Decomposition of natural catastrophe risks: Insurability using parametric CAT bonds

Marvi, Morteza Tavanaie; Linders, Daniël Article Decomposition of natural catastrophe risks: Insurability using parametric CAT bonds Risks Provided in Cooperation with: MDPI – Multidisciplinary Digital Publishing Institute, Basel Suggested Citation: Marvi, Morteza Tavanaie; Linders, Daniël (2021) : Decomposition of natural catastrophe risks: Insurability using parametric CAT bonds, Risks, ISSN 2227-9091, MDPI, Basel, Vol. 9, Iss. 12, pp. 1-19, https://doi.org/10.3390/risks9120215 This Version is available at: https://hdl.handle.net/10419/258297 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ risks Article Decomposition of Natural Catastrophe Risks: Insurability Using Parametric CAT Bonds Morteza Tavanaie Marvi 1,*,† and Daniël Linders 2   Citation: Tavanaie Marvi, Morteza, and Daniël Linders. 2021. Decomposition of Natural Catastrophe Risks: Insurability Using Parametric CAT Bonds. Risks 9: 215. https://doi.org/10.3390/risks9120215 Academic Editor: Mercedes Ayuso Received: 24 October 2021 Accepted: 17 November 2021 Published: 1 December 2021 Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. Copyright: © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/). 1Civil & Environmental Engineering Department, University of Illinois at Urbana-Champaign, Champaign County, IL 61801, USA 2Amsterdam School of Economics, University of Amsterdam, Roeterstraat 11, 1001 NJ Amsterdam, The Netherlands; [email protected] *Correspondence: [email protected] † Current address: Newmark Civil Engineering Laboratory, MC-250, 205 North Mathews Ave., Urbana, IL 61801, USA. Abstract: Nat Cat risks are not insurable by traditional insurance mainly because of producing highly correlated losses. The source of such correlation among buildings of a region subject to a natural hazard is discussed. A decomposition method is proposed to split Nat Cat risk into idiosyncratic (and hence insurable) risk and systematic risk (carrying the correlated part). It is explained that the systematic risk can be transferred to capital markets using a set of parametric CAT bonds. Premium calculation is presented for insuring the decomposed risk. Portfolio risk-return trade-off measures for investing on the parametric CAT bond are derived. Multi-regional and multi-hazard parametric CAT bonds are introduced to reduce the risk of the investment. The methodology is applied on a region with about 3000 residential buildings subject to flood hazards. Keywords: Nat Cat risk; insurability; CAT bond; risk decomposition; parametric bond; correlated risk; systematic risk; catastrophe risk management 1. Introduction Natural Catastrophe (Nat Cat) risks have been argued to be uninsurable as they do not allow the law of large numbers to be exploited for insurance purposes (Grossi 2005). The law of large numbers is a fundamental principle for insurance that requires randomness of the loss occurrence in both time and magnitude. Nat Cat risks violate this criterion by affecting many properties at the same time and causing losses with magnitudes proportional to the event intensity. Thus, the produced losses are highly correlated which violate the condition for the law of large numbers. This correlation issue is the root of creating a heavy tail distribution for Nat Cat losses which means higher maximum possible aggregate loss, higher expected loss, and higher dispersion in the loss distribution. These issues demand insurers to provide large capital (Charpentier and Le Maux 2014), limit their exposure Zanjani (2006), and charge premiums significantly above actuarial fair value (Banks 2005) which ultimately lead to low insurance take-up rates. As traditional insurance failed to provide the necessary capital to cover Nat Cat losses, insurers have invented other instruments to transfer and carry the risk imposed by the nature. Those instruments enable insurers to share risk and capital with other entities. Three major sources of capital include insurers/reinsurers, investment funds, and financial institutions. Reinsurance policies have been used to share risk and capital among peer licensed insurers/reinsurers. A variety of financial instruments have been used to provide insurers a direct access to the capital market and a capability to exchange risk and capital with investment funds and financial institutions. However, only large insurance companies can offer catastrophe coverage, because they have easier access to capital and can pool the risk with independent risks from other regions (Charpentier and Le Maux 2014). Inability Risks 2021,9, 215. https://doi.org/10.3390/risks9120215 https://www.mdpi.com/journal/risks Risks 2021,9, 215 2 of 19 to diversify a risk makes managing it so costly that many insurance companies prefer to limit their exposure as the case for California earthquake risk described by Zanjani (2006). Although the alternative risk transfer instruments have helped insurers to cover Nat Cat risks, their limitations and constraints prevent them from being a reliable solution. The reinsurance process involves some challenges including pricing difficulties, earnings and capital volatility, lack of penetration, and capacity constraints. In reinsurance agreements, historical premiums appear to be 1.5 to 5 times the expected losses for a given layer (Banks 2005). Maintaining large reserves have challenges such as accounting prohibition, tax inefficiency, and conflict of interest with reinsurers’ shareholders (Banks 2005;Doherty 1997;Niehaus 2002). The reinsurance markets have a finite capital at their disposal, and such capital constraint becomes more evident in the market contraction periods after a major disaster such as the Northridge earthquake and Hurricane Andrew (Carpenter 2006;Enz 2002). Instruments providing direct access to capital markets such as catastrophe futures and options, catastrophe bonds, sidecars, and catastrophe equity put options are subject to low trading volume (Jaffee and Russell 1997), low transparency (Cummins 2008), and substantial basis risk (Cummins 2012). In this paper, we investigate the source of correlation among Nat Cat losses. We demonstrate how the correlation is mainly related to the intensity of the event in the region. Based on this understanding, we propose an innovative method to decompose Nat Cat risks into systematic risk (carrying the correlated part) and idiosyncratic risk (insurable part). We propose to transfer the systematic risk into capital markets using a set of parametric CAT bonds. Premium calculation is explained for insuring buildings against a Nat Cat risk using such decomposition and transfer method. Portfolio measures for risk-return trade-off of investing on such parametric CAT bond are derived. Multi-regional and multi-hazard parametric CAT bonds are introduced and shown to be able to reduce the risk of investment for the bond holders. The proposed risk decomposition method is applied to the city of Seaside, Oregon for about three thousand residential buildings subject to flood hazard. The results show that the residual risk after the decomposition is idiosyncratic as expected and follows a normal distribution with an average at zero when aggregating them together. The calculated insurance premium for each building and detailed results for each individual building are presented using an interactive web app. The paper is organized into six sections (this introduction included). The next section is the methodology which discusses insurability issues of Nat Cat risks, explains the risk decomposition method, and introduces the concept of multi-regional and multi-hazard parametric CAT bond. The explanation of the risk decomposition method includes a demonstration of the source of the correlation, elaboration of the risk decomposition method, presentation of the Nat Cat premium calculation, and derivation of the CAT bond portfolio measures. The third section contains an illustrative example that describes how the proposed methodology is applied to a selected region. The purpose of this section is to illustrate how the proposed theory performs in practice. The fourth section presents the results of the illustrative example to show how the decomposition method could transform the risk for the entire region as well as the individual buildings. Furthermore, the results include an estimate of the required premium for the individual buildings based on some assumptions. The fifth section is the discussion of the results, and the last section is the conclusion as the summary of the paper and the findings. 2. Methodology 2.1. Nat Cat Risk and Insurability A type of risk is not insurable unless it meets certain criteria which are referred to as insurability criteria. We categorize the insurance criteria into three classes: actuarial, market, and societal criteria (Biener and Eling 2012). Table 1summarizes these classes with their associated criteria and requirements. If a type of risk passes the requirements of all the criteria, it is considered to be insurable, and insurers may offer coverage for that risk. Risks 2021,9, 215 3 of 19 For a complete discussion on these criteria and their requirements, interested readers are referred to Berliner (1982); Biener and Eling (2012); SwissRe (2005). Table 1. Insurability criteria and their requirements. Insurability Criteria Requirements Actuarial (1) Randomness (of loss occurrence) Measurable and independent (2) Maximum possible loss Manageable (3) Average loss amount and loss frequency Moderate average loss amount and low loss frequency (4) Loss exposure Loss exposure must be large (5) Information asymmetry Moral hazard and adverse selection not excessive Market (6) Insurance premium Cost recovery and affordable (7) Cover limits Acceptable Societal (8) Public policy Consistent with societal values and availability of services (9) Legal restrictions Allow the coverage Nat Cat risks are considered uninsurable as they easily violate the insurability criteria of a risk. Nat Cat losses do not occur independently in both time and magnitude. If a building experiences a loss in an event, it is highly likely that its neighbors also sustain losses of a similar size. An individual building can produce a loss with thousands of dollars in size. The large size of the loss per se does not make the risk uninsurable. For instance, the same building is offered fire coverage from many insurers; while flood coverage might be available from only a few private insurers (if any) at a significantly higher cost. The simultaneity of such large losses is the issue as it makes the average total loss of a Nat Cat and its maximum possible loss to be too large to be manageable by most insurers. In fact, only large insurance companies can offer catastrophe coverage, because they have easier access to capital and can pool the risk with independent risks from other regions (Charpentier and Le Maux 2014). Table 2has comments on the insurability criteria of Nat Cat risks and clarifies their status regarding the satisfaction of the requirements of each criterion. Table 2. Insurability criteria for natural catastrophes. Insurability Criteria Status Comment Actuarial (1) Randomness (of loss occurrence) NOT Satisfied Dependent claim size and simultanous occurrence (2) Maximum possible loss NOT Satisfied Not manageable for most insurers (3) Average loss amount and loss frequency NOT Satisfied Hard to estimate and with a significant high uncertainty (4) Loss exposure NOT Satisfied LLN is not applicable and low take-up rate (5) Information asymmetry Satisfied Insurers have superior understanding about the risk Market (6) Insurance premium NOT Satisfied Premiums are too high (7) Cover limits NOT Satisfied Using limits cannot solve the insurability issue Societal (8) Public policy Satisfied No conflict with societal values (9) Legal restrictions NOT Satisfied Solvency regulations inhibit many insurers to offer coverage Producing positively correlated losses is the main reason that Nat Cat risks become uninsurable. The correlation among the losses does not allow insurers to use the law of large numbers and reduce loss variability around its expected value by selling more policies. In addition, the positive correlation increases the likelihood of observing extremely large losses in an event which creates a heavy-tail loss distribution for a Nat Cat risk. Industry regulations require insurers to maintain a reserve with an amount commensurate with the risk they carry. Thus, a Nat Cat risk with a heavy-tail distribution is unfavorable for insurers to carry as it can easily increase the required reserve by a significant amount. 2.2. Nat Cat Property Loss and Risk Decomposition Natural catastrophes impact society in various forms including financial losses of the order of hundreds of billions of dollars, population dislocation, and loss of lives. The impacts are categorized into direct and indirect as well as tangible and intangible. Direct Risks 2021,9, 215 4 of 19 impacts are those caused directly by the physical forces of the event and happen at the time of the disaster or within a short period of time after. Indirect impacts are caused as the consequence of the direct impacts and may occur in a different time and space than the event. Tangible impacts are those that can be monetized; while, intangible impacts cannot be converted into a monetary value. For example, damage to a flooded building is a direct and tangible impact; loss of life in flood is a direct and intangible impact; delay in transporting goods because of a flooded main road is an indirect and tangible impact; loss of trust in authorities is an indirect and intangible impact. In this paper, we are focusing on direct impacts of Nat Cats on buildings that relate to property and casualty type insurance (P&C). In other words, we are investigating property losses as the result of natural events such as floods, earthquakes, hurricane, etc. The size of the loss for an individual building in a Nat Cat event dependents on the physical characteristics of the building (structural properties, construction material, floor plan, etc.) and the hazard intensity at the building location (magnitude of energy and forces applied to the structure). For a specific type of building (a building with specific physical characteristics), the loss estimation function presents the relationship between the loss size of a building and the intensity of a specific natural hazard at the property location. In a loss estimation function, the loss size is defined as the ratio of the loss to the total value of the property (relative loss). The physical characteristics are captured by categorizing buildings into different classes with a similar response to the hazard. One or more intensity measures of the hazard are used to describe how the loss ratio changes as the intensity increases at the property location. For instance, the loss estimation function for flood is called depth-damage function as the depth of flood at the building location is commonly used as the sole intensity measure for flood damage estimation (Marvi 2020). The depth-damage function describes how flood damage (relative to the property value) increases as the flood water height increases at the building location. Different depth-damage functions have been developed for different types of buildings to capture the role of physical characteristics. Figure 1 illustrates examples of loss estimation functions for flood, earthquake, and hurricane. The loss estimation function of a building is composed of two parts: (1) a deterministic non-decreasing trend which is the function of the hazard intensity at the building location, g(X) ; (2) a stochastic term with a variance that may change by the hazard intensity, σXε . Here, X represents the intensity of a natural hazard that the loss estimation function of the building is constructed for. Since the intensity of a natural hazard varies in a region, note that X is the local evaluation of the intensity measure at the property location. For simplicity, we follow the general assumption that the random variable ε has a standard normal distribution, and σX is the standard deviation of the loss given the intensity of X around its average value or g(X) . Thus, for a specific natural hazard, the building i in a region will have a random loss of Lias Li=gi(Xi) + σXiε, (1) when the intensity measures of the hazard are evaluated as the vector Xi at the property location. 2.2.1. Source of Correlation Intensity of a natural hazard varies spatially, yet is correlated. The variability of the intensity in a region can be described by the physics model governing the natural phenomenon and the event parameters as the initial values and/or boundary conditions of such a model. For example, flood water height, as an intensity of flood, can be determined in an entire region by the governing hydraulic model calibrated for that specific region, and the river water level at the upstream of the region is the only event parameter required in this model. In this example, the intensity measure of the hazard and the event parameter are from the same quantity; both are height and measured by the same unit. However, the intensity measure and the event parameters can be different in number and quantity. The Risks 2021,9, 215 5 of 19 physics model specifies the event parameters required for the estimation of the intensity in a region. As the natural phenomenon occurs with random magnitude, the event parameters representing such magnitude will be a set of random variables. 0123456 Loss Estimation Function (Flood) Flood Depth (m) Relative Loss 0% 20% 40% 60% 80% 100% 120% 140% Average ± 2σ (a) 0.0 0.5 1.0 1.5 2.0 2.5 Loss Estimation Function (Earthquake) Spectral Acceleration (g) Relative Loss 0% 20% 40% 60% 80% 100% 120% 140% Average ± σ (b) 100 120 140 160 180 200 Loss Estimation Function (Hurricane) Peak Gust Wind Speed at 10m Open Terrain (mph) Relative Loss 0% 20% 40% 60% 80% 100% 120% 140% 75th percentile Average 25th percentile (c) Figure 1. Loss estimation functions for different hazards. ( a ) Flood hazard, (adopted from Nofal and van de Lindt 2020). (b) Earthquake hazard, (adopted from Ramirez 2009). (c) Hurricane hazard, (adopted from Vickery et al. 2006). The calibrated hazard model determines the intensity in any part of the region including at building locations using some event parameters. Assume that the hazard model calibrated for a region is known and denoted by a deterministic function D ; the event parameters required for the hazard model are denoted by the vector Θ ; and, the location of the building i in the region is used in the hazard model in the form of the vector zi . Thus, Xican be determined by Xi=D(zi;Θ), (2) where the Xi ’s are all dependent on the event parameters Θ through the deterministic physics model D and the buildings’ fixed geographical location zi . Note that D for any zi ’s is non-decreasing as Θ changes to the event parameters representing a less frequent event (an event with lower exceedance probability). Thus, Xi ’s and consequently gi(Xi) ’s become correlated. Note that the functions gi( . ) are non-decreasing. Furthermore, Xi ’s are non-decreasing as the intensity of the hazard increases. Thus, the correlation among Risks 2021,9, 215 6 of 19 building losses are positive and can be traced back to the event intensity which is captured by parameters in Θ . Given a specific hazard intensity (Xi) , the part gi(Xi) is a deterministic term of the building loss which we call it systematic loss. The stochastic term of the building loss, σXiε , is uncorrelated among the buildings of the region, and we call it residual loss. Thus, the residual risk is idiosyncratic and insurable. The correlated systematic part of the loss for the individual buildings can be aggregated and considered the systematic loss of the region as Ls=∑N i=1gi(Xi) = ∑N i=1gi(D(zi;Θ)) = G(Θ), (3) where Ls represents the systematic loss of the region assuming that there are N buildings in the region (or in our portfolio). Since the location of the buildings ( zi ), their structural characteristics (captured by gi(.) ), and the hazard model of the region (captured by D(.) ) do not change, the Ls is a deterministic function of Θ . So, we can simplify Equation (3) to Ls=G(Θ) . Likewise, the residual loss of the region ( Lr ) is defined by the sum of the residual losses of the buildings in the region as in Lr=∑N i=1σXiε∼ N 0, r∑N i=1σ2 Xi!or N(0, σΘ). (4) Note that Lr is a random variable which is a sum of N normally distributed random variables. Thus, Lr is a normal random variable with E(Lr)= 0 and Var(Lr)=∑N i=1σXi2 . For simplicity, we use σΘ to show the standard deviation of Lr . The total loss of the region ( L ) is the sum of the losses for all the buildings in the region which is equal to the sum of the systematic and the idiosyncratic residual loss of the region as explained in L=∑N i=1Li=∑N i=1gi(Xi) + ∑N i=1σXiε=Ls+Lr. (5) 2.2.2. Risk Decomposition The systematic part of the natural catastrophe risk, Ls , can be transferred to the capital markets by issuing parametric CAT bonds. For this reason, the range of Θ is divided into J mutually exclusive and collectively exhaustive subdomains, Θj for j= 1, . . . , J . The subdomains are constructed in a way that any event in Θk has an annual exceedance probability less than any event in Θh if and only if k>h . The annual exceedance probability of the natural phenomenon is assumed to be known in the region using historical measurements of Θ and statistical estimation methods such as extreme value theory. For the subdomain Θj , a parametric CAT bond Bj is issued with a total principal as its collected asset invested in a trust which is equal to bjdescribed in bj=(E[G(Θ)|Θ∈Θ1],j=1, EG(Θ)Θ∈Θj−bj−1,j6=1, for j=1, . . . , J. (6) Like other bonds, the parametric CAT bond pays interest in the form of coupons; however, the principal and maybe the interest will be withdrawn if the triggering event occurs. Here, we assume that the parametric CAT bond Bj is paying interest with an annual rate of αj , and the whole principal will be withdrawn in case the bond is triggered. An event with the intensity Θ0 triggers the parametric CAT bond Bj , if there exists at least one event in Θj with the annual exceedance probability greater than the annual exceedance probability of Θ0 . Since the annual exceedance probability for Θ is known with a consensus among the involved parties (insurer and investors), and Θ0 is measured fairly quickly by objective third parties (such as government agencies like USGS and NOAA), the involved parties will have an identical recognition in a short time whether the bond has been triggered or not. This increases transparency and helps to release the bond asset quickly to cover the losses from the event. Risks 2021,9, 215 7 of 19 By transferring the systematic risk to the capital markets, the remaining part of the risk (the residual risk) becomes idiosyncratic and hence insurable. Assume that the event with Θ0 triggers the parametric CAT bonds B1 , . . . , Bs . The total fund provided by the triggered bonds equals to b0=∑s i=1bi=E[G(Θ)|Θ∈Θs]. (7) Thus, the remaining part of the loss is [G(Θ0)−b0]+Lr . It is assumed that Θj ’s are small enough that the variability of [G(Θ0)−b0] is negligible compare to the variability of Lr . In other words, the total loss ( L ) is not sensitive to the variability of G(Θ) around its conditional average E[G(Θ)|Θ∈Θs] . As shown in Equation (4), conditional on the realization of Θ , the residual risk has a normal distribution with standard deviation σΘ . Thus, the residual risk, denoted by Lr , has a compound normal distribution as described in Lr∼ N(0, σΘ), Pr(σΘ)is known through Pr(Θ),(8) Instead of using a compound distribution, a conservative or risk averse option is to use the maximum σΘ over possible Θ as the standard deviation of the residual risk Lr . One can add an extra dispersion to σΘ for considering the variability of G(Θ) around its conditional average over Θj ’s intervals. In that case, the modified standard deviation ( σm Θ ) used for the normal distribution of Lris calculated by σm Θ=rσ2 Θ+∑k j=1VarG(Θ)Θ∈Θj, (9) where k is the highest index of the bonds triggered if an event with the intensity of Θ occurs. 2.2.3. Nat Cat Premium Calculation Premium collected from individual buildings needs to cover the costs including a portion of the CAT bond interest, the residual risk, and the underwriting gain. Those costs should be distributed fairly among the individual buildings. The CAT bond interest is partially provided from low-risk investments such as government or AAA corporate bonds and partially from the insureds’ premium. Assume that Ij is the total interest that needs to be provided from insurance premiums for the CAT bond Bj . As the CAT bond is issued to cover the systematic part of the risk, it would be fair if Ij is distributed among the individual buildings proportional to their average systematic risk within Θj . Thus, we define cI ij as the portion of Ijpaid by the property i, and calculate cI ij using cI ij =   E[gi(D(zi;Θ))|Θ∈Θ1] b1,j=1, E[gi(D(zi;Θ))|Θ∈Θj]−E[gi(D(zi;Θ))|Θ∈Θj−1] bj,j6=1, for i=1, . . . , Nand j=1, . . . , J. (10) The residual risk is managed by the insurer based on variety of factors including the industry regulations, market competition, and company’s risk appetite. Based on all factors and the residual risk distribution mentioned in Equation (8), assume that the insurer decides to collect the total of V amount from the insured buildings to cover the residual risk. As the residual risk is more induced by the dispersion of the loss around its average, a fair practice to distribute V among the insureds is to find the contribution of each property to the total dispersion. Thus, we propose to use cV i as the portion of V paid by the property iwhich is explained by cV i=1 PrΘ∈ ∪J 1ΘjE"σ2 Xi σ2 Θ#=1 PrΘ∈ ∪J 1ΘjE"σ2 gi(D(zi;Θ)) σ2 Θ#. (11) Risks 2021,9, 215 8 of 19 The underwriting gain is assumed to be a fixed rate of each policy. Assume that U is the total underwriting gain, and qi is the total portion of the premium paid by the building i for the CAT bond interest and its residual risk. So, the portion of U paid by the building i is denoted by cU iand calculated using cU i=qi ∑N i=1qi qi=∑J j=1cI ij Ij+cV iVfor i=1, . . . , N. (12) Ultimately, the total premium needs to be collected from the building i is the sum of its portions from the CAT bond interest, residual risk, and underwriting gain. The total premium is denoted by Qiand calculated as Qi=∑J j=1cI ij Ij+cV iV+cU iU=qi 1+U ∑N i=1qi!(13) 2.2.4. CAT Bond Portfolio Measures As a risky investment, the parametric CAT bonds are compared with other investment opportunities using their risk-return trade-off. Expected return and standard deviation of the returns are commonly used in risk-return trade-off to compare risky investments and quantify their risk. From two investments with the same expected return, the one with lower standard deviation (lower risk) is more favorable. Likewise, from two investments with the same standard deviation (same risk), the one with higher expected return is more favorable. The parametric CAT bond Bj with a total principal of a generates a stochastic returns with their associated probabilities as R1/β=(a(1+α), 1 −β, −a(1−δα),β.(14) Here, α is the bond’s annual interest rate; δ determines the portion of the interest that is paid if the triggering event occurs. Note that δ= 0 if nothing is paid when the triggering event occurs, and δ= 1 if full interest is paid no matter the triggering event occurs. Finally, β is the highest annual exceedance probability of the events in Bj . This means that any event with an annual exceedance probability less than β will trigger the bond Bj . Thus, the bond Bj can be named as β -bond or, with using return period instead of β , it can be named as 1/β -year bond. For example, Bj corresponding to β= 0.005 is called 200-year bond. One can describe R using a Bernoulli random variable W with P(W=0)=β and P(W=1)=1−βas presented by R1/β=a(1+α)W−a(1−δα)(1−W)=a[2+(1−δ)α]W−a(1−δα). (15) The expected value and standard deviation of the stochastic return R1/β is presented as µR1/β=a[1+α−[2+(1−δ)α]β], σR1/β=a[2+(1−δ)α]pβ(1−β).(16) Thus, the results for special cases of δ= 0 and δ= 1 are derived, respectively, as follows µR1/β=a[1+α−[2+α]β], σR1/β=a[2+α]pβ(1−β).for δ=0 (17) µR1/β=a[1+α−2β], σR1/β=2apβ(1−β).for δ=1 (18) Risks 2021,9, 215 15 of 19 0.0×10+0 5.0×10−9 1.0×10−8 1.5×10−8 2.0×10−8 37.5 46.9 56.3 65.6 75.0 84.4 93.8 103.1 112.5 121.9 131.3 140.6 150.0 159.4 168.8 178.1 187.5 196.9 206.3 215.6 225.0 234.4 243.8 253.1 262.5 271.9 281.3 290.6 300.0 309.4 318.8 328.1 337.5 Total loss ($ millions) Density Histogram of Losses for Entire Region Figure 6. The conditional distribution of flood loss. (flood risk before decomposition) Based on the realized losses, the principal of the parametric CAT bonds were calculated and presented in Table 5. CAT bond interest can be divided into two parts: (1) the interest generated from the bond principle investing on a trust which is governed by risk-free rate, and (2) the spread of the catastrophe bond which is paid to the investors to compensate the involved risk. The bond spread is the portion of the interest that is covered by the insureds’ premium. To calculate this part for each building, an interest rate for each bond’s spread was assumed in this example which is presented in Table 5. The riskier the investment (or the lower the return period of the underlying catastrophe of the bond) the higher the interest. Each bond’s total spread was calculated according to the assumed spread interest rate and the total bond’s principal. The total spread for each bond is presented in Table 5. Table 5. The calculated CAT bond principals. CAT Bond B1B2B3B4 Bond principal (bi) $51.9 million $61.9 million $97.0 million $105.0 million Spread interest rate 4% 3% 2% 1% Bond total spread $2.1 million $1.9 million $1.9 million $1.0 million The residual losses for the simulated 1000 year variants were calculated by subtracting the triggered bi ’s from the total losses. The distribution of the residual losses is presented in Figure 7. The average of the residual losses occurred around zero (the vertical dashed line in Figure 7) and the standard deviation was USD 20.6 million. The red dashed curve in Figure 7shows a fitted normal density to the histogram of the residual losses. The Q-Q plot illustrated in Figure 7verifies the normality of the residual losses. The insurance premium for each building was calculated based on the results of this Monte Carlo simulation. The coefficients cI ij ’s were calculated using Equation (10), and the bonds’ total spread were split and allocated to each building accordingly. We assumed that the total amount collected from the insureds to cover the residual risk is equivalent to the 70% percentile of the residual risk distribution which was V= USD 10.8 million. The coefficients cV i ’s were calculated using Equation (11), and V was distributed among the buildings according to cV i ’s. Using Equation (12), qi ’s were calculated, and an assumed underwriting gain of U= USD 1 million were distributed among the buildings accordingly. Following Equation (13), Qi was derived as the total premium paid by each building. Figure 8presents the histogram of the monthly premiums for each building as a dollar value and a percentage of the property value. The average dollar value of the monthly premiums was USD 230 with standard deviation USD 300, and the average percentage is 0.07% with standard deviation 0.07%. A more detailed report for each building is accessible Risks 2021,9, 215 16 of 19 using a web app at https://mortezatm.shinyapps.io/RiskDecomposition/ (accessed on 16 November 2021). 0.0×10+0 5.0×10−9 1.0×10−8 1.5×10−8 2.0×10−8 −50.4 −37.8 −25.2 −12.6 0 12.6 25.2 37.8 50.4 Residual loss ($ millions) Density Histogram of Residual Losses for Entire Region −5×10+7 0×10+0 5×10+7 −2 0 2 Theoretical Sample Q−Q Plot for Residual Losses Figure 7. The conditional distribution and Q-Q plot of the residual losses. 0×10+0 1×10−3 2×10−3 3×10−3 0 202 404 606 808 1010 1212 1414 1616 1818 Monthly premium ($) Density Histogram of Monthly Premiums of the Insured Buildings 0.0×10+0 2.5×10+2 5.0×10+2 7.5×10+2 1.0×10+3 0% 0.03% 0.06% 0.09% 0.12% 0.15% 0.18% 0.21% 0.24% 0.28% Monthly premium (percentage of property value) Density Histogram of Monthly Premiums Divided by Property Values Figure 8. The histogram of the calculated premiums for each building. 5. Discussion The result of the example demonstrates that the proposed methodology is able to fulfill its purposes including to transform the total loss distribution from a heavy-tail distribution into a normal distribution, and to provide an affordable insurance against a natural hazard to the households. To discuss those achievements, the distribution of the losses before and after decomposition are compared, and the affordability of the calculated premiums are investigated. Comparing Figure 6with Figure 7shows that the decomposition could successfully transform the heavy-tail distribution of the total losses to the residual losses with a normal distribution. The losses before decomposition varies from USD 37.5 million to USD 337.5 million; while, the residual losses after decomposition varies from USD − 50.4 million to USD 50.4 million. This is a significant improvement to the loss distribution regarding insurance purposes. The residual losses after the decomposition follow a normal distribution as shown in Figure 7; while, the losses before decomposition has a heavy-tail exponential-like distribution. As discussed in Section 2.1, losses with heavy-tail distributions are considered uninsurable. Thus, this example shows that the proposed decomposition is capable of making improvements to the loss distribution in this aspect as well. The calculated monthly premiums for the buildings, based on the aforementioned assumptions, are within an affordable range for the households. The average monthly premiums for the households is around USD 230 as shown in Figure 8. The median household income for Seaside, OR is reported to be USD 46, 500 in a year (Data USA 2021). Risks 2021,9, 215 17 of 19 This means that the proposed flood insurance would cost less than 6% of the income for the majority of the households. In addition, the monthly premiums are less than 0.06% of the total property value for majority of the households as shown in Figure 8. This means that the households are insuring their assets (buildings) against an event with occurrence probability of 0.01 for less than 1% of their asset value as the annual cost which is highly economical. As the results show, the methodology is capable of transforming loss distribution from a heavy-tail distribution into a normal distribution by decomposing the risk and taking out the correlated part. The correlated part can be transferred into capital markets using a set of parametric CAT bonds. The calculated insurance premiums based on such decomposition and risk transfer turn out to be affordable and economical for the households. 6. Conclusions Insurability criteria of a risk were briefly discussed. Based on the insurability criteria, natural catastrophe risks were explained to be uninsurable as they cannot fulfill the requirements of many criteria. Producing correlated losses was identified as the main issue for the insurability of Nat Cat risks. The correlation among building losses was shown to be positive and be traced back to the event intensity. Using the event intensity, we demonstrated how Nat Cat risks can be decomposed into systematic risk and idiosyncratic risk. The systematic risk was proposed to be transferred to the capital markets by a set of parametric CAT bonds based on the event intensity. The residual risk was shown to be idiosyncratic and thus insurable by its nature. The calculation of the premium for individual buildings was explained by dividing it into three parts. The first part was a portion of the CAT bond interest called spread. The second part was the building share to cover the residual risk. The third part was the share of the building to provide an underwriting gain to the insurer. We argued that our proposed way of calculation of each part is a fair practice. The total premium needs to be paid by an individual building was shown to be the sum of those three parts. The risk-return trade-off for an investment on such parametric CAT bond was discussed. The distribution of the stochastic returns generated from the parametric CAT bond was explained. The average return and the standard deviation of the returns were derived. Multi-regional and multi-hazard parametric CAT bonds were introduced as methods to lower the risk of investment on the parametric CAT bond. The average return and the standard deviation of the returns for each of those two methods were derived. How saddlepoint approximation can be implemented to find the distribution of the returns for those methods was explained. The proposed Nat Cat risk decomposition was applied to the city of Seaside, Oregon. Floods with 100-year return period or higher were studied as the natural hazard. Four types of residential buildings were chosen to be the asset at risk. A Monte Carlo simulation was implemented to find flood risk before and after decomposition. The result showed that the residual risk after the decomposition follows a normal distribution with average at zero. The insurance premium for each building was calculated. A detailed result of this study example was presented using an interactive web app created with R-Shiny package. The proposed Nat Cat risk decomposition is beneficial to all stakeholders: insurers, insureds, and investors. Insurers do not need to diversify their portfolio geographically; and they can enjoy the advantages of parametric CAT bonds including access to securitized capital, transparent triggers, and fast transaction time. Insureds need to pay considerably less insurance premiums. Investors benefits from receiving returns uncorrelated to economic performance and stock market moves; a return generated from an investment with a transparent, easy-to-understand, and easy-to-assess underlying risk driver. Risks 2021,9, 215 18 of 19 Author Contributions: Conceptualization, M.T.M. and D.L.; methodology, M.T.M.; software, M.T.M.; validation, M.T.M. and D.L.; formal analysis, M.T.M.; investigation, M.T.M.; resources, M.T.M.; data curation, M.T.M.; writing—original draft preparation, M.T.M.; writing—review and editing, M.T.M. and D.L.; visualization, M.T.M.; supervision, D.L. All authors have read and agreed to the published version of the manuscript. Funding: This research received no external funding. Institutional Review Board Statement: Not applicable. Informed Consent Statement: Not applicable. Data Availability Statement: All data, models, or code generated or used during the study are available online at: https://uofi.box.com/v/RiskDecomposition (accessed on 16 November 2021). Conflicts of Interest: The authors declare no conflict of interest. Abbreviations The following abbreviations are used in this manuscript: CAT Catastrophe GEV Generalized Extreme Value GIS Geographic Information System Nat Natural Nat Cat Natural Catastrophe NFIP National Flood Insurance Program (of the U.S.) NOAA National Oceanic and Atmospheric Administration (of the U.S.) OR Oregon State (of the U.S.) P&C Property and Casualty USGS United States Geological Survey References ASCE (American Society of Civil Engineers). 2014. Flood Resistant Design and Construction. Reston: ASCE Press. Banks, Erik. 2005. Catastrophic Risk: Analysis and Management. Hoboken: John Wiley & Sons. Berliner, Baruch. 1982. Limits of Insurability of Risks. Hoboken: Prentice Hall. Biener, Christian, and Martin Eling. 2012. Insurability in microinsurance markets: An analysis of problems and potential solutions. The Geneva Papers on Risk and Insurance-Issues and Practice 37: 77–107. [CrossRef] Butler, Ronald W. 2007. Saddlepoint Approximations with Applications. Cambridge: Cambridge University Press, vol. 22. Carpenter, Guy. 2006. Rating Agency Update: Stepping up to New Criteria. Available online: https://portal.ct.gov/-/media/CID/ App6RatingAgenUp2006pdf.pdf (accessed on 16 November 2021). Charpentier, Arthur, and Benoît Le Maux. 2014. Natural catastrophe insurance: How should the government intervene? Journal of Public Economics 115: 1–17. [CrossRef] Cummins, J. David. 2008. Cat bonds and other risk-linked securities: State of the market and recent developments. Risk Management and Insurance Review 11: 23–47. [CrossRef] Cummins, J. David. 2012. Cat bonds and other risk-linked securities: Product design and evolution of the market. SSRN Electronic Journal. [CrossRef] Data USA. 2021.Property Value for Seaside, Oregon. Technical report, Data from the Cencus Bureau ACS 5-year Estimate. Available online: https://datausa.io/profile/geo/seaside-or/ (accessed on 16 November 2021). Doherty, Neil A. 1997. Innovations in managing catastrophe risk. The Journal of Risk and Insurance 64: 713–18. [CrossRef] Enz, Rudolf. 2002. The Insurance Cycle as an Entrepreneurial Challenge. Zürich: Swiss Reinsurance Company. Grossi, Patricia. 2005. Catastrophe Modeling: A New Approach to Managing Risk. Berlin/Heidelberg: Springer Science & Business Media, vol. 25. Jaffee, Dwight M., and Thomas Russell. 1997. Catastrophe insurance, capital markets, and uninsurable risks. Journal of Risk and Insurance 64: 205–30. [CrossRef] Marvi, Morteza T. 2020. A review of flood damage analysis for a building structure and contents. Natural Hazards: Journal of the International Society for the Prevention and Mitigation of Natural Hazards 102: 1–29. [CrossRef] NFIP (National Flood Insurance Program). 2021. Flood Insurance Manual. Washington: Federal Emergency Management Agency (FEMA). Niehaus, Greg. 2002. The allocation of catastrophe risk. Journal of Banking & Finance 26: 585–96. Nofal, Omar M., and John W. van de Lindt. 2020. Minimal building flood fragility and loss function portfolio for resilience analysis at the community level. Water 12: 2277. [CrossRef] Risks 2021,9, 215 19 of 19 Ramirez, Carlos Marcelo. 2009. Building-Specific Loss Estimation Methods & Tools for Simplified Performance-Based Earthquake Engineering. Stanford: Stanford University. SwissRe. 2005. Innovating to Insure the Uninsurable. Technical Report 4. Setagaya City: Sigma. US Census Bureau. 2019. Quickfacts: Seaside City, Oregon. Technical Report. Suitland: US Census Bureau. Available online: https://www.census.gov/quickfacts/seasidecityoregon (accessed on 16 November 2021). Vickery, Peter J., Peter F. Skerlj, Jason Lin, Lawrence A. Twisdale, Jr., Michael A. Young, and Francis M. Lavelle. 2006. Hazus-mh hurricane model methodology. ii: Damage and loss estimation. Natural Hazards Review 7: 94–103. [CrossRef] Zanjani, George. 2006. Public versus Private Underwriting of Catastrophe Risk: Lessons from the California Earthquake Authority. Berkeley Symposium on Real Estate, Catastrophic Risk, and Public Policy. Available online: http://urbanpolicy.berkeley.edu/pdf/ zanjani.pdf (accessed on 16 November 2021). Zervas, Chris Eugene. 2013. Extreme Water Levels of the United States 1893–2010. Washington, DC: NOAA.