The influence of natural catastrophes on the CAT bond market

(1)

University of Amsterdam

Department of Economics and Business

The influence of natural catastrophes on the

CAT bond market.

Economics and Finance | BSc Thesis 2018

Lotte van Leersum | 11050748

lotte.vanleersum@student.uva.nl

Supervisor: drs. R. Sperna Weiland

(2)

2 Statement of Originality

This document is written by Lotte van Leersum who declares to take full responsibility for the contents of this document.

I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

1) Introduction

The year 2017 has been a big test for the catastrophe bond market (Basak, S., & Kochkodin, B., 2018). Big natural catastrophes like hurricane Harvey, Irma and Maria in 2017 caused global insured losses of US$ 92 billion. Global economic losses even reached a staggering 337 billion dollars, the second highest on record (Swiss Re, 2018). In spite of major insured losses in 2017, there was still an increase in demand for insurance-linked securities such as catastrophe bonds (CAT bonds) (Trottier and Van Son Lai, 2017). The CAT bond market only exists since the early 1990s and is growing fast these days. CAT bonds are an interesting economic development, since economic and insured losses have increased significantly over time (Carayannopoulos and Perez, 2015). Through CAT bonds, the catastrophe risk for insurers and reinsurers is transferred to the financial market. The CAT bond market is interesting for investors, since CAT bonds give a high risk premium for catastrophe risk and since the bonds provide high diversification possibilities (Carayannopoulos and Perez, 2015). These high diversification possibilities are due to the low correlation of these assets with the market, which is provided through the CAT bond structure. Gürtler et al. (2014) and, Carayannopoulos and Perez (2015) showed that CAT bonds correlation with the market can be affected by big events such as the subprime crisis 2008-2009 and hurricane Katrina 2005. The behaviour of CAT bonds on such kind of events is therefore important information to investors.

In earlier research, the effect of natural catastrophes on the correlation of CAT bonds with the market was mostly limited to two sample periods: hurricane Katrina in 2005 and hurricane Ike (where results were mostly due to the financial crisis) in 2008. One of the main reasons for investors to invest in CAT bonds, is due to the low correlation with the market. Therefore, it is important for investors to know when, and to what extent, CAT bonds give diversification possibilities. In this paper, I want to create a more extensive insight on the effects of natural catastrophes on the CAT bond market in the United States for investors. Consequently, the research question that will be central in this paper is the following:

What was the effect of big trigger events during the five insured losses peaks, in the United States during 2004-2017, on the diversification possibilities of the CAT bonds in comparison to corporate bonds with the same credit risk?

This question will be analysed using a data set from 2003 to 2017. To assist in answering the research question, three hypotheses will be created and discussed. A regression analysis and time-varying analyses will be performed to study the effect of big trigger events on the diversification possibilities. The main findings resulting from the analyses are the following: The analyses show that the natural catastrophes do not lead to a consistent effect on the

(5)

5 diversification possibilities of the CAT bond, since two of the betas increased and three decreased during the sample periods. However, despite the increases and decreases found, the beta is very close to zero during the entire period. Therefore, CAT bonds can be seen as a good source of diversification. Further, the BB CAT bond provides higher diversification possibilities, during the entire period, than the BB corporate bond.

As additional test, the correlation and hedge ratios of the CAT bond in relation with the corporate bond will be estimated. The results show an increasing correlation, during the five trigger events, between the two bonds. Thus, when holding a portfolio consisting of CAT bonds and BB corporate bonds, investors should be aware of the effect of trigger events on the diversification possibilities.

The remainder of this thesis is structured as follows. Section 2 focusses on understanding the CAT bond structure through a theoretical framework. This will be done through discussing the effect of the financial crisis on the bonds, discussing the pricing and premiums of CAT bonds and examining the diversification possibilities of the bonds. After this, in section 3, the three hypotheses will be introduced which are created to assist in answering the research question. In section 4 a description of the data and the methodology are given. In section 5 the results of the OLS regression analysis and the time-varying analyses will be discussed. For the time-varying analyses, the rolling correlation and rolling beta will be used. Further, will the rolling correlation and dynamic hedge ratios for CAT bonds in relation with corporate bonds be obtained and discussed as well in this section. Finally, in section 6 the conclusion and some study limitations will be stated.

(6)

6

2) Catastrophe bonds

Global economic and insured catastrophe losses have increased significantly over time (Carayannopoulos and Perez, 2015). An interesting economic development, is the transfer of catastrophe risk for insurers and reinsurers to the financial market due to CAT bonds. The CAT bond is a type of insurance-linked security (ILS). This type of bond is interesting for investors because of its reputation for a relatively high risk-return profile, and for the low correlation of the CAT bond with the market and other asset classes (Braun, 2015). Since the introduction of CAT bonds to the financial market in the early 1990s, they have steadily grown (Artemis, 2018). As shown in figure 1; the CAT bond market grew from US$785 million outstanding in 1997 to US$31 billion outstanding in 2017. There was only a decrease of US$2 billion in capital outstanding in the years 2008 to 2010, this was due to the Subprime financial crisis in 2008-2009 (Carayannopoulos and Perez, 2015).

Figure 1.

Note: the amounts on the y-axis are in millions of US$.

To get a better understanding of the CAT bond structure, I will now discuss it more thoroughly. A visual representation of the structure is shown in figure 2. The insurer or reinsurer gets in a reinsurance agreement with a special purpose vehicle (SPV) so they do not have to issue bonds directly to the capital market (Carayannopoulos and Perez, 2015). The insurer is now protected against catastrophe losses (Braun, 2015). The SPV then issues bonds to investors on the capital market. The proceeds are placed in a trust account and are then used to purchase high rated short-term assets and collateral, such as short-term treasury bonds or AAA-rated bonds (Carayannopoulos and Perez, 2015). To protect the collateral in the trust account, the fixed

0 5000 10000 15000 20000 25000 30000 35000 40000 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 20 08 20 09 20 10 20 11 20 12 20 13 20 14 20 15 20 16 20 17 20 18

(7)

7 returns from the trust account are swapped by the swap counterparty for floating returns mostly based on the London Interbank Offered Rate (LIBOR) (Cummins and Weiss, 2009). Because of the swap, the insurer and investors are immunized from interest rate risk and default risk (Cummins and Weiss, 2009). When a trigger event occurs, CAT bond investors could lose a part, or even all, of their principal investment (Cummins and Weiss, 2009). The SPV then releases proceeds to the insurer, so they can pay the claims that arose from the event. In return for the risk, the insurer pays a premium to the investors. The investors get a regular coupon for bearing the risk, which consists of the variable interest rate (LIBOR) and a spread which is the risk premium (Braun, 2015). When no trigger event occurred, the principal will be returned as well at the expiration of the bond.

Figure 2.

*Coverage is only given to an insurer when a catastrophic event occurred *Principal is only repaid to the investor when no catastrophic event occurred

Because of this structure with the swap counterparty, the investors are solely exposed to the risk of the underlying trigger event. The CAT bonds therefore provide high diversification possibilities. However, research of Cummins and Weiss (2009) and Carayannopoulos and Perez (2015) showed declining diversification possibilities during the subprime financial crisis 2008-2009. During this crisis, some limitations of the CAT bond structure arose (Cummins and Weiss, 2009). The creditworthiness of some swap counterparties was affected and there were problems with the assets in the collateral trust account (Carayannopoulos and Perez, 2015). Lehman Brothers served as a swap counterparty for some CAT bonds, which in September 2008 became insolvent. Another swap counterparty was not a possibility for these CAT bonds, since the assets in the collateral trust account were also reduced in value (Carayannopoulos and Perez, 2015). Carayannopoulos and Perez (2015) explain, that this decrease was a consequence of the structured products that were included in the trust account,

(8)

8 such as mortgage-backed securities (MSB) and asset-backed securities (ABS). These structured products were impacted by the financial crisis since they had artificially high credit ratings. Another problem was that the duration of most of the assets were longer than the duration of the CAT bonds (Braun, 2015).

After the financial crisis it was clear that more transparency and regulation was necessary concerning the CAT bonds credit and collateral issues (Cummins and Weiss, 2009). Illiquid and hard-to-price structured assets, such as ABS and CDOs, were not allowed anymore as an investment possibility in the trust account (Carayannopoulos and Perez, 2015). Furthermore, there was increased monitoring and mark-to-market requirements to ensure all the criteria were satisfied (Braun, 2015). Because of this, the credit quality of the collateral assets and the creditworthiness of the swap counterparties improved. Some CAT bonds do not even use a swap counterparty anymore and the trust accounts consists of money market funds, which invests in short-term sovereign debt, such as T-bills (Braun, 2015). An example of a popular asset that is allowed, because of the liquid characteristic of this bond, is the put option by sovereign or quasi-governmental entities (Carayannopoulos and Perez, 2015).

2.1) Pricing

A lot of research has been done on the pricing of CAT bonds. So did for example Bodoff and Gan (2009) describe the market clearing issuance price as a linear function of expected loss. Lane and Mahul (2008), and, Dieckmann (2011) and Galeotti et al. (2013) conducted empirical study on which factors have an influence on CAT bond risk premiums, however, these researches all used small data sets (Gürtler et al., 2014). Braun (2015) in addition, proposed an econometric pricing model for CAT bonds in the primary market with a big sample size. He concluded that the major drivers for the CAT bond spread are the expected loss, the covered territory, the reinsurance cycle, the sponsor and the BB corporate bond spread. These findings are widely accepted in academia (Lane, 2000; Lane and Mahul, 2008; Dieckmann, 2010; Galeotti et al., 2013). To get a better understanding of which factors influence the market premium, Gürtler et al. (2014) used a large sample set of secondary data of CAT bond premiums, instead of primary data as Braun (2015) did. These findings were corresponding with the findings of Braun (2015).

Cummins and Weiss (2009), Braun (2015) and Gürtler et al. (2014) their research on the risk premium of the CAT bonds showed a decreasing ratio in risk premium to expected loss for CAT bonds over the years. Froot (2001) explains, that the significant reduction in expected loss premium over time is due to better supply of the CAT bonds nowadays, while at the beginning of this market there were supply restrictions. Another explanation according to Froot (2001) is that the CAT bond market used to have market imperfections but became more

(9)

9 efficient. Lei et al. (2008) and Dieckmann (2009) found, that the CAT bonds premiums are typically far higher than the premiums for corporate bonds with the same credit risk. Braun (2015) explains, that the higher risk premium of CAT bonds compared to corporate bonds can be due to higher complexity of the CAT bonds structure, lower liquidity of CAT bonds and investors may demand a higher risk premium since they are not as used to catastrophe risk as to traditional market risk.

The effect of a certain trigger type for CAT bonds on the risk premium has been widely discussed in literature (Lane and Mahul, 2008; Cummins and Weiss, 2009; Dieckmann, 2010; Galeotti et al., 2013). There are four main trigger types for CAT bonds: 1) Indemnity triggers, the payoff on the bond is determined by the actual losses of the issuing insurer; 2) Industry-loss index triggers, the payoff is determined by the value of industry-wide Industry-losses; 3) Modelled loss triggers, the payoff is based on the estimated loss index that is predicted by catastrophe modelling firms based on the specific event parameters that occurred; 4) Parametric triggers, the payoff on the bond is triggered when a specified predetermined physical measure of the catastrophic event occurs, such as the scale of an earthquake or wind speed of a hurricane (Lei et al., 2008). When the CAT bond has an indemnity trigger, the bond is based on insurer-specific losses, which exposes the bond to moral hazard. It is therefore commonly accepted in academia that CAT bonds trading with indemnity triggers require an additional risk premium (Cummins and Weiss, 2009; Dieckmann, 2010; Galeotti et al., 2013). However, Lei et al. (2008) and Gürtler et al. (2014) did not find any significant influence of the indemnity trigger variable on the CAT bonds premium. This could result from a special payment structure of CAT bonds with the indemnity trigger, which provides an incentive provision (Cummins and Weiss, 2009). This structure reduces problems that arise from asymmetric information (Gürtler et al., 2014).

2.2) Diversification

As shown in section 2.1, the pricing of CAT bonds is discussed extensively in literature, there is, however, less research conducted concerning the diversification possibilities of these bonds. One of the reasons for the growing interest of investors in the CAT bonds market, is due to these diversification possibilities. Since the occurrence of natural catastrophes is independent from market development, the correlation of CAT bonds with the asset market is very low or even zero (Gürtler et al., 2014). Galeotti et al. (2013) did not find any correlation of the CAT bond market with the S&P 500, based on a data set from 1999-2008. Dieckmann (2011) however, did find some correlation between CAT bond returns and the capital market with a data set between 2002 and 2011. However, Gürtler et al. (2014) discussed that the financial crisis, during 2008-2009, could be the reason for this higher correlation. Gürtler et al (2014)

(10)

10 and Carayannopoulos and Perez (2015) examined if CAT bonds are a good diversification source in the context of the 2008-2009 Subprime financial crisis and during hurricane Ike (which occurred at the same time as the bankruptcy of Lehman Brothers) and Katrina (2005). Gürtler et al. (2014) used a big data set of secondary market CAT bond premiums from 2002 to 2012, to do research on the CAT bond premiums and their correlation with corporate bond spreads during hurricane Katrina and the fall of Lehman Brothers (2008). They found a significant dependence between corporate spreads and CAT bond premiums, and state that CAT bonds cannot be seen as zero-beta assets. Carayannopoulos and Perez (2015) used secondary data from 2002 to 2013, to compare hedge ratios of CAT bonds with hedge ratios of BB-rated corporate bonds and government bonds. They did this pre- and post- the financial crisis and hurricane Katrina. Their main findings are that CAT bonds had a correlation with the market and there were significant hedge ratios during the financial crisis. They describe that this could be due to the bad creditworthiness of the trust account and the decreasing value of the collateral, as described earlier in this paper. However, after the financial crisis the correlation returned to the zero pre-crisis level (Carayannopoulos and Perez, 2015). Carayannopoulos and Perez (2015) also found that hurricane Katrina significantly increased the correlation of CAT bonds with the market. Nevertheless, they conclude that the impact of natural catastrophes have been small compared to the impact of the financial crisis. Another important finding that they stressed out, is that the hedge ratios of CAT bonds are very small compared to those of corporate bonds and government bonds during the financial crisis. Because of this, the CAT bonds are a better diversification instrument for investors than these other financial assets (Carayannopoulos and Perez, 2015).

Both Gürtler et al (2014) and Carayannopoulos and Perez (2015) limited their research on the risk premium and correlation of CAT bonds during natural catastrophes hurricane Katrina and Ike (where the results were mostly due to the financial crisis). It is important for investors to know to what extent, and when, CAT bonds give diversification possibilities. In this paper, to get a more elaborated insight, five periods where natural catastrophes occurred during 2004 and 2017 will be used to test the diversification possibilities pre- and post- this period in comparison with BB-rated corporate bonds.

(11)

11

3) Hypotheses

To get a better understanding on the effects of natural catastrophes on the diversification possibilities of CAT bonds, three hypotheses are composed. These hypotheses will be tested for all the sample periods.

To know what the effect of the natural catastrophes was on the diversification possibilities of CAT bonds, the period prior the catastrophe should be analysed as well. Braun (2015) did not find any significant correlation between CAT bond returns and the market during normal market conditions. The period before the Catastrophes can be seen as normal market conditions. Because of this, I expect to find a zero-beta on the periods before the trigger events. Based on this expectation the first hypothesis is proposed:

Hypothesis 1: The beta of CAT bonds with the market pre-trigger event is zero Gürtler et al. (2014) and Carayannopoulos and Perez (2015) showed that hurricane Katrina influenced the CAT bonds risk premium and its correlation with the market. Because of this I expect the beta of the CAT bonds to increase during big natural catastrophes. Therefore, the following hypothesis is assumed:

Hypothesis 2: The beta of CAT bonds with the market pre-trigger events is smaller than the beta during the trigger events.

Carayannopoulos and Perez (2015) found that the hedge ratio of CAT bonds is very small compared to those of other asset classes. So, despite the expectation of an increasing beta for CAT bonds during the trigger events, I still expect the corporate bonds to be more correlated with the market. This is stated in the following Hypothesis:

(12)

12

4) Data and Methodology

In this research, weekly data obtained from Datastream is used. For all data, there are 784 weekly observations starting from 29-May-2003 till 31-May-2018. For the CAT bond data, one of the Swiss Re indices will be used. Swiss Re maintains four total return insurance-linked securities indices: an overall ILS index, a BB-rated index, a California earthquake index and a U.S. windstorm index. The broadest indices are the ILS index and the BB-rated index, which are both highly correlated with each other. Braun (2015) shows that the majority of the CAT bonds exhibit a BB-rating. Following Cummins and Weiss (2009) and Carayannopoulos and Perez (2015) in this research the BB-rated index will be used. Specifically, weekly data from the BB-rated CAT bond total return index. Carayannopoulos and Perez (2015) showed that the dynamic behaviour of the CAT bonds returns of the overall index, the ‘BB CAT bond total return index’ and the ‘BB CAT bond price index’ are very similar, so all of the indices can be used. For the corporate bonds ‘the Bank of America Merill Lynch BB U.S. high yield total return index’ is used. This index includes all securities with a BB-rated investment grade, issued in the U.S. domestic market. For the overall market the ‘Standard & Poors 500 (S&P500) total return index’ is used, which consists of the 500 biggest American companies. Five periods, where timewise two or more catastrophes occurred relatively close to each other, that created the highest insured losses from 2004 till 2017, will be used in this research. On Artemis and Aon Benfield, the following big natural catastrophes are provided with their amount of insured losses: Hurricane Katrina, Rita and Wilma August-October 2005 (US$87billion) (economist), Hurricane Ike and Gustav Augustus-September 2008 (US$21.585 billion), severe weather April-June 2011 (US$29,5 billion), Hurricane Sandy and Isaac August and October 2012 (US$34 billion), Hurricane Harvey, Irma and Maria August and September 2017 (US$47,7 billion).

(13)

13

Table 1.

CAT Bond

returns Corp. Bond returns S&P500 returns Descriptive statistics Mean 0.11 0.14 0.21 Maximum 5.51 4.85 16.79 Minimum -5.69 -5.53 -16.19 Std. Dev. 0.48 0.78 2.19 Observations 783 783 783

Five periods effect:

2005 Mean before (01/2005-07/2005) 0.12 0.14 0.83 Std. dev. before 0.06 0.63 1.31 Mean during (08/2005-12/2005) -0.24 -0.03 0.06 Std. dev. during 0.84 0.57 1.28 2008 Mean before (01/2008-07/2008) 0.13 -0.02 -0.36 Std. dev. before 0.06 0.53 2.16 Mean during (08/2008-12/2008) -0.13 -1.03 -1.43 Std. dev. during 0.94 2.43 7.14 2011 Mean before (09/2010-03/2011) 0.02 0.25 0.83 Std. dev. before 0.69 0.39 1.57 Mean during (04/2011-06/2011) 0.14 0.12 0.14 Std. dev. during 0.21 0.30 1.70 2012 Mean before (01/2012-07/2012) 0.11 0.27 0.33 Std. dev. before 0.20 0.44 1.55 Mean during (08/2012-12/2012) 0.17 0.25 0.15 Std. dev. during 0.60 0.40 1.73 2017 Mean before (01/2012-07/2012) 0.04 0.19 0.50 Std. dev. before 0.17 0.39 1.00 Mean during (08/2012-12/2012) 0.07 0.06 0.43 Std. dev. during 1.74 0.27 0.77

Note: The descriptive statistics of return data is given in percentage points and the mean returns are on a weekly basis.

The table represents the mean and standard deviation, for the entire period and the before and during periods of the five trigger events, of the CAT bond returns, Corporate bond returns and S&P500 returns.

(14)

14

4.1) Descriptive statistics

The descriptive statistics of the returns are given in table 1. The mean, maximum, minimum and standard deviations are shown for the entire period. Also, to see if there are changes in returns, the mean and standard deviation before and during the five periods are given. Table 1 shows that CAT bonds have the lowest standard deviation, for the entire period, in comparison with the BB-rated corporate bonds and the market. This can be evidence for CAT bonds having an advantage in being a more stable asset. When comparing the results of Carayannopoulos and Perez (2015) of the mean of returns, before and during the periods of 2005 and 2008, similar results are found. As expected, the catastrophes had a negative impact on the returns. However, hurricane Ike took place at the same time that Lehman Brothers became insolvent. Gürtler et al. (2014) have shown that most of the effect was due to the financial crisis, but hurricane Ike also contributed some. The other sample periods during the trigger events of 2011, 2012 and 2017 showed increasing mean returns for the CAT bonds. This is against the expectations. Conclusively, while the descriptive statistics show that CAT bonds are pretty stable assets, since they have a lower standard deviation over the entire period for the CAT bonds compared to the other assets, it is still not clear, by looking at the returns per period, what the effect of natural catastrophes are.

4.2) Regression analyses

To get more insight on the diversification possibilities of the CAT bonds, an OLS regression will be done for the CAT bonds and corporate bonds with the market, before and during the five periods with the highest insured losses. With the regression the estimated beta will be obtained. The following regression analyses will be used:

𝑟𝐴 = 𝛼 + 𝛽𝑟𝑀+ 𝜀𝑖 (1)

For the assets, the CAT bonds and the corporate bonds will be used. The beta is a ratio of the covariance of the asset with the market divided by the variance of the market. The Measure is used to determine the volatility of an asset in relation to the market. A beta of one means that the asset moves exactly the same as the market does. When the asset has a zero-beta or even a negative-beta, the asset does not move or even decreases (relatively) when the market goes up. Hypothesis 1, if the betas of CAT bonds before the trigger events are zero, can be tested with the p-value obtained from the regressions results. To test if the beta of CAT bonds with the market pre-trigger events is smaller than the beta during the trigger events, a Chow-test will

(15)

15 be executed. The Chow-test is a test for equality or a structural-break between sets of coefficients in two linear regressions (Toyoda, 1974). The formula for the Chow-test is (Booij, 2016, p. 67):

𝐹 =

(𝑆𝑆𝑅𝑡−(𝑆𝑆𝑅1+𝑆𝑆𝑅2))/2

(𝑆𝑆𝑅1+𝑆𝑆𝑅2)/(𝑛−4) (2)

Where SSR stands for Sum of Squares Residuals. SSRt stands for the combined regression line, SSR1 is the regression line before the breaking-point and SSR2 is the regression line after the breaking-point. The SSR’s will be obtained from the regression analyses.

4.3) Time varying analyses

The beta’s obtained from the regressions have the assumption that they are constant over time. However, financial time series seems to exhibit volatility clustering (Carayannopoulos and Perez, 2015). To tackle this problem, the time-varying relation of the assets will be observed as well. This will be done by using the rolling correlation and rolling beta. First, the rolling correlation will be performed between the CAT bonds and the corporate bonds with the market, to get a first thought about the relation of these bonds with the market. The rolling correlation estimator is an estimator for the conditional correlation and is defined as follows (Carayannopoulos and Perez, 2015):

𝜌12,𝑡= ∑ 𝑟1,𝑠𝑟2,𝑠 𝑡−1 𝑠=𝑡−𝑛−1 √(∑ 𝑟1,𝑠2) 𝑡−1 𝑠=𝑡−𝑛−1 (∑ 𝑟1,𝑠 2₎ 𝑡−1 𝑠=𝑡−𝑛−1

The rolling window chosen, will determine the time span on which the correlation coefficient is based. The correlation lies in the interval [-1, 1]. A correlation of one means they are perfect correlated. A correlation of zero, or minus one, means that there is respectively no correlation or even a perfect negative correlation.

To get a better insight about what the effect is of big natural catastrophes on the diversification possibilities of CAT bonds, the time-varying beta will be used as well. As described before, the beta is a ratio of the covariance of the asset with the market divided by the variance of the market. The rolling window will determine the time span on which the beta coefficients are based. To stay consistent, the same rolling window will be chosen for the correlation and beta. Hypothesis 3, if the betas of the CAT bonds are higher than those of the corporate bonds, will be tested by using the rolling beta.

(16)

16 As final analysis, the hedge ratios of CAT bonds in relation with BB-rated corporate bonds will be observed. This is interesting since some investors will hold a diversified portfolio, consisting of corporate bonds and CAT bonds (Carayannopoulos and Perez, 2015). Because of this, it is important to know the correlation and hedge ratios between these two assets, and if natural catastrophes have an impact on it. The hedge ratio of the CAT bond is estimated as the ratio of the covariance with the corporate bond and the variance of the corporate bond. The rolling correlation and time-varying hedge ratios will be obtained for the CAT bond in relation with the corporate bond. If the correlation and the hedge ratios are low, investing in the CAT bonds can lead to high diversification possibilities for investors already holding BB-rated corporate bonds.

5) Results

5.1) Regression analysis

To get a first thought about the three hypotheses, the betas of the CAT bond index and the corporate bond index will be analysed. The regression as described in equation 1 will be used for this. Table 2 gives an overview of the OLS regression analyses, done for the CAT bond with the market and for the corporate bond with the market. As described earlier, the proxies for the assets are the BB-rated CAT bond total return index, the Bank of America Merill Lynch BB U.S. high yield total return index and the S&P500 total return index. In table 2 the beta, p-value and SSR are given for: the entire period (29-May-2003 till 31-May-2018), the periods pre-trigger events, the periods during the trigger events and for the pre- and during trigger event periods combined.

To test our first hypothesis, the p-values of the betas of the before periods will be analysed. All the betas of the ‘before period’ of the CAT bonds are not significantly different from zero with a significance level of 10%. This corresponds with the expectation and hypothesis 1, that the beta of the CAT bond index with the market, pre-trigger event, is zero. This confirms Braun’s (2015) results for no significant correlation of CAT bonds with the market during normal market conditions.

For the second hypothesis, the beta before and during the trigger events will be compared to each other, and a Chow-test will be performed to see if there is a structural break. It is expected that the betas during the trigger event periods are larger than before the trigger events. Gürtler et al. (2012) gave two reasons for a potential increase in correlation, and thus the beta, between CAT bond markets and the capital market. First, the CAT bond market can

(17)

17 change due to a natural catastrophe, and the damage of the catastrophe can also affect the overall economy and thus the capital market. Second, a huge crisis such as the financial crisis has a big impact on the capital market. It may also affect investors risk aversion and therefore affect the CAT bonds prices. When comparing the betas before and during the trigger events of the CAT bonds in table 2, in 2005, 2011 and 2012 the beta increased during the catastrophe period, and in 2008 and 2017 the beta decreased during the catastrophe period. However, all the betas during the trigger events are not significantly different from zero with a significance level of 10%. These results are consistent with the results of Carayannopoulos and Perez (2015) who found increasing, but not significant, correlation during the catastrophes in 2005, 2011 and 2012. However, the results differ during the financial crisis and hurricane Ike and Gustav in 2008. The beta analysed is smaller and not significantly different from zero, while Carayannopoulos and Perez (2015) found a significant positive correlation coefficient of 0.29. A possible reason for a decreasing correlation during the financial crisis, could be a shift of investors from the capital market to the CAT bond market. This would be due to the fact that CAT bonds are known for not having traditional market risk, and the capital market actually was at risk. However, this substitution effect is not confirmed, and no further research will be conducted about it in this paper. The decreasing beta obtained for 2017 cannot be compared with earlier results, since comparable researches did not include 2017 in their datasets. When performing the Chow-test, as defined in equation 2, only during hurricane Katrina, Rita and Wilma (2005) the betas differ significantly from each other with F(2,48)=2.704 and a significance level of 10%. The betas during the trigger events of 2008, 2011, 2012 and 2017 do not significantly differ from the beta before the trigger events. For a significance level of 5%, all the results from the Chow-test become not significant. These findings are against hypothesis 2, where a significant break was expected for the betas during the trigger events. With these preliminary results in mind, the catastrophic events do not seem to have a large effect on the diversification possibilities of the CAT bonds.

To get a first thought about hypothesis 3, the beta of the CAT bond index will be compared with the betas of the corporate bond index. Hypothesis 3 states that the betas of the CAT bond are smaller than the betas of the corporate bond. When looking at the entire period in table 2, the CAT bond has a beta of -0.01 which is not significant different from zero and the corporate bond has a significant beta of 0.16. Table 2 shows, that during all the periods before and during the trigger events the beta of the BB-rated corporate bond index is larger than the beta of the BB-rated CAT bond index. These results confirm hypothesis 3. This suggests that the CAT bond gives better diversification possibilities than the corporate bond does. Summarizing: three out of the five betas were higher during the catastrophic events, however, they failed to reach significant levels. During 2008 and 2017 the beta decreased, these betas also failed to reach significant levels. Performing the Chow-test with a significance

(18)

18 level of 10%, only the catastrophic events in 2005 caused a significant structural break for the beta during the trigger events in comparison with the beta before the trigger event. With a 5% significance level, none of the betas during the catastrophes showed a structural break in comparison with the betas before the catastrophes. Further, CAT bond betas seem to be smaller than corporate bond betas. This together, can be preliminary evidence for CAT bonds being a good source of diversification.

(19)

19

Table 2.

CAT bond Corp. bond CAT bond Corp. bond Entire period Before (09/2010-3/2011)

Beta -0,01 0,16 Beta -0,03 0,11 P-value 0,12 0,00 P-value 0,71 0,01 SSR 0,0181 0,0380 SSR 0,0014 0,0004 N 783 783 2005 During (04/2011-06/2011) Beta 0,01 0,15 Beta 0,03 0,07 P-value 0,86 0,02 P-value 0,40 0,19 SSR 0,0017 0,0017 SSR 0,0000 0,0001 N 52 52 2012 Before (01/2005-07/2005) Beta 0,01 0,10 Beta -0,01 0,22 P-value 0,83 0,00 P-value 0,24 0,01 SSR 0,0009 0,0007 SSR 0,0000 0,0009 N 52 52 During (08/2005-12/2005) Before (01/2012-07/2012) Beta 0,04 0,06 Beta 0,00 0,11 P-value 0,81 0,53 P-value 0,94 0,04 SSR 0,0015 0,0007 SSR 0,0001 0,0005 2008 During (08/2012-12/2012) Beta -0,02 0,15 Beta 0,02 0,10 P-value 0,24 0,00 P-value 0,83 0,04 SSR 0,0018 0,0113 SSR 0,0007 0,0003 N 52 52 2017 Before (01/2008-07/2008) Beta -0,15 0,16 Beta 0,01 0,10 P-value 0,40 0,00 P-value 0,21 0,02 SSR 0,0063 0,0005 SSR 0,0000 0,0007 N 52 52 During (08/2008-12/2008) Before (01/2017-07/2017) Beta -0,03 0,15 Beta 0,00 0,16 P-value 0,35 0,05 P-value 0,89 0,03 SSR 0,0017 0,0097 SSR 0,0001 0,0004 2011 During (08/2017-12/2017) Beta -0,02 0,10 Beta -0,50 0,15 P-value 0,77 0,00 P-value 0,32 0,04 SSR 0,0015 0,0005 SSR 0,0060 0,0001 N 44 44

The OLS regression, as described in equation 1, is done for the CAT bond returns and corporate bond returns in relation with the S&P500 returns. The betas, p-values and Sum of Square Residuals are given for: the entire period (29-May-2003 till 31-May-2018), the periods pre- and during the trigger events and for the period of pre- and during trigger events combined.

(20)

20

5.2) Time-varying analyses

The betas obtained from the regression in table 2 are constant over time. This is a restrictive assumption, since financial time series seem to exhibit volatility clustering (Carayannopoulos and Perez, 2015). To get rid of this assumption, a time-varying analysis will be performed. To get a time-varying analysis, the rolling correlation as described in equation 3 is estimated. A rolling window of 52 weeks will be used, the same as Carayannopoulos and Perez (2015) used in their research. Figure 3 shows the rolling correlation coefficients for the BB-rated CAT bond and the BB-rated corporate bond with the market. Gürtler et al. (2012) discussed, that a catastrophe can have an impact on both the CAT bond market and the capital market. When this is true, the correlation between the CAT bond and the market will increase during a catastrophe. Figure 3 shows an increasing correlation during the trigger events in 2005 and 2012, however, the coefficient is still below 0.2. In 2008, at the start of the during period in August, there is first a little increase in correlation, but then it drops to a correlation of -0.2. During the trigger events of 2011 and 2017, a decreasing correlation coefficient is obtained. Since two of the sample periods showed an increasing correlation coefficient, and three sample periods showed a decreasing correlation coefficient, a clear effect of natural catastrophes cannot be concluded yet. When comparing the rolling correlation of the CAT bond with the rolling correlation of the corporate bond, it can be said that the corporate bonds are more correlated with the market than CAT bonds are.

(21)

21 To get a more refined result of the diversification possibilities and the effect of natural catastrophes on it, the time-varying betas are obtained. In figure 4, the estimated rolling beta of the CAT bond and the corporate bond with the market are obtained. Again, a rolling window of 52 weeks is used. The figure shows that the beta of the CAT bond is very close to zero during the entire period. However, during our sample periods of 2005 and 2012, the beta coefficient increases to 0.03 and 0.01 respectively. During the other sample periods, there is a decrease in the coefficient, which is consistent with the rolling correlation results. In 2017 the beta even decreased to -0.15. This large decrease in beta could be due to the effect of hurricane Irma (September, 2017). The threat of hurricane Irma to Florida made the CAT bond price return index drop by 16% (Artemis, 2017). While the CAT bond market declined with 16%, the S&P500 did not decrease and even kept increasing steadily. Therefore, a negative beta arose. As seen in figure 4, the natural catastrophes do not have a consistent effect on the beta of the CAT bonds. When comparing the betas of the CAT bond to the betas of the corporate bond, the betas of the CAT bond are significantly lower during the entire period (with exception of May 2007, where the beta of the corporate bond is a fraction lower than the beta of the CAT

-0,4 -0,2 0 0,2 0,4 0,6 0,8 1

Jan-05 Jan-06 Jan-07 Jan-08 Jan-09 Jan-10 Jan-11 Jan-12 Jan-13 Jan-14 Jan-15 Jan-16 Jan-17 Jan-18

Rolling correlation - Rolling window 52 weeks

BB-rated CAT bond BB-rated corporate bond

Figure 3. Rolling correlations of the CAT bond and corporate bond.

The solid lines represent the rolling correlation coefficient of the BB CAT bond index returns and BB corporate bond index return with the returns of S&P500 index, as defined in equation 3. The dotted lines represent the starting points of the five major trigger events during the sample periods: August 2005, Hurricane Katrina, Rita and Wilma (US$87billion); August 2008, Hurricane Ike and Gustav (US$21.585 billion); April 2011, severe weather (US$29,5 billion); August 2012, Hurricane Sandy and Isaac (US$34 billion); August 2017, Hurricane Harvey, Irma and Maria (US$47,7 billion).

(22)

22 bond). This corresponds with the third hypothesis, the beta of CAT bonds being lower than the beta of the corporate bonds with the same amount of risk.

As a final test, the correlation and hedge ratios of the CAT bond index, in comparison with BB-rated corporate bond index, will be analysed. This is interesting, since investors may want to hold a diversified portfolio consisting of BB-corporate bonds. If there is a low beta for the CAT bond in relation to the corporate bond, the CAT bond can be used to hedge the negative effects of the market on the corporate bonds (Carayannopoulos and Perez, 2015). Figure 5 shows the rolling correlation between the CAT bond and the corporate bond. Again, a rolling window of 52 weeks is used. All five sample periods show an increasing correlation between the two assets. During the financial crisis, and hurricane Ike and Gustav, it even reached a correlation coefficient of almost 0.48. During 2011 and 2017 the correlation increases, but still remains below 0.1. To get a better insight on the diversification possibilities, the dynamic hedge ratio is estimated. A rolling window of 52 weeks is used, the results are

-0,2 -0,1 0 0,1 0,2 0,3 0,4 0,5

Rolling beta - Rolling window 52 weeks

BB CAT bond BB corporate bond

Figure 4. Rolling Beta of the CAT bond and corporate bond.

The solid lines represent the rolling beta coefficient of the BB CAT bond index returns and BB corporate bond index return in relation with the returns of S&P500 index. The dotted lines represent the starting points of the five major trigger events during the sample period: August 2005, Hurricane Katrina, Rita and Wilma (US$87billion); August 2008, Hurricane Ike and Gustav (US$21.585 billion); April 2011, severe weather (US$29,5 billion); August 2012,

Hurricane Sandy and Isaac (US$34 billion); August 2017, Hurricane Harvey, Irma and Maria (US$47,7 billion).

(23)

23 shown in figure 6. During all the five periods, the hedge ratios between the CAT bond and corporate bond increased. During the trigger events in 2005 and 2008 the hedge ratios even increased towards 0.3. These results are in line with Gürtler et al. (2014) their finding of a positive dependence between CAT bond premiums and BB-rated corporate bond premiums. Gürtler et al. (2014) gave the ‘flight to quality’ in extreme market conditions as an explanation for this, where investors move to safer assets as a consequence of increased risk aversion. Both of the assets are affected by this, since the CAT bonds and BB-rated corporate bonds exhibit the same credit rating. However, after 2005 and 2008, the trigger events seem to have a smaller effect on the hedge ratios. This could be due to the improved structure of CAT bonds after the financial crisis, and since CAT bonds became more competitively priced over the years (Cummins and Weiss, 2009). Even though not all the catastrophes had a huge impact on the hedge ratios, investors should still be aware of the increasing correlation between the CAT bonds and BB-rated corporate bonds during big trigger events or financial crises.

-0,3 -0,2 -0,1 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7

Rolling correlation - Rolling window 52 weeks

CAT bond - corporate bond

Figure 5. Rolling correlation of the CAT bond in relation with the corporate bond.

The solid line represents the rolling correlation coefficient of the BB CAT bond index returns in relation with the BB corporate bond index return, as defined in equation 3. The dotted lines represent the starting points of the five major trigger events during the sample period: August 2005, Hurricane Katrina, Rita and Wilma (US$87billion); August 2008, Hurricane Ike and Gustav (US$21.585 billion); April 2011, severe weather (US$29,5 billion); August 2012,

Hurricane Sandy and Isaac (US$34 billion); August 2017, Hurricane Harvey, Irma and Maria (US$47,7 billion).

(24)

24 -0,2 -0,1 0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8

Rolling Hedge ratio - Rolling window 52 weeks

CAT bond - corporate bond

Figure 6. Rolling hedge ratios of the CAT bond with the corporate bond.

The solid line represents the estimated dynamic hedge ratios of the BB CAT bond index return with respect to the BB corporate bond index return. The hedge ratios are estimated as the covariance of the two assets divided by the variance of the corporate bond return. The dotted lines represent the starting points of the five major trigger events during the sample period: August 2005, Hurricane Katrina, Rita and Wilma (US$87billion); August 2008, Hurricane Ike and Gustav (US$21.585 billion); April 2011, severe weather (US$29,5 billion); August 2012, Hurricane Sandy and Isaac (US$34 billion); August 2017, Hurricane Harvey, Irma and Maria (US$47,7 billion).

(25)

25

6) Conclusion

Over the years, CAT bonds have gained increasing popularity by market participants. The idea of this asset class is transferring catastrophe risk to the capital market, while it provides high risk premiums and diversification possibilities to investors. However, it is important for investors to know if CAT bond behaviour is stable or may change during certain events. In this paper, a regression analysis and time-varying analyses were performed to study the effect of big trigger events, during the five insured losses peaks from 2004 till 2017 in the United States, on the diversification possibilities of CAT bonds, in comparison to corporate bonds with the same credit risk.

With the results obtained from the regression analyses, it is shown that none of the betas were significantly different from zero. The Chow-test showed that, with a significance level of 5%, all the betas during the trigger events of the five sample periods did not significantly differ from the beta before the trigger events. Furthermore, all of the betas from the corporate bond during the trigger events were significantly higher than the betas of the CAT bond. These results led to preliminary evidence, and showed that CAT bonds were not significantly affected by the natural catastrophes, and their diversification possibilities are high in comparison with corporate bonds.

To get rid of the assumption that the betas are constant over time, time-varying analyses were conducted as well. The analysis of the rolling correlation and rolling beta did not lead to a consistent effect of the natural catastrophes. While during 2005 and 2012 the correlation and beta increased, during the other three sample periods they decreased. Still, during the entire period the betas were very close to zero. When comparing the betas of the corporate bond to the beta of the CAT bond, the CAT bond is considerably lower correlated with the market than the corporate bonds are.

Conclusively, the natural catastrophes do not show a clear effect on the diversification possibilities of CAT bonds. Further, since the CAT bonds betas stay very close to zero during the entire dataset, and are significantly lower than the betas of the corporate bonds, they can be seen as assets that provide high diversification possibilities and should not be ignored by investors. However, the five trigger events did affect the correlation and hedge ratio of the CAT bond index in relation to the corporate bond index. Thus, when holding a portfolio consisting of BB-rated corporate bonds, investors should be aware of the effect of the trigger events on the diversification possibilities. Yet, to what extent they are correlated and the precise reasons for it, would be a subject for future research.

(26)

26

6.1 Study limitations

In this paper the rolling correlation and rolling beta are used. However, a shortcoming of the rolling correlation estimator is that all the observations, less than n, are equally weighted, and older observations get zero weight (Carayannopoulos and Perez, 2015). In future research, I therefore suggest using a model that takes this limitation into account. Furthermore, Figure 3 shows an increasing correlation in 2005 and 2012 and a decreasing correlation in 2008, 2011 and 2017. Additional studies should be conducted to get a deeper understanding about why the correlation increased or decreased during these periods.

(27)

27

References

Aonbenfield (2018, June 12). Insured losses. Retrieved from http://catastropheinsight.aonbenfield.

com/pages/regionschart.aspx?region=US&losstype=Insured

Artemis (2018, June 12). Catastrophe bond & ILS risk capital issued & outstanding by year. Retrieved from http://www.artemis.bm/deal_directory/cat_bonds_ils_issued_outstanding.html Artemis (2017, September 9). Cat bonds drop 16% on hurricane Irma, prices discounted heavily.

Retrieved from http://www.artemis.bm/blog/2017/09/09/cat-bonds-drop-16-on-hurricane-irma-prices-discounted-heavily/

Basak, S., & Kochkodin, B. (2018, January 4). Even 2017 Couldn’t Crack the $90 Billion Disaster-Bond Market. Retrieved from https://www.bloomberg.com/news/articles/2018-01-04/even-2017-couldn-t-crack-the-90-billion-disaster-bond-market

Bodoff, N. M., & Gan, Y. (2009, May). An analysis of the market price of cat bonds. In CAS E-Forum. 101

Booij, A., (2016) Syllabus Research Project. Amsterdam School of Economics, University of Amsterdam.

Braun, A. (2016). Pricing in the primary market for cat bonds: new empirical evidence. Journal of

Risk and Insurance, 83(4), 811-847.

Carayannopoulos, P., & Perez, M. F. (2015). Diversification through catastrophe bonds: lessons from the subprime financial crisis. The Geneva Papers on Risk and Insurance-Issues and

Practice, 40(1), 1-28.

Cummins, J. D., & Weiss, M. A. (2009). Convergence of insurance and financial markets: Hybrid and securitized risk‐ transfer solutions. Journal of Risk and Insurance, 76(3), 493-545.

Dieckmann, S. (2010). By force of nature: explaining the yield spread on catastrophe bonds.

Froot, K. A. (2001). The market for catastrophe risk: a clinical examination. Journal of Financial

Economics, 60(2-3), 529-571.

Galeotti, M., Gürtler, M., & Winkelvos, C. (2013). Accuracy of premium calculation models for CAT bonds—an empirical analysis. Journal of Risk and Insurance, 80(2), 401-421.

(28)

28

Gürtler, M., Hibbeln, M., & Winkelvos, C. (2014). The impact of the financial crisis and natural catastrophes on CAT bonds. Journal of Risk and Insurance, 83(3), 579-612.

Gürtler, M., Hibbeln, M., & Winkelvos, C. (2012). The impact of the financial crisis and natural catastrophes on CAT bonds. Technische Universität Braunschweig, Institure of Finance Working Paper No. IF40V1

Lane, M. N. (2000). Pricing Risk Transfer Transactions 1. ASTIN Bulletin: The Journal of the IAA,

30(2), 259-293.

Lane, M., & Mahul, O. (2008). Catastrophe risk pricing: an empirical analysis.

Lei, D. T., Wang, J. H., & Tzeng, L. Y. (2008). Explaining the Spread Premiums on Catastrophe Bonds. In NTU International Conference on Finance, Taiwan.

Swiss Re (2018, April 10). At USD 144 billion, global insured losses from disaster events in 2017 were the highest ever, sigma study says. Retrieved from http://www.swissre.com/media/news_ releases/nr20180410_sigma_global_insured_loses_highest_ever.html

Toyoda, T. (1974). Use of the Chow test under heteroscedasticity. Econometrica: Journal of the

Econometric Society, 601-608.

The influence of natural catastrophes on the CAT bond market