a model for investor preferences and for aggregate consumption.
These are calibrated to aggregate consumption in the data, to the equity pre-
mium, and to stock market volatility. We calibrate cash flows on firms to match
the low default rates in the data. Preferences, consumption, and firm cash flows
are thus the basic building blocks of the model.
Based on these microfoundations, we build a model for the CDX from the
ground up. From cash flows and the pricing kernel, we solve for firm valua-
tions in equilibrium. A firm defaults when its value falls below a prespecified
boundary.3 Firm values and default determine losses on credit default swaps,
which in turn determine losses on the CDX. Given a process for losses on the
CDX, we obtain payoffs on CDX tranches and therefore equilibrium prices. Our
resulting model can account for tranche prices, and in particular the prices of
senior tranches, both before and during the financial crisis.
Our findings relate to a recent debate concerning the pricing of CDX tranches.
Coval, Jurek, and Stafford (2009) examine the precrisis behavior of CDX
tranches, pricing these tranches with a static options model that assumes cash
flows occur at a fixed maturity.4 Relative to the options data, and assuming the
CDX itself is correctly priced, they find that senior spreads were too low in the
precrisis period.
Collin-Dufresne, Goldstein, and Yang (2012) question these conclusions. They
note that the pricing of the equity tranche (the most junior tranche) is sensitive
to the timing of defaults and the specification of idiosyncratic risk. Coval, Jurek,
and Stafford (2009) assume that default occurs at the five-year horizon, which
1 See Longstaff and Rajan (2008) for a description of structured finance products.
2 Salmon, Felix, Recipe for disaster: The formula that killed Wall Street, Wired Magazine,
February 23, 2009.
3 One strand of the credit derivative literature models default as an exogenous event that is the
outcome of a Poisson process (Duffie and Singleton (1997)). Such models are known as “reduced
form.” A second strand, which we build on, assumes that default occurs when firm value passes
through a lower boundary. Such models are known as “structural.” Structural models typically
take the price process and the risk-neutral measure as exogenous. In our paper, both are derived
from investor preferences and beliefs.
4 Specifically, if all cash flows occur at year five, then in principle CDX prices can be calculated
using state prices derived from five-year options.

Do Rare Events Explain CDX Tranche Spreads? 2345
need not be the case.5 When payoffs occur at a horizon other than five years,
the static method of extracting state prices from options no longer applies.
Collin-Dufresne, Goldstein, and Yang (2012) therefore specify a dynamic model
of the pricing kernel. They find that this model comes closer to matching tranche
spreads prior to the crisis.
However, the results of Collin-Dufresne, Goldstein, and Yang (2012) point to
the limitations in both papers. They find that the dynamic model can only ex-
plain tranche spreads during the crisis if one incorporates some probability of
catastrophic economy-wide losses. But if the probability of an economic catas-
trophe is nonzero, the available options do not complete the market: Options
that are sufficiently deep out of the money do not exist.6 In such an economy,
CDX senior tranches are nonredundant and no-arbitrage alone cannot deter-
mine their prices.
It is for this reason that we turn to an equilibrium model. Our model gives
risk-neutral probabilities of an economic catastrophe based on assumptions for
preferences and aggregate consumption. These are calibrated based on prior
studies without regard to the CDX. The model explains senior tranches on
the CDX by the same mechanism that allows it to explain the high equity
premium. A severe economic disaster raises the probability that multiple firms
go into default. This is precisely the event that affects the payoff on CDX senior
tranches. At the same time, this catastrophe affects investors by reducing their
consumption and raising their marginal utility. The price that investors are
willing to pay to insure against such an event is consistent with the premium
they demand to hold the aggregate stock market. This connection between the
equity premium and CDX senior tranches is tight: When we recalibrate the
economy to a lower equity premium, we find that implied senior tranche levels
fall far short of their levels in the data.
Like any complete-markets equilibrium model, our model offers implications
for virtually any asset class given assumptions on its cash flows. We investigate
implications for index options and for the term structure of government bonds,
as well as for the aggregate market, the risk-free rate, and of course the CDX
and its tranches. Because of the link between CDX tranches and options, we pay
particular attention to index option prices, using one-month implied volatilities
to determine, through the lens of the model, the time series of the latent state
variables. We find that while the model captures average implied volatilities,
it cannot jointly account for the low autocorrelation in implied volatilities on
index options and the high autocorrelation of the price-dividend ratio on the
5 Assuming that defaults occur at the five-year horizon makes the junior tranches look more
attractive. The model spreads will thus be artificially low on the junior tranches, and, because
the model is calibrated to match the index, the model-implied spreads on the senior tranches will
be too high. Collin-Dufresne, Goldstein, and Yang (2012) emphasize the need for fat tails in the
idiosyncratic risk of firms to capture the CDX spread at the three-year, as well as the five-year,
maturity. Introducing fat-tailed risk also raises the spread of junior tranches in the model and
lowers the spread of senior tranches.
6 Collin-Dufresne, Goldstein, and Yang (2012) treat the risk-neutral probability of a disaster as
a free parameter that undergoes an unanticipated and permanent shift.

aggregate market. Moreover, while the model can match the behavior of short-
term government bills and Treasury Inflation Protected Securities (TIPS) yields
at the five-year maturity, it predicts yields on longer term real bonds that are
counterfactually low. These aspects of the data represent important challenges
for future work.
Our results show that it is possible to account for the prices of senior tranches
on the CDX in a model without explicit frictions. We thus show that at least
some of the pricing behavior that was attributed to market failures during
the crisis can be explained using the benchmark framework of representative-
agent asset pricing. Our findings also support the view that beliefs about rare
disasters are an important determinant of stock market behavior. That said,
we do not explain the cause of rare disasters in consumption; rather, our model
derives implications of a nontrivial probability of such events. Such disasters
might have various causes, including a widespread failure of the financial
system.
The rest of this paper proceeds as follows. Section I describes the model,
Section II describes the data, and Section III describes the calibration of the
model. In Section IV, we evaluate the model using data on the CDX and
CDX tranches. In Section V, we discuss the fit of the model to other asset
classes, namely, long-term government bonds, equity prices, and option prices.
Section VI concludes.
I. Model
A. Model Primitives and the State-Price Density
We assume an endowment economy with complete markets and an infinitely
lived representative agent. Aggregate consumption (the endowment) solves the
following stochastic differential equation (SDE):
dCt
= μc dt + σc dBc,t + (e Zc,t − 1)dNc,t , (1)
Ct −
where Bc,t is a standard Brownian motion and Nc,t is a Poisson process. The
intensity of Nc,t is given by λt and assumed to be governed by the following
system of equations:

dλt = κλ (ξt − λt )dt + σλ λt dBλ,t (2a)

dξt = κξ (ξ̄ − ξt )dt + σξ ξt dBξ,t , (2b)
where Bλ,t and Bξ,t are Brownian motions (independent of each other and of
Bc,t ).
The process defined by (1) has both normal-times risk, as represented by the
Brownian component σc dBc,t , and a risk of rare disasters, as represented by
the Poisson term (e Zc,t − 1)dNc,t . Thus at time t, the economy will undergo a
disaster with probability λt .7 Given a disaster, the change in consumption (as a

fraction of the total) is e Zc,t − 1, where Zc,t < 0 is a random variable. By writing
the change in consumption as an exponential, we ensure that consumption
itself remains positive. We assume for simplicity that the distribution of Zc,t is
time invariant.
The system of equations ((2a)) implies that the probability of a disaster λt is
time varying and that it mean-reverts to a value ξt that itself changes over time.
This type of multifrequency process has often been used for modeling asset
price volatility and for option pricing in reduced-form models (see Duffie, Pan,
and Singleton (2000) for discussion and references). Two-factor multifrequency
processes are also used in the CDX literature (Collin-Dufresne, Goldstein, and
Yang (2012)). Because return volatility will inherit the multifrequency vari-
ation of λt , (2a) is a natural choice for the disaster probability process. The
process (2) can capture long memory in a time series, that is, autocorrelations
that decay at a slower-than-geometric rate. For example, the 2008 financial
crisis was characterized by both a spike in λt that decayed quickly and higher
disaster probabilities in subsequent years. The system of equations ((2a)) cap-
tures this feature of the data, while a univariate autoregressive process would
not. Setting σξ to zero and assuming that ξt is at its (then-deterministic) steady
state of ξ̄ results in the one-factor model of Wachter (2013).
We assume a recursive generalization of time-additive power utility that
allows for preferences over the timing of the resolution of uncertainty. Our
formulation comes from Duffie and Epstein (1992), and we consider the special
case in which the elasticity of intertemporal substitution (EIS) is equal to 1.
Specifically, we define continuation utility Vt for the representative agent using
the recursion
∞
Vt = Et f (Cs , Vs ) ds,
t
where

1
f (Ct , Vt ) = β(1 − γ )Vt log Ct − log((1 − γ )Vt ) .
1−γ
The parameter β is the rate of time preference and γ is relative risk aversion.
This utility function is equivalent to the continuous time limit of the utility
function defined by Epstein and Zin (1989) and Weil (1990). Assuming an EIS of
one allows for closed-form solutions for equity prices up to ordinary differential
equations (ODEs), and facilitates the computation of options and CDX/CDX
tranche prices.
7 This description of Poisson shocks, which we adopt throughout the text, is approximate. In
any finite interval, there could theoretically be more than one shock since λt is an intensity, not a
probability.
In Section I of the Internet Appendix,8 we show that the pricing kernel is

characterized by the process
dπt
= −rt − λt E e−γ Zc,t − 1 dt
πt−

−γ σc dBc,t + bλ σλ λt dBλ,t + bξ σξ ξt dBξ,t + (e−γ Zc,t − 1)dNc,t , (3)
with
2
κλ + β κλ + β E e(1−γ )Zc,t − 1
bλ = − −2 , (4)
σλ2 σλ2 σλ2
2
κξ + β κξ + β bλ κλ
bξ = − −2 2 . (5)
σξ
2
σξ
2
σξ
In the special case of time-additive utility, γ = 1 and bλ = bξ = 0. The only risk

that matters for computing expected returns is consumption risk, given by the
terms −γ σc dBc,t and (e−γ Zc,t − 1)dNc,t , the latter of which captures the effect
of rare disasters. For γ > 1 (which implies a preference for early resolution of
uncertainty), bλ and bξ are positive. Assets that increase in value when λt and
ξt rise provide a hedge against disaster risk. All else equal, these assets will
have lower expected returns, and higher prices, than otherwise.
The risk-free rate is given by

rt = β + μc − γ σc2 + λt E e(1−γ )Zc,t − e−γ Zc,t . (6)
Equation (6) implies that the risk-free rate is decreasing in the probability of
an economic disaster. The greater is this probability, the more investors want
to save for the future and the lower the risk-free rate must be in equilibrium.
B. The Aggregate Market

φ
We assume that the aggregate market has payoff Dt = Ct (Abel (1990),
Campbell (2003)). Empirically, dividends are more variable than consumption
and more sensitive to economic disasters (Longstaff and Piazzesi (2004)). We
capture this fact by setting φ > 1. The process for dividends then follows from
Ito’s Lemma:
dDt
= μddt + φσc dBc,t + (eφ Zc,t − 1)dNc,t ,
Dt−
where μd = φμc + 12 φ(1 − φ)σc2 .
8 The Internet Appendix is available in the online version of the article on The Journal of Finance
website.
In equilibrium, the price of the dividend claim is determined by the cash

flows and the pricing kernel:
∞
πs
F(Dt , λt , ξt ) = Et Ds ds . (7)
t πt
Let G(λt , ξt ) be the price-dividend ratio:
∞
πs Ds
G(λt , ξt ) = Et ds
t πt Dt
∞

= exp aφ (τ ) + bφλ (τ )λt + bφξ (τ )ξt dτ, (8)
0
where aφ (τ ), bφλ (τ ), and bφξ (τ ) satisfy ODEs given in Section II of the

Internet Appendix. Under the reasonable assumption of φ > 1, bφλ (τ ) < 0
(Wachter (2013)). Further, using the reasoning of Tsai and Wachter (2015),
it can be shown that bφξ (τ ) < 0. The value of the aggregate market is thus
decreasing in the disaster probability λt and its time-varying mean ξt .
Figure IA.1 in the Internet Appendix depicts the functions bφξ (τ ) and bφλ (τ ),
given the parameter values discussed in Section III. Both functions are negative
and decreasing as a function of τ . The function bφλ (τ ) converges after about 20
years, which reflects the relative lack of persistence in λt . In contrast, the
function bφξ (τ ) takes nearly 70 years to converge. For dividend claims with
maturities of 10 years or less, bφλ (τ ) is greater in magnitude than bφξ (τ ), that
is, λt -risk is more important. However, in the limit as the horizon approaches
infinity, the effect of ξt is nearly three times as large as the effect of λt .9
Applying Ito’s Lemma to Ft = Dt G(λt , ξt ) gives the equilibrium law of motion
for the aggregate market:
dFt ∂G 1 ∂G 1
= μ F,t dt + φσc dBc,t + σλ λt dBλ,t + σξ ξt dBξ,t
Ft − ∂λ G ∂ξ G

+ eφ Zc,t − 1 dNc,t . (9)
The terms φσ dBc,t and (eφ Zc,t − 1)dNc,t represent normal-times and disaster-
times variation in dividends, respectively.10 At our parameter values, the latter
is a much more important source of risk than the former. Variation in λt and ξt
produces variation in the price-dividend
√ ratio G and √ thus in stock prices. This
is reflected in the terms ∂G 1
σ
∂λ G λ
λ t dBλ,t and ∂G 1
σ
∂ξ G ξ
ξt dBξ,t . These can lead to
9 See Lettau and Wachter (2007) and Borovička et al. (2011) for discussion of fixed-maturity
dividend claims and the effect of exposures to risks of varying frequencies.

10 The drift rate μ
F,t is determined in equilibrium from the instantaneous expected equity return
rte ,
D
μ F,t = rte − λt E eφ Zc,t − 1 +
t
,
Ft
where rte is determined by (10).
highly volatile stock prices, even during normal times. Combining equations
for the pricing kernel and for the aggregate market leads to the equation for
the equity premium (Tsai and Wachter (2015)):

rte − rt = γ φσ 2 − λt Et e−γ Zc,t − 1 e−φ Zc,t − 1
1 ∂G 1 ∂G
−λt bλ σλ − ξt bξ σξ . (10)
G ∂λ G ∂ξ
The first term, negligible in our calibration, represents the consumption capital
asset pricing model (Consumption CAPM). The next term is the risk premium
due to disasters itself, and is positive and large. This term represents the
comovement of marginal utility and firm value during disaster times. The last
two terms arise from time variation in the risk of a disaster, due to both changes
in λt and changes in ξt . Because disasters increase marginal utility (bλ and bξ
are positive) and decrease prices, these terms are positive. A calibration, like
ours, that matches aggregate stock market volatility also implies that they
have a significant impact on the equity premium.
Our analysis below relies on the prices of index options. We therefore briefly
describe option pricing in our setting. A European put option gives the holder
the right to sell the underlying security at some expiration date T for an
exercise price K. Because the payoff on the option is (K − FT )+ , no-arbitrage
implies that

πT +
P(Ft , λt , ξt , T − t; K) = Et ( K − FT ) .
πt
Let Kn = K/Ft denote the normalized strike price (“moneyness”) and Ptn =
Pt /Ft the normalized put price. Like the price-dividend ratio, the normalized
put price is a function of λt and ξt alone:

πT FT +
P (λt , ξt , T − t; K ) = Et
n n
K −
n
. (11)
πt Ft
We calculate normalized put prices from (11) and implied volatilities as in

Black and Scholes (1973) (see Seo and Wachter (2018) for details). As we show
in Section VII of the Internet Appendix, the transform analysis of Duffie, Pan,
and Singleton (2000) allows us to compute (11) analytically, which avoids the
need for extensive simulations.
C. Individual Firm Dynamics

CDX and CDX tranche pricing requires a model for individual firms. Let Di,t
be the payout amount of firm i, for i = 1, . . . , N f , where N f is the number of
firms in the CDX (the number has been 125). While we use the notation Di,t ,
we intend this to refer to the payout to not only the equity holders but to the
bondholders as well. The firm payout is subject to three types of risk:
dDi,t
= μi dt + φi σc dBc,t + (eφi Zc,t − 1)dNc,t + Ii,t (e ZSi − 1)dNSi ,t + (e Zi − 1)dNi,t ,
Di,t−
aggregate risk sector risk idiosyncratic risk
(12)
where μi is defined similarly to μd, that is, μi = φi μc + 12 φi (1 − φi )σc2 . The sys-

tematic risk is standard: Dit has a multiplicative component that behaves like
φ
Ct i , analogously to dividends. Firms are exposed to both normal-times aggre-
gate risk and aggregate consumption disasters. Because the total payout is
less sensitive to aggregate shocks than is the equity payout, φi < φ. However,
we will still allow firms to have greater exposure to aggregate disasters than
consumption, that is φi > 1 (labor income could account for the wedge between
unlevered cash flows and consumption).
Firms are also exposed to idiosyncratic negative events that occur with con-
stant probability λi . For simplicity, we assume that all idiosyncratic risk is Pois-
son.11 When a firm is hit by its idiosyncratic shock (which we model as an incre-
ment to the counting process Ni,t ), the firm’s payout falls by Di,t− × (1 − e Zi ). For
parsimony, we assume that μi , σi , φi , Zi , and λi are the same for all i and that Zi
is a single value (rather than a distribution). The shocks themselves, dNi,t , are
of course independent of one another and independent of the aggregate shock
dNc,t .
Longstaff and Rajan (2008) estimate that a portion of the CDX spread is
attributable to risk that affects a nontrivial subset of firms. Following Longstaff
and Rajan (2008) and Duffie and Garleanu (2001), we refer to this as sector
risk. Let S denote a finite set of sectors.12 Each firm is in exactly one sector;
we let Si ∈ S denote the sector for firm i and dNSi ,t the sector shock. When a
sector shock arrives, the firm is hit with probability pi , that is the sector term
in (12) is multiplied by Ii,t , which takes a value of 1 with probability pi and
0 otherwise. If a firm happens to be affected by this sector shock, the firm’s
payout drops by Di,t− × (1 − e ZSi ).13 Again, for parsimony, pi and ZSi are the
same across firms. The shocks Ii,t are independent across firms and dNSi ,t are
independent across sectors.
Intuitively, sector risk should be correlated with aggregate consumption risk.
To capture this correlation, we allow the intensity of NSi ,t , λ Si ,t to depend on the
state variables λt and ξt . We solve for firm values under general specification
11 Campbell and Taksler (2003) show that idiosyncratic risk and the probability of firm default
are strongly linked in the data.
12 For concreteness, we can think of the sector classification as corresponding to that given by
our data provider Markit. There are five sectors: Consumer, Energy, Financials, Industrial, and
Telecom, Media, and Technology, so S = {C, E, F, I, T}. We use this classification to discipline our
calibration. However, neither the equations nor our empirical results require this interpretation.
13 This structure is isomorphic to one in which the set of firms is partitioned into a greater
number of sectors and firms are hit by sector shocks with probability 1.
λ Si ,t = w0 + wλ λt + wξ ξt . For parsimony, we will calibrate the simpler model

λ Si ,t = wξ ξt .14
Given this payout definition, we solve for the total value of firm i (equity plus
debt), which we denote by Ai (Di,t , λt , ξt ):
∞
πs
Ai (Di,t , λt , ξt ) = Et Di,s ds . (13)
t πt
Define the price-to-payout ratio as
Ai (Di,t , λt , ξt )
Gi (λt , ξt ) ≡ .
Di,t
Similar to the price-dividend ratio, the price-to-payout ratio of an individual
firm can be expressed as an integral of an exponential-linear function of the
state variables,
∞

Gi (λt , ξt ) = exp ai (τ ) + biλ (τ )λt + biξ (τ )ξt dτ, (14)
0
where ai (τ ), biλ (τ ), and biξ (τ ) solve a system of ODEs (see Section III of the
Internet Appendix). Like aggregate market value, firm values decrease as a
function of the disaster probability λt and of its time-varying mean ξt .
The dynamics of firm values Ait = Ai (Di,t , λt , ξt ) follow from Ito’s Lemma:
dAi,t ∂Gi 1 ∂Gi 1
= μ Ai ,t dt + φi σc dBc,t + σλ λt dBλ,t + σξ ξt dBξ,t
Ai,t − ∂λ Gi ∂ξ Gi

+ eφi Zc,t − 1 dNc,t + Ii,t (e ZSi − 1)dNSi ,t + (e Zi − 1)dNi,t , (15)
where μ Ai ,t is the asset drift rate, which is determined in equilibrium.

Equation (15) has some similarity to processes that use reduced-form models
for asset prices. As in Collin-Dufresne, Goldstein, and Yang (2012), there is a
Brownian term with stochastic volatility (following a multifrequency process),
a risk of an adverse idiosyncratic event, and a risk of catastrophic market-wide
decline. Here, however, the process is an endogenous outcome of our assump-
tions on fundamentals and on the utility function. Specifically, volatility occurs
because of changes in agents’ rational forecasts of economic disasters.
D. CDX Pricing
A credit default swap contract provides a means to trade on the risk of a sin-
gle firm’s default. Under this bilateral contract, the protection buyer commits
to paying an insurance premium to the protection seller, who pays the protec-
tion buyer the loss amount in the case of default. Recent statistical models of
single-name credit default swaps suggest that market participants price in the
14 This is a shortcut. Ideally one would model sector shocks as having some impact on aggregate
consumption.
risk of rare idiosyncratic and market-wide events (Kelly, Manzo, and Palhares
(2016), Seo (2017)). Our focus in this paper is on the CDX North American
Investment Grade, an index whose value is determined by default events on
a set of underlying firms, also known as reference entities. An investor who
buys protection on the CDX buys some protection on default events of all the
underlying firms.
At each time t, we price a CDX contract initiated at t and maturing at T on
N f reference entities whose asset prices are governed by (15). Define default
as the event whereby a firm’s value falls below a threshold AB (Black and Cox
(1976)). The default time for firm i is therefore

Ai (Di,τ , λτ , ξτ )
τt,i = inf τ > t : ≤ AB . (16)
Ai (Di,t , λt , ξt )
To maintain stationarity, we define the threshold relative to the value of the firm
at initiation.15 This formulation allows the distance to the default boundary to
vary over the life of the contract. It also has the effect of keeping this distance
invariant across CDX series, that is, CDX series closer to the crisis do not have
reference entities that are closer to default. Given that Markit attempts to
keep the credit worthiness of firms constant as it issues new CDX series, the
assumption of a constant distance to default seems reasonable.
Let Rτt,i denote the recovery rate for a firm defaulting at τt,i . This recovery
rate is a random variable that depends only on the outcome of dNc,τt,i , that is,
whether default co-occurs with a disaster. Note that this specification implies
that the time-t distribution of both τt,i − t and of Rτt,i is determined completely
by λt and ξt .
Below we discuss contracts on the CDX as a whole; in Section I.E we consider
tranches. Following the convention in our data, we assume that the protection
buyer pays to insure $1 (i.e., $1 is the notional). If firm i defaults, the loss on
the CDX increases by N1f (1 − Rτt,i ). Let Lt,s denote the cumulative loss at time
s. Then Lt,s is given by
1
Nf
Lt,s = 1{t<τt,i ≤s} (1 − Rτt,i ). (17)
Nf
i=1
Increases in Lt,s trigger payments from the protection seller (the party provid-
ing insurance) to the protection buyer. No-arbitrage implies that the value of
these payments equals

T s
ProtCDX (λt , ξt ; T − t) = EtQ e− t ru du
dLt,s ,
t
15 If we interpret A A as the amount of debt that the firm has at time t, then (16) implies that
B i,t
the firm is in default whenever the value of equity is below 0. This formulation is consistent with
stationary leverage ratios and slow reversion to long-run means (Lemmon, Roberts, and Zender
(2008)).
where E Q denotes the expectation taken under the risk-neutral measure Q and
ru is the risk-free rate, both of which are implied by the pricing kernel (3).16
Equation (18) is sometimes referred to as the “protection leg” of the contract.
In a CDX contract, the protection buyer makes payments at quarterly inter-
vals. If no firms default, these premium payments add up to a fixed value S
over the course of a year. If default occurs, the premium payments fall to reflect
the fact that a lower amount is now insured under the contract. Let nt,s denote
the fraction of firms that have defaulted s − t years into the contract:
1
Nf
nt,s = 1{t<τt,i ≤s} . (18)
Nf
i=1
Like Lt,s , nt,s is a random variable whose value is realized at s and whose time-t
distribution depends only on λt , ξt , and s − t. Let ϑ = 1/4, the interval between
premium payment dates. For a given spread S, the premium leg is equal to
4T
Q 1
t+ϑm
PremCDX (λt , ξt ; T − t, S) = SEt e− t rudu(1 − nt,t+ϑm)
4
m=1
t+ϑm s

− t ru du
+ e (s − t − ϑ(m − 1))dnt,s . (19)
t+ϑ(m−1)
The first term in (19) is the value of the scheduled premium payments. Note that
nt,t+ϑm is the fraction of the pool that has defaulted as of the mth payment, and so
1 − nt,t+ϑm ≤ 1 is the notional at that point in time. The second term represents
the accrued premium. Consider a firm default at some time s between payments
m − 1 and m. Until time s, the protection buyer is insured against default of
this firm. However, this is not reflected in the scheduled payments—the value
of the mth payment is the same as it would be if the default occurred at time
t + ϑ(m − 1) < s. Thus, upon default the protection buyer pays the fraction of
S/(4N f ) that accrues between the (m − 1)th payment and the default time s.
This quantity is known as the accrued premium.
The CDX spread SCDX (λt , ξt ; T − t) is the value of S that equates the premium
leg (19) with the protection leg (18). We describe the computation of this spread
in Section IV of the Internet Appendix.
E. CDX Tranche Pricing

CDX tranches are derivative contracts written on the CDX. They partition
the total loss on the index (in a sense that will be clear below), and they have
different levels of subordination. Note that the CDX is a synthetic collateralized
debt obligation—it is not a claim to underlying physical assets. Payoffs on the
tranches are defined in terms of the CDX loss Lt,s .
16 Following longstanding practice in the literature on credit derivatives, we express prices as
discounted cash flows under the risk-neutral measure rather than as cash flows multiplied by πt
under the physical measure. The mapping between the two is well known (Duffie (2001)).
Tranches are defined by two numbers: the attachment point, which gives the
level of CDX loss at which the tranche is penetrated, and the detachment point,
after which further losses detach from the tranche. For example, consider the
10% to 15% tranche. This tranche loses value if the CDX accumulates more
than a 10% loss. Losses of between 10% and 15% attach to this tranche. If
losses reach 15%, the notional amount of the tranche is exhausted. Further
losses attach to the next tranche. During our sample period, six tranches were
commonly traded, with attachment-detachment pairs 0% to 3%, 3% to 7%, 7%
to 10%, 10% to 15%, 15% to 30%, and 30% to 100%. The most junior tranche
(0% to 3%) is referred to as equity, the second as mezzanine, and the remaining
four tranches as senior. The most senior tranche (30% to 100%), is often called
“super senior.”
Let K j−1 be the attachment point and K j the detachment point of the jth
tranche for j = 1, . . . , J, where K0 = 0 and KJ = 1. Given a CDX loss Lt,s , the
tranche loss is given by
min{Lt,s , K j } − min{Lt,s , K j−1 }
L
T j,t,s = T jL(Lt,s ) = . (20)
K j − K j−1
If the CDX loss is below both K j and K j−1 , it has not attached to the tranche
and the loss is 0%. If the loss is greater than both K j and K j−1 , it detaches from
the tranche and the tranche loss is 100%. If the loss is between K j−1 and K j ,
L −K j−1
then the loss equals 0 < Kt,sj −K j−1 < 1. Definition (20) implies that the notional
amount on each tranche is equal to $1, which is the convention in our data.
Note that the weighted sum of tranche losses equals the total loss:

J
(K j − K j−1 )T j,t,s
L
= Lt,s . (21)
j=1
Given the specification for the tranche loss, the protection seller for tranche j
pays

T s
ProtTran, j (λt , ξt , T − t) = EtQ e− t ru du L
dT j,t,s . (22)
t
The protection buyer for tranche j makes quarterly premium payments. As in

the case of the CDX, the amount he or she pays depends on the tranche notional.
However, adjusting the tranche notional for default is more complicated than
for the CDX. The adjustment in notional depends not only on the tranche loss,
but also on something called tranche recovery. If a firm defaults, the notional
on the CDX falls by N1f . However, the notional on the most junior tranche
falls by a smaller amount, N1f (1 − Rτt,i ).17 For the notional amount on the CDX
tranches to be consistent with that of the CDX, the total change in notional on
17 To be precise, this is the change in notional multiplied by the width of the tranche, K j − K j−1 .
the tranches following default must also add up to N1f . This extra reduction in
notional is the tranche recovery.
It is customary to apply the tranche recovery to the most senior tranche. Note
that nt,s − Lt,s is the amount recovered to date from defaults. Then tranche
recovery for the super senior tranche is defined as
nt,s − Lt,s
R
T J,s =
KJ − KJ−1
(recall that KJ = 1). In the very rare event that this recovery exhausts the
notional on the senior tranche, the remaining recovery amount detaches from
this tranche and attaches to the next most senior tranche. A general definition
for tranche recovery is thus
min{nt,s − Lt,s , 1 − K j−1 } − min{nt,s − Lt,s , 1 − K j }

R
T j,t,s = T jR(nt,s − Lt,s ) = . (23)
K j − K j−1
Note that if nt,s − Lt,s < 1 − K j , then no recovery applies to tranche j (it can
all be applied to the more senior tranches). It follows that nt,s − Lt,s < 1 − K j−1
as well, and (23) is equal to 0. If nt,s − Lt,s > 1 − K j and nt,s − Lt,s < 1 − K j−1 ,
then the numerator in (23) equals nt,s − Lt,s − (1 − K j ), which is the amount
of the recovery not reflected in the tranche loss for more senior tranches. If
nt,s − Lt,s is above both 1 − K j and 1 − K j−1 , then (23) is equal to 1. As is the
case for tranche losses (24), the weighted sum of tranche recovery is equal to
total recovery:

J
(K j − K j−1 )T j,t,s
R
= nt,s − Lt,s . (24)
j=1
Combining (21) and (24), we see that the weighted change in notional from a
default is nt,s , which is the change in notional for a contract on the CDX itself.
Given these definitions of tranche loss and recovery, we define the premium
payments for a given tranche as follows. Let S be the spread, and U the upfront
payment (to be discussed further below). Then the premium leg for tranche j
is given by
PremTran, j (λt , ξt ; T − t, U , S)
4T t+ϑm
1 t+ϑm
= U + SE Q e− t rudu 1 − T j,t,s
L
− T j,t,s
R
ds (25)
4 t+ϑ(m−1)
m=1
(the definitions of tranche loss and recovery imply that their sum cannot ex-
ceed 1). Except for the upfront payment U , the tranche premium leg is nearly
equivalent to that of the CDX, (19), as can be shown by integration by parts.
The difference is in the timing of the accrued premium payment: for the CDX,
this payment is made upon occurrence of default, whereas for the tranche,
this payment is made at the next scheduled premium payment date.18 The
difference in the contract terms may reflect the fact that the event of a loss or
recovery attaching to a tranche is more difficult to determine than a default
event because it requires the computation of Lt,s rather than nt,s . The computa-
tion of Lt,s requires firm i’s recovery, Rτt,i , which may depend on the outcome of
International Swaps and Derivatives Association (ISDA) Credit Event Auction
proceedings and hence is not typically known in real time.
For all but the equity tranche, the upfront payment U is set to zero in our
data, and the spread S is determined in the same way as the CDX as a whole.
However, the equity tranche trades assuming a set spread of 500 basis points
(bps), with U determined so as to equate the premium and the protection legs.19
Following the conventions in the data, therefore, define UTran,1 (λt , ξt ; T − t) to
be the value of U that equates (25) and (22), for j = 1, and S = 0.05. For
j = 2, . . . , J, define STran, j (λt , ξt ; T − t) to be the value of S that, for U = 0,
equates (25) with (22). Section IV of the Internet Appendix describes CDX
tranche pricing.
II. Data
Our analyses require the use of pricing data from options and CDX markets.
The options data, provided by OptionMetrics, consist of daily implied volatili-
ties on S&P 500 European put options from January 1996 to December 2012.
To construct a monthly time series, we use data from the Wednesday of ev-
ery option expiration week. We apply standard filters to extract contracts with
meaningful trade volumes and prices. To obtain an implied volatility curve for
each date, we fit a second-order polynomial in strike price and maturity. We fit
the polynomial for options with maturities ranging from 30 to 247 days, and
with moneyness below 1.1.
Our CDX data come from Markit and consist of daily spreads and upfront
amounts for the October 2005 to September 2008 period on the five-year CDX
North American Investment Grade index and its tranches, excluding the su-
per senior. The CDX North American Investment Grade index is the most
actively traded CDX product. We refer to it in what follows as “the CDX.” This
index represents 125 equally weighted large North American firms that are
investment-grade at the time the series is initiated.
18 Recognizing that n is the CDX equivalent of T L + T R , the CDX equivalent of the integral
t,s j,t,s j,t,s
in (25) is
t+ϑm t+ϑm
1
(1 − nt,s ) ds = (1 − nt,t+ϑm) + (s − t − ϑ(m − 1)) dnt,s ,
t+ϑ(m−1) 4 t+ϑ(m−1)
where the right-hand side is derived using integration by parts. Note that this right-hand side is
equivalent to the change in notional in (19), except for a change in the discounting of the accrued
premium payment.
19 Why this difference? As Collin-Dufresne, Goldstein, and Yang (2012) point out, the spread
payments for the equity tranche are particularly sensitive to defaults. Reducing the spread pay-
ments and adding an upfront amount makes the protection sell position much less risky than it
would be otherwise.
To maintain an (approximately) fixed-maturity contract, a new series for the

CDX is introduced every March and September and the previous series becomes
off-the-run. Our sample corresponds to CDX series 5 through 10.20 We use data
from the series that is most recently issued, and hence most actively traded, in
our analyses. For comparability with prior studies (Collin-Dufresne, Goldstein,
and Yang (2012), Coval, Jurek, and Stafford (2009)), we report average spreads
for two subperiods, with September 2007 being the end of the first subperiod. In
our sample, the CDX and all tranches except for the equity tranche are quoted
in terms of spreads. The equity tranche is quoted in terms of upfront payment
with a fixed spread of 500 bps.21
III. Calibration
In this section, we describe how we calibrate the model. Section III.A de-
scribes the calibration of the preference parameters and the consumption pro-
cess. Section III.B describes firm cash flows. In this section, we show that the
model can match physical default probabilities. Finally, Section III.C describes
how we obtain a time series of state variables from option data.
A. The Stochastic Discount Factor

We keep risk aversion γ , the discount rate β, the normal-times consumption
distribution, the disaster distribution, and leverage φ the same as in Wachter
(2013).22 Note that κλ and σλ do not have the same interpretation in this model.
Our first goal in calibrating the model is to generate reasonable predictions
for the aggregate market and for the consumption distribution. One challenge
in calibrating representative agent models is matching the high volatility of the
price-dividend ratio. There is an upper limit to the amount of volatility that can
be assumed in the state variable before a solution for utility fails to exist—the
more persistent the processes, the lower the respective volatilities must be for
the discriminants in (4) and (5) to be nonnegative. We choose parameters so
that the discriminant equals 0. Thus, there are three parameters to match the
aggregate market, the risk-free rate, and consumption.
20 The liquidity of CDX tranches shrank significantly after CDX10, the last series introduced
before the Lehman crisis. From series 11 on, these products were traded too infrequently for prices
to be meaningful. Moreira and Savov (2016) present a model with time-varying disaster risk that
accounts for qualitative features of structured finance around the crisis, including the lack of
trading in these securities following the Lehman default.
21 Trading conventions changed with the introduction of the Standard North American Contract
(SNAC) in April 2009. Under SNAC, CDX products trade with upfront amounts and fixed coupons
of either 100 or 500 bps.
22 This disaster distribution comes from Barro and Ursúa (2008), who build on the work of Barro
(2006). The distribution is based on international consumption data over roughly the last century.
We also assume a 40% probability of default on the government bill in the case of disaster as
in Barro (2006). Our results are robust to deviations in these distributions, as long as disasters
remain rare and large.
Table I
Properties of Aggregate Cash Flows and Utility
Panel A shows parameters for normal-times consumption and dividend processes, and for the
preferences of the representative agent. Panel B shows the parameter values for λ and ξ processes:
√
dλt = κλ (ξt − λt )dt + σλ λt dBλ,t
√
dξt = κξ (ξ̄ − ξt )dt + σξ ξt dBξ,t .
Note that ξ̄ is the average level of the probability of a disaster. Panel C shows population statistics
for the disaster probability λt . Except for the autocorrelation of λt , values are in annual terms.
Panel A: Parameters for Utility, Consumption, and Dividends
Relative risk aversion γ 3

EIS ψ 1
Rate of time preference β 0.012
Average growth in consumption (normal times) μc 0.0252
Volatility of consumption growth (normal times) σc 0.020
Leverage φ 2.6
Panel B: Parameters for the Disaster Probability Process
Mean reversion κλ 0.20

Volatility parameter σλ 0.1576
Mean reversion κξ 0.10
Volatility parameter σξ 0.0606
Mean ξ̄ 0.02
Panel C: Population Statistics for the Disaster Probability λt
Median 0.0037
SD 0.0386
Monthly autocorrelation 0.9858
The resulting parameter choices are reported in Table I. The parameters κλ

and σλ are relatively high, consistent with our interpretation of λt as the fast-
moving component of disaster risk, while κξ and σξ are relatively low, consistent
with our interpretation of ξt as a slow-moving component. The unconditional
average of the disaster probability, ξ̄ , is 2% per annum. This is a lower average
disaster probability than in Wachter (2013), and thus, our calibration is con-
servative. The extra persistence created by the ξt process implies that λt can
deviate from its average for long periods of time. We report population statis-
tics on λt in Panel C of Table I. The median disaster probability is only 0.37%,
which indicates that the distribution is highly skewed. The standard deviation
is 3.9% and the monthly first-order autocorrelation is 0.986.
Implications for the risk-free rate and the market are shown in Table II.
We simulate 100,000 samples of length 60 years to capture features of the
small-sample distribution (Section V of the Internet Appendix provides details
on the simulation). We also simulate a long sample of 600,000 years to cap-
ture the population distribution. Statistics are reported for both the full set of
100,000 samples and for the subset for which there are no disasters (38% of
the sample paths). As in earlier work (Wachter (2013)), a time-varying disaster
Table II
Moments for the Government Bill Rate and the Market Return in the
Data and the Model
Data moments are calculated using annual data from 1947 to 2010. Population moments are
calculated by simulating data from the model at a monthly frequency for 600,000 years and then
aggregating monthly growth rates to an annual frequency. We also simulate 100,000 60-year
samples and report the 5th , 50th , and 95th percentile for each statistic, both for the full set of
simulations and for the subset of samples for which no disasters occur. Rb denotes the government
bill return, Rm denotes the return on the aggregate market, and p − d denotes the log price-dividend
ratio.
No-Disaster Simulations All Simulations
Data 0.05 0.50 0.95 0.05 0.50 0.95 Population
E[Rb ] 1.25 1.68 2.96 3.46 −0.47 2.41 3.37 2.02

σ (Rb ) 2.75 0.34 1.07 2.71 0.48 2.06 7.14 3.69
E[Rm − Rb ] 7.25 5.40 8.01 12.36 5.30 8.49 14.25 9.00
σ (Rm) 17.8 13.24 19.26 27.91 14.59 22.59 34.38 24.13
Skew(Rm) −0.40 0.01 0.72 2.03 −0.44 0.62 1.99 1.06
Kurt(Rm) 3.11 2.62 4.08 10.51 2.73 4.48 10.95 9.09
Sharpe ratio 0.41 0.32 0.42 0.55 0.26 0.39 0.53 0.37
exp(E[ p − d]) 32.5 28.96 40.63 48.88 22.93 36.95 47.41 35.36
σ ( p − d) 0.43 0.15 0.27 0.47 0.17 0.33 0.59 0.43
AR1( p − d) 0.92 0.59 0.79 0.91 0.62 0.82 0.92 0.90
probability implies a high equity premium, low risk-free rate, and high equity
volatility.23 Only the autocorrelation of the price-dividend ratio falls at the edge
of the 90% confidence bounds.24 Because prices fall and risk premia rise when
the disaster probability increases, the model also captures excess return pre-
dictability without generating counterfactual cash-flow predictability (there is
no cash-flow predictability in normal times).
Table II also reports the skewness and kurtosis of the aggregate market in
both the model and the data. Both fall well within the 90% confidence bounds
when sample paths with disasters are included. In no-disaster samples, returns
are slightly positively skewed, while they are slightly negatively skewed in the
data. Log returns in the model have a small positive median skewness over
no-disaster samples and small negative median skewness over the full set
of samples. Disasters, not surprisingly, lead to substantial skewness: the 95%
critical value in simulations is −3.4. The kurtosis in the model is 4 with samples
without disasters and 4.5 in samples with disasters. In the data, the kurtosis
is 3, which is well within the confidence bounds for both types of samples.
The simulation results in Table II point to substantial downward bias in the
23 The Treasury bill rate predicted by the model slightly exceeds that in the data. This could be
fixed by decreasing β or by decreasing the probability of government default.
24 The price-dividend ratio over the postwar period may be driven by very long-run fluctuations
in the tendency to pay dividends (Fama and French (2001)), which are outside of the scope of the
model. Note that these fluctuations could also be partially responsible for the observed volatility
of the price-dividend ratio, which is also somewhat higher in the data than the model.
measurement of kurtosis: the true (population) kurtosis in the model is 9, while

the median kurtosis observed in samples of length 60 years is 4.5.
B. Firm Cash Flows

To go from state variables to CDX/CDX tranche prices requires dynamics
for individual firm cash flows and assumptions about what constitutes a de-
fault. For parsimony, and following standard practice in this literature, we
assume that firms are ex ante identical. However, firms face distinct idiosyn-
cratic shocks and potentially distinct sector-wide shocks.
To calibrate individual firm dynamics, we use results from Collin-Dufresne,
Goldstein, and Yang (2012). The default boundary AB is set to 19.2%. This
implies that if the asset value falls below 0.192 multiplied by what it was at
the initiation of the contract, the firm is assumed to be in default. This value
derives from the average leverage ratio from firm-level data (32%). Firms are
then considered to be in default if their value is 60% of their debt outstanding.
Following Collin-Dufresne, Goldstein, and Yang (2012), we assume that the
recovery rate is 40% in normal times and 20% in the event of rare disasters.25
For simplicity, we set the idiosyncratic jump size to be large enough that it
makes default virtually certain. We set this value at −95%, but results are not
sensitive to the precise value.
To calibrate the sensitivity of assets to consumption, we use the asset betas
for CDX firms estimated by Collin-Dufresne, Goldstein, and Yang (2012). The
asset betas are found to be between 0.5 and 0.6 for precrisis series and between
0.6 and 0.7 for crisis series. This reflects a slight increase in leverage for the
firms included in the series. In our model, the asset beta is determined largely
by the ratio φi /φ; however, the connection between sector-wide and aggregate
risk adds an additional degree of covariance. We therefore choose φi /φ = 0.5
precrisis and φi /φ = 0.6 during the crisis. Given our assumption of φ = 2.6, this
corresponds to a precrisis φi value of 1.3 and a crisis φi value of 1.6.26
We use the results of Longstaff and Rajan (2008) to calibrate the parameters
for the sector-wide shocks. They estimate a loss rate of 5% on the portfolio in
the event of a sector shock. Because we assume a recovery rate of 40%, the
fraction of firms defaulting in the event of a sector shock is equal to 0.05/(1 −
0.4)=0.08. Given a total of 125 firms, a sector shock corresponds to a default
of about 10 firms. Recall that we assume 25 firms in each of the five sectors.
Thus, the probability of firm default given a sector shock equals 10/25 = 0.4.
We estimate the remaining parameters using the term structure of phys-
ical default probabilities. These default probabilities are an important disci-
plining device for the model because of the well-known credit spread puzzle
(Chen, Collin-Dufresne, and Goldstein (2009), Chen (2010), Huang and Huang
25 Collin-Dufresne, Goldstein, and Yang (2012) do not have consumption disasters in their
reduced-form model, but they do have catastrophic declines in firm value.
26 For simplicity we treat φ as a parameter rather than a process, that is, the shift in φ is a
i
one-time event not anticipated by the agents.
Figure 1. Term structure of physical default probabilities. This figure plots cumulative
default probabilities for horizons ranging from 1 to 10 years in the data and in the model. The solid
line shows the cumulative default probabilities as calculated by Moody’s Investors Services using
data from 1920 to 2010 for BBB bonds. The dotted line shows cumulative default probabilities in
the benchmark version of the model when the state variables are at their unconditional means.
Dashed lines show cumulative default probabilities when the probability of an idiosyncratic shock
λi is zero, when the probability of a sector shock pi is zero, and when the probability of a disaster is
increased by two percentage points relative to its unconditional mean. (Color figure can be viewed
at wileyonlinelibrary.com)
(2012)).27 We want to make sure that we do not achieve high credit spreads
merely because we are introducing a counterfactually high rate of default.
Figure 1 plots the cumulative default probability for horizons ranging from
1 to 10 years, as calculated by Moody’s Investor Services (Ou, Chiu, and Metz
(2011), Exhibit 33) for the period from 1920 to 2010.28 The cumulative default
probability is 0.3% at one year, rising to 7% at 19 years. We search over the
parameter space to find the remaining three parameter values: the sector jump
size e ZSi ,t (assumed to be constant for parsimony), wξ , and λi . To keep the
exercise tractable, we calculate the default distribution assuming that the state
variables are at their unconditional means. Table III reports the resulting
parameter values for our model.29
27 Recent papers that examine credit spreads from a disaster risk perspective include Christof-
fersen, Du, and Elkamhi (2016), Gabaix (2012), and Gourio (2013). These papers do not look at the
CDX.
28 We follow the literature on the credit spread puzzle in using default probabilities for BBB-
rated firms. In fact, most reference entities in the CDX index are BBB rated (Amato and Gyntelberg
(2005))
29 To maintain a parsimonious model, we assume that firm cash flows do not exhibit idiosyncratic
diffusive and sector risk. If we added diffusive risk of this type, the idiosyncratic and sector
Table III
Parameter Values for an Individual Firm
This table reports the parameters for the payout process on an individual firm. Note that de-
fault boundary AB is calculated as 60% of the average leverage ratio of the reference entities.
Idiosyncratic jump intensity λi is in annual terms.
Default boundary AB 19.2%

Recovery rate Rt (normal times) 40%
Recovery rate Rt (disaster times) 20%
Aggregate risk loading φi (precrisis) 1.3
Aggregate risk loading φi (crisis) 1.6
Idiosyncratic jump size (e Zi − 1) −95.0%
Bernoulli parameter for sector shocks P(Iit = 1) = pi 0.4
Sector-wide jump size (e ZSi − 1) −75.9%
Coefficient for sector-wide jump intensity wξ 1.402
Idiosyncratic jump intensity λi 0.0032
The resulting fit to the term structure of physical default probabilities is

shown in Figure 1. The model and data are nearly indistinguishable. Figure 1
also plots the curves that result when we take away idiosyncratic shocks and
sector shocks, and when we raise the disaster probability. Eliminating sector
shocks flattens the term structure because these shocks are autocorrelated
(their probability varies with ξt ). The disaster probability adds very little to the
term structure of default probabilities: increasing the disaster probability by
a full two percentage points results in a negligible change in the curve. Under
our calibration, disasters that are sufficiently large to cause default are quite
rare. When such disasters do occur, however, a large number of defaults occur
together, with implications for the pricing of senior tranches.
C. State Variables
In an earlier paper (Seo and Wachter (2018)), we establish that a special
case of the model in Section III.A can fit average option-implied volatili-
ties in the data. Our previous model (which has one state variable rather
than two) captures the fact that implied volatilities are higher than realized
volatilities and that out-of-the money (OTM) put options have higher implied
volatilities than at-the-money (ATM) options. Our model explains these facts
through the mechanisms of excess kurtosis arising from disasters and high
normal-times volatility arising endogenously from variation in the disaster
probability.30
probabilities might change. However, the fit to physical default probabilities and CDX tranche
spreads would change very little.
30 As Drechsler (2013) shows, standard models with recursive preferences generate an implied
volatility curve that is too flat. For this reason, Drechsler (2013) and others (e.g., Liu, Pan, and
Wang (2005), Shaliastovich (2015)) consider belief dynamics that depart from rational expectations.
Our results show that rare events offer another avenue for explaining option prices.
Because the current model nests the previous one and is similarly calibrated,
it is not surprising that this model can explain average implied volatilities.
Moreover, the second state variable allows it to capture time variation in the
slope of the implied volatility curve. Because we use the model to infer the
state variables for pricing CDX tranche spreads, this property is desirable.
Figure IA.VII.C in the Internet Appendix plots averages and volatilities of
implied volatilities in the data and the model.
Because the model can account for both the level of and time variation in
implied volatilities, we can use the implied volatilities to discipline our choice
of state variables. Section III.B shows that implied volatilities are functions
of λt and ξt . Given the implied volatilities in the data, we can determine a
time series for the state variables by inverting the implied volatility func-
tion. While our period of interest begins in October 2005, the options data
go back to 1996. We choose state variables to exactly fit one-month ATM
and OTM (0.85 moneyness) in the data. Figure IA.3 in the Internet Ap-
pendix shows a good fit to the time series of the three- and six-month implied
volatilities.
We plot the time series of the extracted state variables in Panel A of Figure 2.
For most of this sample, the disaster probability λt and its mean ξt lie below
5%. The state variables rise in late 1998, which corresponds to the rescue
of long-term capital management following the Asian financial crisis and the
moratorium on payments on Russian debt. There are increases corresponding
to market declines following the NASDAQ boom in the early 2000s. The sample,
however, is dominated by the financial crisis of late 2008 to 2009. At the time of
the Lehman default, the disaster probability rises as high as 20%. The disaster
probability remains high and volatile, as compared to the prior period, through
the end of the sample.
Panel B of Figure 2 focuses on the sample period for which we have CDX
tranche data. This period captures the first hints of the crisis in early 2007,
with slight increases in the level and volatility of λt and ξt . These variables
become markedly higher and more volatile in 2007 and early 2008, culminating
in the near-default of Bear Sterns. The CDX tranche sample ends right before
the Lehman default; the last values of λt are higher than before, but notably
lower than one month later.
Table IV reports statistics from the time series of extracted state variables
(under the column headed “Data”), as well as medians and critical values
from simulations of the model. Panel A reports data and critical values for
the options sample. The table shows that the average disaster probability is
1.6% per annum, and volatility is 2.4%. Both are well within the 90% criti-
cal values, as are the mean and volatility of ξ . The autocorrelation in λt is
also similar in the model and the data. However, the autocorrelation for ξt is
estimated at 0.70 per month in the data, which is well below the 5% confi-
dence bound of 0.90. The low autocorrelations in the extracted λt and ξt re-
flect the low autocorrelation of implied volatility. We return to this issue in
Section V.
Figure 2. Time series of state variables. This figure plots monthly time series of the state
variables λt (annual disaster probability) and ξt (long-run mean of λt ) extracted from option prices.
At each time point, the state variables are chosen to match implied volatilities of one-month ATM
and OTM (moneyness of 0.85) index put options in the data. Panel A shows these time series for
the January 1996 to December 2012 period over which option data are available. The shaded area
represents the period over which CDX tranche data are available. Panel B zooms into the CDX
sample period. (Color figure can be viewed at wileyonlinelibrary.com)
Table IV
State Variable Moments in the Data and in the Model
A monthly time series of the state variables λt (annual disaster probability) and ξt (long-run mean
of λt ) are extracted from option prices for 1996 to 2012. The table reports means, volatilities,
and first-order autocorrelations for the extracted state variables for both the full sample and the
subsample over which the CDX data are available. Means and volatilities are in percentage terms.
We simulate 100,000 samples from the model of length matching that of the relevant sample and
report the 5th , 50th , and 95th percentile for each statistic, both for the full set of simulations and
for the subset of samples for which no disasters occur.
No-Disaster Simulations All Simulations
Data 0.05 0.50 0.95 0.05 0.50 0.95
Panel A: Options Sample (January 1996 to December 2012)
E[λ] 1.57 0.05 0.69 4.44 0.06 1.01 7.36

σ (λ) 2.37 0.13 1.05 4.26 0.16 1.40 5.88
AR1(λ) 0.74 0.71 0.92 0.97 0.74 0.94 0.98
E[ξ ] 1.70 0.43 1.41 4.31 0.46 1.58 4.97
σ (ξ ) 1.06 0.32 0.83 2.03 0.34 0.89 2.21
AR1(ξ ) 0.70 0.90 0.96 0.98 0.91 0.96 0.98
Panel B: CDX Sample (October 2005 to September 2008)
E[λ] 0.74 0.00 0.43 7.60 0.00 0.51 9.02

σ (λ) 0.90 0.01 0.49 2.95 0.01 0.54 3.23
AR1(λ) 0.16 −0.01 0.74 0.91 0.00 0.75 0.91
E[ξ ] 1.41 0.21 1.42 5.43 0.21 1.47 5.62
σ (ξ ) 1.14 0.12 0.42 1.11 0.13 0.42 1.12
AR1(ξ ) 0.83 0.59 0.83 0.93 0.59 0.83 0.93
IV. Evaluation
Section IV.A compares the model’s predictions for average CDX and CDX
tranche spreads to those in the data. Section IV.B considers the time series in
the model and data, and Section IV.C performs comparative static exercises.
A. Average Spreads
Before discussing the tranches, we first discuss the fit to CDX spreads them-
selves. Unlike spreads on the tranches, which are informative about the corre-
lational structure of the underlying firms, and specifically about the properties
of extremely rare events, CDX spreads themselves are informative about the
average level of credit risk in the pool. Figure 3 plots the term structure of aver-
age CDX spreads in the model for both the precrisis period and the crisis period.
We also plot data for the precrisis period, available from Collin-Dufresne, Gold-
stein, and Yang (2012).31 Recall that, for the CDX and all tranches except the
31 Our data are for the five-year maturity only. For other maturities, we use the values that
Collin-Dufresne, Goldstein, and Yang (2012) report. They do not report average spreads for the
crisis period, which is why average spreads do not appear on Figure 3.
Figure 3. Term structure of CDX spreads. This figure plots average spreads on the CDX index
as a function of maturity in the data and in the model. Averages in the data are from the precrisis
period (September 2004 to September 2007) and are reported by Collin-Dufresne, Goldstein, and
Yang (2012). Average spreads in the model are computed using the state variables from one-month
options over the same period. The figure also shows averages implied by the model for the crisis
period. Spreads are in terms of basis points per unit of notional and are annual. (Color figure can
be viewed at wileyonlinelibrary.com)
equity tranche, these spreads represent the annual payment per unit notional.
For example, the average spread of 27 bps for the 15% to 30% (third senior
tranche) indicates that a protection buyer pays $0.0027 per year to insure
$1.00 of value for this tranche.32 For the equity tranche, the spread is fixed at
500 bps, so we report the upfront payment (in percentage points per unit of
notional). Figure 3 shows that the term structure is upward sloping in both the
model and the data in the precrisis period. The model implies a much higher
level during the crisis than in the precrisis period.
One way to understand these spreads is to consider the loss distributions
under the physical and risk-neutral measures. These are shown in Figure 4
for five-year contracts.33 The top two panels present loss distributions under
the physical and risk-neutral measures for each date in the sample. The figure
32 As described in Section III.E, there are adjustments based on defaults in the portfolio.
33 To obtain loss distributions under both measures, we first simulate the model to acquire
100,000 five-year samples of the CDX loss together with the Radon-Nikodym derivative process
(Section V of the Internet Appendix). Estimating physical loss distributions is straightforward be-
cause the simulated CDX losses are under the physical measure. For risk-neutral loss distributions,
we use the following relation, which holds for l, > 0:

dQ
P Q( − ≤ LT ≤ + ) = E Q 1{−≤LT ≤+ } = E 1{−≤LT ≤+ } .
dP
To construct smooth distributions, we apply kernel density estimation using the normal kernel
smoothing function with a bandwidth of 0.005.
Figure 4. Loss distributions. The top panels show the distributions of cumulative five-year
losses on the CDX for each of the 36 months in our data; the left panel shows losses under the
physical distribution while the right panel shows losses under the risk-neutral distribution. The
bottom panels show the physical and risk-neutral distributions for specific dates, one precrisis and
one during the crisis. (Color figure can be viewed at wileyonlinelibrary.com)
shows that the risk-neutral distributions shift toward the right, toward greater
loss distributions. More importantly, the risk-neutral measure increases tail
thickness, especially on some dates in the sample. The bottom two panels show
a closer look at two particular dates, one precrisis (March 2007) and one during
the crisis (March 2008). The physical and risk-neutral measures are shown on
the same figure. Precrisis, the risk-neutral measure has a thicker tail, but the
effect is small compared to the crisis. The greater probability of large losses
during the crisis leads to the much larger CDX spreads.
Table V moves beyond these results to report spreads for the tranches for
the five-year CDX, as well as for the index itself. The table shows that the
model can match the average spreads on the five-year CDX and its tranches for
the full sample period. More importantly, the model can match the very large
shift in spreads in senior tranches between the precrisis and crisis periods. For
Table V
Average CDX and CDX Tranche Spreads (Five-Year Maturity)
This table reports historical and model-implied average five-year CDX and CDX tranche spreads
in basis points per year. For the equity tranche (0% to 3%), the spread is fixed at 500 bps, so we
report the upfront payment. The other tranches (3% to 7%, 7% to 10%, 10% to 15%, 15% to 30%)
and the CDX itself have no upfront payments. Data are monthly, from October 2005 to September
2008, corresponding to CDX series 5 to 10. We divide the data into two subsamples: precrisis and
crisis. The precrisis sample is from October 2005 to September 2007 (CDX5 to CDX8). The crisis
sample is from October 2007 to September 2008 (CDX9 to CDX10). We compute model values from
state variables fit to the time series of one-month ATM and OTM implied volatilities on equity
index options.
Upfront (%) Annual Spread (bps)
0% to 3% 3% to 7% 7% to 10% 10% to 15% 15% to 30% CDX
Panel A: Precrisis (October 2005 to September 2007)
Data 31 108 25 12 6 42
Model 20 152 61 25 7 38
Panel B: Crisis (October 2007 to September 2008)
Data 54 498 255 136 69 116

Model 34 504 279 156 65 102
Panel C: Full Sample (October 2005 to September 2008)
Data 39 238 102 54 27 67

Model 25 270 133 69 27 59
example, for the 15% to 30% tranche, the average spread changes tenfold (from
6 to 69 bps) between the samples. While the model generates a substantial
change in the senior tranches, it correctly predicts much less of a shift in the
junior tranches.
Table VI shows that the model can also match the spreads on the index
and on the tranches for the three-year maturity. The model can match the
very low average spreads prior to the crisis, and the high spreads during the
crisis. The model also captures the fact that three-year senior tranche spreads
are lower, but not much lower, than five-year spreads. High three-year senior
tranche spreads present a particular challenge to models that assume normally
distributed risk, as discussed in Collin-Dufresne, Goldstein, and Yang (2012).34
Finally, Table VII reports average spreads on the super senior tranche. As
is the case for Table VI, data are from Collin-Dufresne, Goldstein, and Yang
(2012). The super senior tranche is affected only if there is a loss of 30% or
more on the CDX itself. Given the 20% recovery rate that we assume for dis-
aster periods, about 40% of the 125 firms would need to go into default for
insurance on the tranche to pay off. Nowhere can rare disaster fears be seen
34 See also the literature on single-name corporate bonds and credit default swap (CDS) (Zhou
(2001), Culp, Nozawa, and Veronesi (2014)).

Table VI
Average CDX and CDX Tranche Spreads (Three-Year Maturity)
This table reports historical and model-implied average three-year CDX and CDX tranche spreads
in basis points per year. For the equity tranche (0% to 3%), the spread is fixed at 500 bps, so we
report the upfront payment. The tranches (3% to 7%, 7% to 10%, 10% to 15%, 15% to 30%) and the
CDX itself have no upfront payments. Data, from Collin-Dufresne, Goldstein, and Yang (2012), are
monthly from September 2004 to September 2008. The precrisis sample is from September 2004
to September 2007 (CDX3 to CDX8). The crisis sample is from October 2007 to September 2008
(CDX9 and 10). We compute model values from state variables fit to the time series of one-month
ATM and OTM implied volatilities on equity index options.
Upfront (%) Annual Spread (bps)
0% to 3% 3% to 7% 7% to 10% 10% to 15% 15% to 30% CDX
Panel A: Precrisis (September 2004 to September 2007)
Data 11 20 8 3 2 27
Model 10 81 23 8 3 30
Panel B: Crisis (October 2007 to September 2008)
Data 43 364 168 87 48 –

Model 20 390 174 86 44 84
Panel C: Full Sample (September 2004 to September 2008)
Data 21 127 58 29 16 –
Model 12 157 60 27 13 43
more clearly than in the increase in average spreads, from near zero to 30
bps, on these tranches. This implies that investors were willing to pay about
$0.0150 to insure $1 of notional, that is, the risk neutral probability was (very
approximately) about 1.5%. Even if one attributes all of the volatility of asset
prices to rare events, which implies correlations close to one, and assumes zero
recovery, a model with lognormally distributed prices is unable to come close
to these probabilities.35
To summarize, Table VII shows that the model captures the level of spreads
before the crisis, the dramatic change during the crisis, and the relative spreads
between the three- and five-year maturities.
B. Time Variation in Spreads

We now calculate the implied time series of spreads on CDX and CDX
tranches based on our state variables extracted from options data, and com-
pare them to the historical time series. Figure 5 plots the monthly time series of
five-year maturity CDX and CDX tranche spreads in the data and in the model.
The blue solid line represents the data and the red dotted line represents the
35 Standard calculations show that, in particular, default probabilities for three-year tranches
are two orders of magnitude below these values under a normal distribution.
Table VII
Average Super Senior Tranche Spreads
This table reports historical and model-implied average three- and five-year super senior tranche
spreads in basis points per year. The super senior tranche has a 30% attachment point and a 100%
detachment point. Data, from Collin-Dufresne, Goldstein, and Yang (2012), are monthly from
September 2004 to September 2008. The precrisis sample is from September 2004 to September
2007 (CDX3 to CDX8). The crisis sample is from October 2007 to September 2008 (CDX9 and 10).
We compute model values from state variables fit to the time series of one-month ATM and OTM
implied volatilities on equity index options.
Super Senior Tranche Spread (Annual bps)
Precrisis (September 2004 to September 2007) Crisis (October 2007 to September 2008)
Three-Year Five-Year Three-Year Five-Year
Data 1 4 23 35
Model 2 3 27 34
benchmark model. In the same figure, we also plot results for the case without
idiosyncratic risk, discussed further below. The top left panel plots spreads on
the CDX, and the other five panels plot spreads on tranches ranging from eq-
uity to the third senior tranche (15% to 30%). As discussed in Section II, when
the equity tranche is traded, the protection buyer makes an upfront payment
to the protection seller in addition to fixed annual premium payments of 500
bps. Thus, for the equity tranche (top right panel), we show the amount of this
upfront payment.
Table V above shows that the model accurately captures the levels of the
spreads on the CDX and its tranches for both the precrisis and the crisis
periods. What is new in Figure 5 is that the timing of the increase in spreads
and the fluctuations both before and during the crisis are accurately captured
by our model. That is, the same period of low spreads and low volatility that
characterized stock and options markets in 2006 and 2007 is also apparent
in the low spreads on structured finance products. The increased prices for
protection, which slightly predate the collapse of Bear Sterns, appear almost
simultaneously in options and CDX/CDX tranche spreads. The crisis period
was one of high volatility in both markets, with the two fluctuating in tandem.
The model captures the increase in ATM and OTM options over this period
with an increase in both the probability of disaster and the long-run mean of
this probability. The latter increase helps the model explain the magnitude of
the increase in CDX spreads, which have a longer maturity than do options.
As a check on these results, we also examine implications for the time series
of the super senior tranche. For this tranche, we have data during the crisis.
We also have a precrisis average from Collin-Dufresne, Goldstein, and Yang
(2012). Figure 6 plots the spreads in the data and implied by the model. The
figure confirms the results of Table VII: the model matches the precrisis and
crisis levels of this series well. In addition, the model (based on option-implied
state variables) captures the peaks in March and September of 2008 and the
Figure 5. Time series of CDX/CDX tranche spreads in the benchmark model and in a
case without idiosyncratic risk. This figure plots monthly time series of five-year CDX and
CDX tranche spreads in the data (blue solid lines), in the benchmark model (red dotted lines), and
in a calibration without idiosyncratic risk (black dashed lines). Spreads are annual and reported
in terms of basis points per unit of notional. For the equity tranche, we report the upfront payment
assuming the spread is fixed at 500 bps. We compute model values from state variables fit to
the time series of one-month ATM and 0.85 OTM implied volatilities. For the model without
idiosyncratic risk, we set the probability of an idiosyncratic shock λi to zero. (Color figure can be
viewed at wileyonlinelibrary.com)
intermediate dips. As can also be seen from Figure 5, the increase in option
prices leads the increase in CDX prices by about two months.
The only tranche that the model does not fit closely is the equity one. While
the model can match the timing of the increase in the upfront payment on the
Figure 6. Time series of super senior tranche spreads. This figure plots monthly time series
of five-year CDX super senior tranche spreads (attachment point = 30%) in the model and in
the data. Spreads are annual and reported in terms of basis points per unit of notional. We
compute model values from state variables fit to the time series of one-month ATM and 0.85 OTM
implied volatilities. For the precrisis data, we show the average value, reported by Collin-Dufresne,
Goldstein, and Yang (2012). (Color figure can be viewed at wileyonlinelibrary.com)
equity tranche, it does not entirely capture its magnitude. Nor does it capture
some of the variation in the early part of the sample, perhaps related to a
credit crisis triggered by Ford and General Motors’ downgrades. Nonetheless,
the approximate magnitude of these spreads is well matched, despite the fact
that we constrain idiosyncratic risk to take the same value throughout the
sample period.36
C. Comparative Statics
We now consider the effect of changes in our assumptions on firm cash flows
(12). The dashed line in Figure 5 shows the effect of setting the probability of
an idiosyncratic decline in firm cash flows, λi , to zero. We see that the tranches
respond quite differently to elimination of this risk. Specifically, the upfront
payment on the equity tranche drops precipitously to a negative value, indicat-
ing that the spread of 500 bps is higher than what would be required under no
upfront payment. It appears that almost all of the spreads on this tranche are
36 One important simplifying assumption in our model is that firms have the same level of
idiosyncratic risk. If firms have heterogeneous risks, as is reasonable, equity tranche upfront
payments will be sensitive to the highest idiosyncratic volatilities in the pool. On the other hand,
average default probabilities, depicted in Figure 1, will still be determined by the average idiosyn-
cratic volatilities.
case without sector shocks. This figure plots monthly time series of five-year CDX and CDX
tranche spreads in the data (blue solid lines), in the benchmark model (red dotted lines), and in a
calibration without sector shocks (black dashed lines). Spreads are annual and reported in terms of
basis points per unit of notional. For the equity tranche, we report the upfront payment assuming
the spread is fixed at 500 bps. We compute model values from state variables fit to the time series
of one-month ATM and 0.85 OTM implied volatilities. For the model without sector risk, we set
the probability that a firm will suffer a sector shock pi to zero. (Color figure can be viewed at
wileyonlinelibrary.com)
due to idiosyncratic risk. As tranches increase in seniority, idiosyncratic risk

has a diminishing effect. For the third senior tranche, there is no discernable
effect of idiosyncratic risk at all.
Figure 7 shows the effect of eliminating sector risk (the dashed line). This
limiting case is achieved by setting pi equal to 0. We find that eliminating sector
risk has the largest effect on tranches with intermediate levels of seniority.
Without sector risk, the spreads on the mezzanine and first senior tranche are
near zero for the entire sample. CDX tranche data thus clearly require a level
of commonality among firms somewhere between the firm-specific idiosyncratic
risk and market-wide risk.
Figure 7 shows that the third senior tranche, with an attachment point
of 15% and detachment point of 30%, is barely affected when the probabil-
ity of a sector shock is set to zero. This confirms the intuition that only rare
market-wide shocks have a significant effect on the pricing of this tranche.
We can also see the important role of rare disasters in pricing this tranche
by recalculating CDX/CDX tranche prices under a two-percentage-point in-
crease in the probability of rare disasters at every point in time (two per-
centage points is the annual mean of the disaster probability). Figure 8 plots
the results. Spreads on the third senior tranche increase by about 25 bps in
the precrisis period and 50 bps in the crisis period. The increase is substan-
tially less than the full two percentage points because disaster risk is mean-
reverting, and λt gives an instantaneous probability of disaster. Thus, over
the full five years of the contract, investors expect λt to be elevated, but by
much less than two percentage points. The reason the effect is larger during
the crisis period is that ξt is larger, and thus λt falls more slowly. Moreover,
because of the square root term in (2a), higher values of λt (which prevail
over the crisis period) are associated with greater volatility of λt . Because
λt is positively skewed, this raises the probability that the high values will
persist.
Note that the effects of a change in the disaster probability on the tranches
have the opposite pattern as a change in the probability of an idiosyncratic
rare event: the more senior the tranche, the greater the effect of a change
in the probability of disaster. The top right panel of Figure 8 shows that the
effect on the upfront payment of the equity tranche is nearly undetectable,
while the increase in the spread of the senior tranche is large. On some level,
it is surprising that such a large increase in risk would not affect the equity
tranche. The reason that the disaster probability does not much affect the
equity tranche is that this tranche is likely to be penetrated regardless of
whether a rare disaster occurs.
Figure 8 shows how a temporary increase in the probability of a rare dis-
aster impacts CDX tranche prices. We now ask how a permanent change in
disaster risk affects tranche prices. We consider a version of the model in
which disasters have the same probability but are much less severe (specifi-
cally, we assume that log consumption declines by 10 percentage points less
than in our benchmark calibration). This reduction leads to a smaller eq-
uity premium of 4% rather than 8%. Figure 9 shows that crisis-level senior
tranche spreads are strongly affected by this change in calibration. While
spreads rise during the crisis period, the increase is less than one-third of
that in the previous calibration. This result highlights the role of equilibrium
restrictions. The disaster distribution, average probabilities, and risk aver-
sion are calibrated to the equity premium puzzle. Figure 9 shows that the
case with higher disaster risk. This figure plots monthly time series of five-year CDX and CDX
tranche spreads in the data (blue solid lines), in the benchmark model (red dotted lines), and in a
calibration with higher disaster risk. Spreads are reported in terms of basis points. For the equity
tranche, we report the upfront payment because the spread is fixed at 500 bps. The benchmark
model values are computed using the option-implied state variables fit to the time series of one-
month ATM and 0.85 OTM implied volatilities. The calibration with higher disaster risk is obtained
by uniformly increasing the disaster probability λt by two percentage points. (Color figure can be
viewed at wileyonlinelibrary.com)
connection between the equity premium puzzle and the anomalous behavior of
tranche spreads is tight. Had the equity premium required a different disas-
ter distribution, it is very unlikely that our model could have explained CDX
spreads.
Figure 9. Time series of CDX/CDX tranche spreads when the equity premium is low.
This figure plots monthly time series of five-year CDX and CDX tranche spreads in the data, in
the model under the benchmark calibration, and in the model under a calibration to an equity
premium of 4%. To obtain this calibration, we translated the distribution of ZC,t by 0.10 (e.g., a
change in log consumption of −0.20 would be −0.10 in the new calibration). Spreads are reported
in terms of basis points . For the equity tranche, we report the upfront payment because the spread
is fixed at 500 bps. (Color figure can be viewed at wileyonlinelibrary.com)
V. Implications for Other Asset Classes

In this section, we discuss the model’s fit to option prices and long-term
bonds, and the ability of the model to explain the comovement (or lack thereof)
in equity prices and options.
A. Option-Implied Volatilities
As discussed above, we use implied volatilities from one-month options to
back out a time series of state variables. We would be unable to do this if the
model did not deliver a plausible level and variability of ATM and OTM implied
volatilities (as Table IV reports, the mean and volatility of the extracted state
variables are close to the ex ante model-implied values).37
Nonetheless, the model fails to capture some aspects of implied volatility
data. Specifically, the model cannot simultaneously account for the high auto-
correlation of the aggregate price-dividend ratio and the low autocorrelation of
implied volatilities. Because the calibration comes close to the former, it fails on
the latter. Table IV shows that the state variables extracted from option prices
have a lower autocorrelation than the model would predict, with the autocorre-
lation of ξt being outside the 10% confidence intervals. In a reduced-form model,
price volatility is exogenous. In an equilibrium model, however, volatility and
prices are driven by the same underlying factors. It is puzzling, from the point
of view of this model (and potentially others), that their autocorrelations are
so different.
In addition, the model predicts excess movement on the long end of the
implied volatility curve in response to shocks at the short end. Figure IA.3
in the Internet Appendix shows that six-month implied volatilities move too
much in response to a change in one-month implied volatilities. This may also
reflect the persistence implied by the model (so that shocks to short-term claims
carry through to long-term claims), relative to the true persistence of implied
volatilities in the data.
B. The Time Series of the Price-Dividend Ratio

Given the time series of state variables, extracted from options data, we
can compute the price-dividend ratio on the aggregate market using (8).
Figure 10 compares the results to the price-dividend ratio in the data
(from Robert Shiller’s webpage, http://www.econ.yale.edu/shiller/data.htm).
The model correctly captures the level of the price-dividend ratio af-
ter 2004, and approximates the timing and magnitude of the finan-
cial crisis and subsequent recovery. Indeed, between 2004 and 2013,
the correlation between the price-dividend ratio generated by the model and
the actual price-dividend ratio is 0.84. This suggests that options and equity
do indeed share a common source of risk.
However, the model does not capture the increase in the price-dividend ratio
during the 1990s and its subsequent decline. Thus, whatever factors were
driving the aggregate market in this period do not appear to be reflected in
options data, at least to the extent that our model can detect.
37 While reduced-form models of the pricing kernel can be calibrated to match implied volatilities
exactly, this is not necessarily the case for equilibrium models. Indeed, Du (2011) and Drechsler
(2013) discuss the difficulty that habit and long-run risk models have in matching the level of
implied volatilities.
Figure 10. The price-dividend ratio in the data and in the model. The solid line shows
the time series of the price-dividend ratio in the data. The red line shows the price-dividend ratio
implied by the model for state prices chosen to fit the one-month ATM and OTM (0.85 moneyness)
put options. (Color figure can be viewed at wileyonlinelibrary.com)
C. Long-Term Bonds
Table II shows that the model can capture the low mean and volatility of the
government bill return. What about long-term bonds?
Section VI of the Internet Appendix solves for bond prices in the model. As
is standard (e.g., Barro (2006)), we assume a 40% default rate on government
bills in the case of disaster, with a percentage loss equaling the decline in con-
sumption. In Section VI of the Internet Appendix, we extend this formulation
to long-term bonds. We assume that a disaster results in a 40% reduction in the
face value of all bonds. While tractable, this likely underestimates the effect of
sovereign default on bond prices because it implies that, in the event of default,
bonds still retain most of their hedging properties.
Figure 11 plots the time series of the five-year continuously compounded
bond yield implied by the model and in the data (in the model we assume zero
coupons; in the data we use the TIPS yield).38 This figure shows that the model
fits the level of the five-year TIPS yield and produces reasonable time variation.
For instance, the model matches the slow decrease in interest rates from 2%
to 0% that took place in late 2007 and early 2008. This arises because of the
precautionary savings effects of the fear of rare disasters on interest rates.
For sufficiently long-term bonds, however, the model’s fit to the data breaks
down. Backus, Chernov, and Zin (2014) demonstrate a tension, in a wide class
of models, between the ability to match the equity premium and the ability
38 As with the price-dividend ratio, we compute the bond yield in the model given the state
variables extracted from one-month options.

Figure 11. Time series of five-year government bond yields in the model and in the data.
This figure plots the time series of model-implied real zero-coupon bond yields and of real yields on
Treasury Inflation Protected Securities (TIPS). Bond yields in the model are computed assuming
a probability of default q, at which point the face value of the bond falls by the same percentage as
consumption, but the maturity of the bond remains the same. Bond yields in the data come from
the U.S. Department of the Treasury. (Color figure can be viewed at wileyonlinelibrary.com)
to match the term structure. Our model shares this tension. The real term
structure in the model becomes sharply downward-sloping at a maturity of 16
years and, at 30 years, bond prices fail to exist.39
The average real term structure is downward-sloping in the model because
long-term bonds protect against an increasing disaster probability.40 The value
of this protection depends on the representative agent’s belief about its reliabil-
ity in the worst states. Sovereign default, either outright or through inflation
(and an inability or unwillingness to compensate TIPS holders), would com-
promise this reliability. Whether or not sovereign default can help explain this
aspect of the data is an open question.
VI. Conclusion
In this paper, we build a quantitative model of spreads on the CDX and
its tranches based on underlying economic fundamentals. When the model is
calibrated to match the equity premium and equity volatility, and when state
variables are chosen to match the time series of implied volatilities on option
prices, the model can explain the level and time series of spreads both before
and during the 2008 to 2009 financial crisis.
39 See Wachter (2013) for a discussion.

40 Inflation creates a countervailing effect for nominal bonds. See Gabaix (2012) and Tsai (2016).
CDX senior tranches are extremely deep OTM put options on the U.S.
economy because they incur losses only when a substantial portion of large
investment-grade firms default. We explain the level of spreads on these in-
struments by introducing a time-varying probability of economic disaster. This
economic disaster causes large simultaneous declines in the consumption of
the representative agent, in aggregate cash flows, and in cash flows on individ-
ual firms. When agents foresee an increased probability of economic disaster,
risk premia rise, asset prices decline and become more volatile, and systemic
defaults become more likely. The economic disasters to which we calibrate the
model are reasonable in light of what has been observed in international data
over the last 100 years. The probability of such a disaster need not be high to
explain senior tranche spreads—4% is sufficient.
While our model can price the CDX and its tranches from the ground up,
that is, in a way that is consistent with consumption data and the equity pre-
mium, several puzzles remain. Most notably, our model implies that the same
set of state variables drive option and equity prices. However, the first-order
autocorrelation of implied volatilities on options is low, while equity prices are
persistent. A second puzzle relates to the real term structure, which is sharply
downward-sloping for bonds of sufficient maturity. These puzzles suggest di-
rections for future research.
Initial submission: September 14, 2016; Accepted: November 7, 2017

Editor: Stefan Nagel, Philip Bond, Amit Seru, and Wei Xiong
REFERENCES
Abel, Andrew, 1990, Asset prices under habit formation and catching up with the Joneses, American
Economic Review Papers and Proceedings 80, 38–42.
Amato, Jeffery D., and Jacob Gyntelberg, 2005, CDS index tranches and the pricing of credit risk
correlations, BIS Quarterly Review March, 73–87.
Backus, David, Mikhail Chernov, and Stanley Zin, 2014, Sources of entropy in representative agent
models, The Journal of Finance 69, 51–99.
Barro, Robert J., 2006, Rare disasters and asset markets in the twentieth century, Quarterly
Journal of Economics 121, 823–866.
Barro, Robert J., and José F. Ursúa, 2008, Macroeconomic crises since 1870, Brookings Papers on
Economic Activity 39, 255–350.
Black, Fischer, and John C. Cox, 1976, Valuing corporate securities: Some effects of bond indenture
provisions, The Journal of Finance 31, 351–367.
Black, Fischer, and Myron Scholes, 1973, The pricing of options and corporate liabilities, Journal
of Political Economy 81, 637–654.
Borovička, Jaroslav, Lars Peter Hansen, Mark Hendricks, and José A. Scheinkman, 2011, Risk-
price dynamics, Journal of Financial Econometrics 9, 3–65.
Campbell, John Y., 2003, Consumption-based asset pricing, in George Constantinides, Milton
Harris, and Rene Stulz, eds.: Handbook of the Economics of Finance, vol. 1b (Elsevier Science,
North-Holland, Amsterdam).
Campbell, John Y., and Glen B. Taksler, 2003, Equity volatility and corporate bond yields, Journal
of Finance 58, 2321–2349.
Chen, Hui, 2010, Macroeconomic conditions and the puzzles of credit spreads and capital structure,
Journal of Finance 65, 2171–2212.
Chen, Long, Pierre Collin-Dufresne, and Robert S. Goldstein, 2009, On the relation between the
credit spread puzzle and the equity premium puzzle, Review of Financial Studies 22, 3367–
3409.
Christoffersen, Peter, Du Du, and Redouane Elkamhi, 2016, Rare disasters, credit, and option
market puzzles, Management Science 63, 1341–1364.
Collin-Dufresne, Pierre, Robert S. Goldstein, and Fan Yang, 2012, On the relative pricing of long-
maturity index options and collateralized debt obligations, Journal of Finance 67, 1983–2014.
Coval, Joshua D., Jakub W. Jurek, and Erik Stafford, 2009, Economic catastrophe bonds, American
Economic Review 99, 628–666.
Culp, Christopher L., Yoshio Nozawa, and Pietro Veronesi, 2014, Option-based credit spreads,
NBER Working Paper No. 20776.
Drechsler, Itamar, 2013, Uncertainty, time-varying fear, and asset prices, The Journal of Finance
68, 1843–1889.
Du, Du, 2011, General equilibrium pricing of options with habit formation and event risks, Journal
of Financial Economics 99, 400–426.
Duffie, Darrell, 2001, Dynamic Asset Pricing Theory, 3rd edition (Princeton University Press,
Princeton, NJ).
Duffie, Darrell, and Larry G. Epstein, 1992, Asset pricing with stochastic differential utility, Review
of Financial Studies 5, 411–436.
Duffie, Darrell, and Nicolae Garleanu, 2001, Risk and valuation of collateralized debt obligations,
Financial Analysts Journal 57, 41–59.
Duffie, Darrell, Jun Pan, and Kenneth Singleton, 2000, Transform analysis and asset pricing for
affine jump-diffusions, Econometrica 68, 1343–1376.
Duffie, Darrell, and Kenneth J. Singleton, 1997, An econometric model of the term structure of
interest-rate swap yields, Journal of Finance 52, 1287–1321.
Epstein, Larry, and Stan Zin, 1989, Substitution, risk aversion and the temporal behavior of
consumption and asset returns: A theoretical framework, Econometrica 57, 937–969.
Fama, Eugene F., and Kenneth R. French, 2001, Disappearing dividends: Changing firm charac-
teristics or lower propensity to pay? Journal of Financial Economics 60, 3–43.
Gabaix, Xavier, 2012, An exactly solved framework for ten puzzles in macro-finance, Quarterly
Journal of Economics 127, 645–700.
Gorton, Gary, and Andrew Metrick, 2012, Securitized banking and the run on repo, Journal of
Financial Economics 104, 425–451.
Gourio, François, 2013, Credit risk and disaster risk, American Economic Journal: Macroeconomics
5, 1–34.
Huang, Jing-Zhi, and Ming Huang, 2012, How much of the corporate-Treasury yield spread is due
to credit risk? Review of Asset Pricing Studies 2, 153–202.
Kelly, Bryan T., Gerardo Manzo, and Diogo Palhares, 2016, Credit-implied volatility, Working
paper, University of Chicago.
Lemmon, Michael L., Michael R. Roberts, and Jaime F. Zender, 2008, Back to the beginning:
Persistence and the cross-section of corporate capital structure, Journal of Finance 63, 1575–
1608.
Lettau, Martin, and Jessica A. Wachter, 2007, Why is long-horizon equity less risky? A duration-
based explanation of the value premium, Journal of Finance 62, 55–92.
Liu, Jun, Jun Pan, and Tan Wang, 2005, An equilibrium model of rare-event premia and its
implication for option smirks, Review of Financial Studies 18, 131–164.
Longstaff, Francis A., and Monika Piazzesi, 2004, Corporate earnings and the equity premium,
Journal of Financial Economics 74, 401–421.
Longstaff, Francis A., and Arvind Rajan, 2008, An empirical analysis of the pricing of collateralized
debt obligations, Journal of Finance 63, 529–563.
Moreira, Alan, and Alexi Savov, 2016, The macroeconomics of shadow banking, Working paper,
New York University and Yale University.
Ou, Sharon, David Chiu, and Albert Metz, 2011, Corporate default and recovery rates, 1920–2010,
Special Comment, Moody’s Investors Service.
Reinhart, Carmen M., and Kenneth S. Rogoff, 2009, This Time is Different: Eight Centuries of
Financial Folly (Princeton University Press, Princeton, NJ).
Rietz, Thomas A., 1988, The equity risk premium: A solution, Journal of Monetary Economics 22,
117–131.
Seo, Sang Byung, 2017, CDS-implied risk of economic catastrophes, Working paper, University of
Houston.
Seo, Sang Byung, and Jessica A. Wachter, 2018, Option prices in a model with stochastic disaster
risk, Management Science, https://doi.org/10.1287/mnsc.2017.2978.
Shaliastovich, Ivan, 2015, Learning, confidence, and option prices, Journal of Econometrics 187,
18–42.
Tsai, Jerry, 2016, Rare disasters and the term structure of interest rates, Working paper, University
of Oxford.
Tsai, Jerry, and Jessica A. Wachter, 2015, Disaster risk and its implications for asset pricing,
Annual Review of Financial Economics 7, 219–252.
Wachter, Jessica A., 2013, Can time-varying risk of rare disasters explain aggregate stock market
volatility? Journal of Finance 68, 987–1035.
Weil, Philippe, 1990, Nonexpected utility in macroeconomics, Quarterly Journal of Economics 105,
29–42.
Zhou, Chunsheng, 2001, The term structure of credit spreads with jump risk, Journal of Banking
& Finance 25, 2015–2040.
Supporting Information
Additional Supporting Information may be found in the online version of this
article at the publisher’s website:
Appendix S1: Internet Appendix.
Replication Code.

11、Do Rare Events Explain CDX Tranche Spreads（2018JF）

Uploaded by

Copyright:

Available Formats

11、Do Rare Events Explain CDX Tranche Spreads（2018JF）

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

11、Do Rare Events Explain CDX Tranche Spreads（2018JF）

Uploaded by

Copyright:

Available Formats

THE JOURNAL OF FINANCE • VOL. LXXIII, NO.

Do Rare Events Explain CDX Tranche Spreads?

Wachter is at the Wharton School, University of Pennsylvania. We thank Pierre Collin-Dufresne;

using state prices derived from five-year options.

a free parameter that undergoes an unanticipated and permanent shift.

disaster with probability λt .7 Given a disaster, the change in consumption (as a

In Section I of the Internet Appendix,8 we show that the pricing kernel is

In the special case of time-additive utility, γ = 1 and bλ = bξ = 0. The only risk

B. The Aggregate Market

where μd = φμc + 12 φ(1 − φ)σc2 .

In equilibrium, the price of the dividend claim is determined by the cash

where aφ (τ ), bφλ (τ ), and bφξ (τ ) satisfy ODEs given in Section II of the

dividend claims and the effect of exposures to risks of varying frequencies.

We calculate normalized put prices from (11) and implied volatilities as in

C. Individual Firm Dynamics

bondholders as well. The firm payout is subject to three types of risk:

where μi is defined similarly to μd, that is, μi = φi μc + 12 φi (1 − φi )σc2 . The sys-

λ Si ,t = w0 + wλ λt + wξ ξt . For parsimony, we will calibrate the simpler model

where μ Ai ,t is the asset drift rate, which is determined in equilibrium.

E. CDX Tranche Pricing

The protection buyer for tranche j makes quarterly premium payments. As in

min{nt,s − Lt,s , 1 − K j−1 } − min{nt,s − Lt,s , 1 − K j }

To maintain an (approximately) fixed-maturity contract, a new series for the

A. The Stochastic Discount Factor

Panel A: Parameters for Utility, Consumption, and Dividends

Relative risk aversion γ 3

Panel B: Parameters for the Disaster Probability Process

Mean reversion κλ 0.20

Panel C: Population Statistics for the Disaster Probability λt

The resulting parameter choices are reported in Table I. The parameters κλ

No-Disaster Simulations All Simulations

Data 0.05 0.50 0.95 0.05 0.50 0.95 Population

E[Rb ] 1.25 1.68 2.96 3.46 −0.47 2.41 3.37 2.02

measurement of kurtosis: the true (population) kurtosis in the model is 9, while

B. Firm Cash Flows

Default boundary AB 19.2%

The resulting fit to the term structure of physical default probabilities is

No-Disaster Simulations All Simulations

Data 0.05 0.50 0.95 0.05 0.50 0.95

Panel A: Options Sample (January 1996 to December 2012)

E[λ] 1.57 0.05 0.69 4.44 0.06 1.01 7.36

Panel B: CDX Sample (October 2005 to September 2008)

E[λ] 0.74 0.00 0.43 7.60 0.00 0.51 9.02

Upfront (%) Annual Spread (bps)

0% to 3% 3% to 7% 7% to 10% 10% to 15% 15% to 30% CDX

Panel A: Precrisis (October 2005 to September 2007)

Panel B: Crisis (October 2007 to September 2008)

Data 54 498 255 136 69 116

Panel C: Full Sample (October 2005 to September 2008)

Data 39 238 102 54 27 67

(2001), Culp, Nozawa, and Veronesi (2014)).

Upfront (%) Annual Spread (bps)

0% to 3% 3% to 7% 7% to 10% 10% to 15% 15% to 30% CDX

Panel A: Precrisis (September 2004 to September 2007)

Panel B: Crisis (October 2007 to September 2008)

Data 43 364 168 87 48 –

Panel C: Full Sample (September 2004 to September 2008)

B. Time Variation in Spreads

Super Senior Tranche Spread (Annual bps)

Three-Year Five-Year Three-Year Five-Year