Professional Documents
Culture Documents
Research Article
Valuing Catastrophe Bonds Involving Credit Risks
Copyright © 2014 Jian Liu et al. This is an open access article distributed under the Creative Commons Attribution License, which
permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Catastrophe bonds are the most important products in catastrophe risk securitization market. For the operating mechanism, CAT
bonds may have a credit risk, so in this paper we consider the influence of the credit risk on CAT bonds pricing that is different
from the other literature. We employ the Jarrow and Turnbull method to model the credit risks and get access to the general pricing
formula using the Extreme Value Theory. Furthermore, we present an empirical pricing study of the Property Claim Services data,
where the parameters in the loss function distribution are estimated by the MLE method and the default probabilities are deduced
by the US financial market data. Then we get the catastrophe bonds value by the Monte Carlo method.
1. Introduction bonds and options. Moreover, for the highly skewed property
of the catastrophe risk distribution, valuing CAT bonds has
The securitization of catastrophe risk springing up in the become very complicated.
early 1990s has created a direct link between the insurance The main pricing models of CAT bonds, including Kreps
industry and capital market. There are a variety of catastrophe model [2], LFC model [3, 4], Christofides model [5], and
risk securitization instruments, such as options, swaps, and Wang two-factor model [6], are all using the quantitative
bonds. Catastrophe bonds (CAT bonds), the largest issued methods to estimate essential factors of the price and then
and most successful instrument among them, have topped price the CAT bonds with them. Zimbidis et al. [7] use
$6 billion in 2013 and are set to be the highest since 2007, Extreme Value Theory to get the numerical results of CAT
according to Artemis, a deal tracker [1]. CAT bonds are bonds prices under stochastic interest rates in an incomplete
usually insurance-linked and meant to raise money in case market framework. Z.-G. Ma and C.-Q. Ma [8] derive a
of a catastrophic event such as a hurricane or earthquake, so bond pricing formula under stochastic interest rates with
that insurance companies can hedge their exposure by trans- the losses following a compound nonhomogeneous Poisson
ferring catastrophe risk to a wide pool of willing investors. process and find the numerical solution for the price of
Compared with other fixed income asset classes, CAT bonds catastrophe risk bonds. Li et al. [9] study a representative
offer high yield as the total returns were almost 9.5% in 2013, agent-pricing model of the multievent CAT bonds by the
according to a Swiss Re index [1]. Furthermore, CAT bonds data of catastrophic insured property losses of typhoon in
are not closely linked with the stock market or economic China. Based on the idea of layered pricing, Xiao and Meng
conditions, so they offer significant attractions to investors. [10] use the extreme value model to discuss the pricing
The development of CAT bonds market depends on the of the excess-of-loss reinsurance premium with different
reasonable prices, so the scientific pricing is the key problem attachment points. Nowak and Romaniuk [11] use Monte
to the field of CAT bonds research. As a kind of catastrophe Carlo simulation method to price the CAT bonds with
risk securitization product,, the value of CAT bonds results different payoff functions. However, most of the literature
from the probability of the catastrophe risk and the loss in in CAT bonds pricing research does not take the credit
the catastrophe, for CAT bonds have the dual properties of risk influence into account. Actually, the credit risk has the
2 Mathematical Problems in Engineering
probability of existence for the operating mechanism of CAT An increasing filtration F𝑡 ⊂ F, 𝑡 ∈ [0, 𝑇]. The investor
bonds. So it is necessary to fully consider the credit risk in receives the predetermined interest at the end of each period
valuing research, which can improve the pricing validity. and the interest of the current period and the whole principal
This paper presents a pricing model of CAT bonds with in time 𝑇. The amount of the interest or the principal depends
credit risks and conducts an empirical analysis of the US on whether the loss of the catastrophic event exceeds the
catastrophe market data using the Extreme Value Theory, trigger level. In this paper, when considering the credit risk
where the parameters in the loss function distribution are of the CAT bond, the investment income also depends on the
estimated by the MLE method and the default probabilities probability of default at each period. Suppose the probability
are deduced by the US financial market data. Furthermore, of default at period 𝑖 is 𝜆 𝑖 and the recovery rate is a constant 𝜃.
based on the theoretical pricing formula, we get the CAT Let 𝐻𝑖 denote the payment amount of the CAT bond at period
bonds value by the Monte Carlo method. 𝑖 which can be expressed as follows.
When 1 ≤ 𝑖 ≤ 𝑛 − 1,
3.2. Credit Risk. This paper uses the methodology of Jarrow 500.00
and Turnbull [14] to model the credit risk of the CAT bond.
The credit risk can be described by the probability of default
400.00
and the recovery value which are deduced by the arbitrage-
free valuation techniques. This methodology is simple and
Adjusted loss
fits the existing term structure of interest rate well. As the 300.00
absolute priority rule is often violated and lots of other
factors affect the payoff of bonds, such as the percentage
200.00
of managerial ownership [14], modeling the actual payoff
in default is a complicated problem. Therefore, we take the
recovery rate in the event of default when the CAT bond does 100.00
not trigger as an exogenously given constant. The recovery
rate is denoted by 𝜃 and it is assumed to be the same for all
0.00
bonds in a given credit ranking. Then, we can use the risk-
free interest rate to discount the cash flow of the CAT bond 1985 1990 1995 2000 2005 2010 2015
but not use the discount rate involving risk premium [15]. In Year
fact, the probability of default has reflected the credit risk of
the CAT bond. Figure 1: Scatter plot of the annual maximum magnitude of
Suppose the probability of default in time interval [𝑖−1, 𝑖] catastrophe losses in the USA (1985–2011).
is denoted by 𝜆 𝑖 , and let 𝜆∗𝑖 = −(1/𝑖) ∑𝑖𝑡=1 ln(1 − 𝜆 𝑡 ); then we
∗
get ∏𝑖𝑡=1 (1 − 𝜆 𝑡 ) = 𝑒−𝜆 𝑖 ⋅𝑖 . If 𝜆 𝑖 is small, then we can obtain
It is defined on the set {𝑧 : 1 + 𝜉(𝑧 − 𝜇)/𝜎 > 0}, where the
∑𝑖𝑡=1 𝜆𝑡 scale parameter satisfies 𝜎 > 0, and the trail index satisfies
𝜆∗𝑖 ≈ . (6) −∞ < 𝜉 < ∞ and location parameter satisfies −∞ < 𝜇 <
𝑖 ∞. If we got the estimated values of the three parameters,
It means that 𝜆∗𝑖 is the average intensity of default. the distribution function of the maximum loss amount in a
The interest rate of the risk bond in time interval [𝑖 − 1, 𝑖] catastrophic event 𝐼𝑖 can be gotten [17].
is denoted by 𝑟𝑖∗ and the risk-free interest rate is 𝑟𝑖 . For the There are different kinds of considerable estimation
no-arbitrage principle, parameters 𝑟𝑖 , 𝑟𝑖∗ , 𝜃, 𝜆∗𝑖 meet the techniques for three parameters {𝜎, 𝜉, 𝜇}, such as General-
∗ ized Method of Moments, the graphical techniques based
equation 𝑒−𝑟𝑖 = [1 ⋅ (1 − 𝜆∗𝑖 ) + 𝜃 ⋅ 𝜆∗𝑖 ]𝑒−𝑟𝑖 , and then we get
on versions of probability plots, and maximum likelihood
∗ estimation (MLE) method. The MLE method is a classical
1 − 𝑒𝑟𝑖 −𝑟𝑖
𝜆∗𝑖 = . (7) estimation technique, and it is effective for the estimations
1−𝜃 of parameters {𝜎, 𝜉, 𝜇} studied by many researchers [18–20].
Consequently, we can obtain the average intensity of default We employ MLE method to estimate parameters. Suppose
𝜆∗𝑖 by (6) and deduce the probability of default in each time that the loss amount processes 𝐼1 , 𝐼2 , . . . , 𝐼𝑛 are independent
interval [𝑖 − 1, 𝑖] which is 𝜆 𝑖 , 𝑖 = 1, 2, . . . , 𝑛. variables following GEV distribution; then the log-likelihood
function for parameters 𝜇, 𝜎, 𝜉 is
3.3. Distribution of Loss Function. The loss function 𝐼𝑖 is the 1 𝑛 𝐼 −𝜇
maximum loss amount of the catastrophic event at period 𝑖 𝐿 (𝜎, 𝜉, 𝜇) = 𝑛 log 𝜎 − (1 + ) ∑ log [1 + 𝜉 ( 𝑖 )]
and 𝐼𝑖 = max{𝑋1𝑖 , 𝑋2𝑖 , . . . , 𝑋𝑚𝑖 }, where 𝑚 is the number of 𝜉 𝑖=1 𝜎
days at each period and 𝑋1𝑖 , 𝑋2𝑖 , . . . , 𝑋𝑚𝑖 is the sequence of (10)
𝑛 −1/𝜉
loss amount at each period. They are independent random 𝐼 −𝜇
− ∑ [1 + 𝜉 ( 𝑖 )] ,
variables with a common unknown distribution function. If 𝑖=1 𝜎
there exist sequences of constants {𝑎𝑘 : 𝑎𝑘 > 0, ∀𝑘 ∈ N},
{𝑏𝑘 }𝑘∈N and a nondegenerate distribution function 𝐺(𝑧), such where 1 + 𝜉((𝐼𝑖 − 𝜇)/𝜎) > 0, as 𝑖 = 1, 2, . . . , 𝑛.
that
𝐼𝑘 − 𝑏𝑘 4. Numerical Analysis
𝑃( ≤ 𝑧) → 𝐺 (𝑧) , as 𝑘 → ∞, 𝑧 ∈ R, (8)
𝑎𝑘
4.1. Data and Parameter Estimations. The analysis is based
then by Fisher-Tipptt Gnedenko Theorem, 𝐺 is a member of on the catastrophe data from Property Claim Services (PSC).
the generalized extreme value (GEV) family of distributions The data covers all losses resulting from natural catastrophic
or von Mises Type Extreme Value distribution or the von events in the USA over the period 1985–2011, with 3628
Mises-Jenkinson type distribution [16] and its distribution original data. We get the series of annual maximum magni-
function form can be denoted as tude of catastrophe losses in the USA. In order to facilitate
comparison, the data is adjusted for inflation given the time
𝑧 − 𝜇 −1/𝜉 value of the capital, using the Consumer Price Index (CPI)
𝐺 (𝑧) = exp {−[1 + 𝜉 ( )] } . (9)
provided by the US Department of Labor. Figure 1 shows the
𝜎
4 Mathematical Problems in Engineering
Skewness Kurtosis
Sample size (𝑛) Minimum Maximum Mean Standard deviation
Statistic Std. error Statistic Std. error
27 0.168 4.7337 7.046 10.697 2.539 0.448 7.193 0.872
adjusted annual maximum magnitude of losses. As described Table 2: Risk-free interest rates and risk interest rates.
in Table 1, the mean of catastrophe losses is 7.046 billion, the Time 1 2 3 4 5
standard deviation is 10.697 billion, the skewness is 2.54, and
Risk-free 0.17% 0.39% 0.74% 1.18% 1.61%
the kurtosis is 7.19. It means that the data is right-skewed and
tail-dispersed. Risk 3.78% 5.24% 6.83% 7.09% 7.35%
Assume that the data is independent random variables
with GEV distribution function. By the extRemes program Table 3: Default rates of risk bonds.
package in R software, we get the Maximum Likelihood Time period 0-1 (𝜆 1 ) 1-2 (𝜆 2 ) 2-3 (𝜆 3 ) 3-4 (𝜆 4 ) 4-5 (𝜆 5 )
estimations of parameters in distribution function (10) as Default rate 5.92% 9.86% 13.77% 8.72% 8.23%
follows:
(𝜇, 𝜎, 𝜉) = (16.22, 20.25, 1.02) . (11) of the loss amount; that is to say, 𝑀 = 7.046 billion dollars.
The covariance matrix of the three parameters estimations is The obtained ratio of the interest and the principal when the
CAT bond triggers is 𝑘 = 0.8. Then, we get the probability of
23.40 25.99 −0.53 trigger 𝑃(𝐼 > 𝑀) = 0.2404 by expression (9) substituting the
cov = [ 25.99 38.36 0.14 ] . (12) above-mentioned parameters. Then by (9), we deduce
[−0.53 0.14 0.11 ] −𝜉
𝑧 = 𝜇 + [𝜎 (1 − (− log 𝐺 (𝑧)) )] . (13)
Figure 2 shows the various plots for assessing the accuracy
of the GEV model, fitted to the annual maximum magnitude As the function 𝐺(𝑧) is the distribution probability, it meets
of catastrophe losses of US data. Each set of plotted points 0 ≤ 𝐺(𝑧) ≤ 1. So by generating random numbers on
is near-line in the probability plot and the quantile plot, so the interval [0, 1], we simulate the variable 𝑧 employing the
the fitted model is valid. The respective estimated curve in Monte Carlo method with the number of simulation paths 𝑙 =
the return level plot is close to a straight line and the corre- 10000. For expression (4), we get the pricing result of the CAT
sponding density plot seems consistent with the histogram bond which is 𝑉 = 118.86 dollars by the Matlab software. In
of the data. Therefore, the estimated GEV distribution fits the same way, the price of fixed income instrument without
the real data well and the GEV model, deduced by the MLE credit risk is calculated as 139.27 dollars with the same interest
method, is acceptable and valid. Then, we use the estimations rates. The former is much less than the latter that attributes to
of the model parameters to calculate the probability of the the risk of catastrophe and the credit risk. For a higher risk,
catastrophe loss by (9). the CAT bond gives a higher yield.
Now, we consider the credit risk. The credit ratings of
most CAT bonds are BB, while a CAT bond rated AA was
issued in 2006 for the first time. So we assume that the
5. Conclusions
credit rating of the CAT bond is BB. Furthermore, we take As catastrophes are small probability and high loss events
the treasury rates in the US market as the risk-free interest and present high positive correlation among individuals,
rate and take the corporation bonds rates as the risk rates. the traditional insurance company and reinsurance company
Suppose the recovery rate 𝜃 = 40% when the CAT bond may not spread risk completely. However, the securitization
defaults. Then, according to the US market data in July, of catastrophe risk brings an effective solution to transfer and
2013, the treasury rates of 6 months, 2 years, 3 years, and 5 spread the catastrophe risk. In order to develop the CAT bond
years are 0.06%, 0.39%, 0.74%, and 1.61%,respectively, and the market, it is necessary to make effective pricing to CAT bond
corresponding rates of corporation bonds rated BB are 3.05%, which is the most mature instrument in catastrophe securi-
5.24%, 6.83%,and 7.35%, respectively. We get the treasury tization at the present. To make the pricing result accurate,
rates and corporation bonds rates from the risk-free interest the credit risk in the CAT bond should not be ignored. This
rates and the risk interest rates, respectively, over the time paper builds the model of the CAT bond involving the credit
period of 1–5 years by the linear interpolation method as risk. This model has the characteristics of an analytic method
Table 2 shows. Moreover, we get the default rates of risk bonds and numerical method and has a good practical feasibility. We
in each period by expressions (6) and (7), just as Table 3 employ the Jarrow and Turnbull method to model the credit
shows. risk and get access to the general pricing formula using the
Extreme Value Theory. Furthermore, we present an empirical
4.2. Pricing Analysis. Consider a CAT bond with the duration pricing study of the Property Claim Services data where the
𝑛 = 5 years, the face value 𝐹 = 100 dollars, and the coupon parameters in the loss function distribution are estimated by
rate 9%. The amount of trigger loss is assumed to be the mean the MLE method and the default probabilities are deduced
Mathematical Problems in Engineering 5
0.8 400
0.6 300
Model
Empirical
0.4 200
0.2 100
0.0 0
0.0 0.2 0.4 0.6 0.8 1.0 0 100 200 300 400 500 600
Empirical Model
0.020
15000
Return level
f(z)
0.010
5000
0 0.000
Figure 2: Diagnostic plots for GEV fit to the annual maximum magnitude of catastrophe losses in the USA.
by the US financial market data. Consequently, we get the of China (no. 12YJC630118), and the Hunan Social Science
catastrophe bonds value by the Monte Carlo method which Planning Project of China (no. 11YBA009).
is lower than the price of fixed income instruments.
Further research is directed to extend the model in more
complex situations, for example, given suitable stochastic References
process to describe the trigger amount of loss, and it will [1] http://finance.sina.com.cn/world/mzjj/20131018/171517040051.
perfect the model which we have discussed. shtml.
[2] R. Kreps, “Investment-equivalent reinsurance pricing,” in Pro-
Conflict of Interests ceedings of the Casualty Actuarial Society (PCAS ’98), vol. 85,
May 1998.
The authors declare that there is no conflict of interests [3] M. N. Lane, Price, Risk, and Ratings for Insurance-Linked
regarding the publication of this paper. Notes: Evaluating Their Position in Your Portfolio, Derivatives
Quarterly, 1998.
Acknowledgments [4] M. N. Lane and O. Y. Movchan, “Risk cubes or price risk and
ratings (Part II),” Journal of Risk Finance, vol. 1, no. 1, pp. 71–86,
The authors would like to express their gratitude to the 1999.
support given by the Natural Science Foundation of China [5] S. Christofides, Pricing of Catastrophe Linked Securities, ASTIN
(no. 71201013, no. 71171024, and no. 71371195), the Humanities Colloquium International Actuarial Association, Brussels, Bel-
and Social Sciences Project of the Ministry of Education gium, 2004.
6 Mathematical Problems in Engineering
International
Journal of Journal of
Mathematics and
Mathematical
Discrete Mathematics
Sciences