Chapter 2 An Introduction to Catastrophe Models

and Insurance

Major Contributors:
Patricia Grossi
Howard Kunreuther
Don Windeler

This chapter provides an overview of the history of catastrophe

models and their role in risk assessment and management of natural disasters.
It examines the insurability of catastrophe risk and illustrates how the output
from catastrophe models aids insurers in meeting their goals for risk
management. Throughout the chapter, there is an emphasis on understanding
catastrophe modeling for earthquake and hurricane hazards and how it is used
to manage natural hazard risk. In the final section, a framework for integrating
risk assessment with risk management via catastrophe modeling is presented.

2.1 History of Catastrophe Models

Catastrophe modeling is not rooted in one field or discipline. The
science of assessing and managing catastrophe risk originates in the fields of
property insurance and the science of natural hazards. Insurers may well argue
that catastrophe modelings history lies in the earliest days of property
insurance coverage for fire and lightning. In the 1800s, residential insurers
managed their risk by mapping the structures that they covered. Not having
access to Geographic Information Systems (GIS) software, they used tacks on
a wall-hung map to indicate their concentration of exposure. This crude
technique served insurers well and limited their risk. Widespread usage of
mapping ended in the 1960s when it became too cumbersome and time-
consuming to execute (Kozlowski and Mathewson, 1995).
On the other hand, a seismologist or meteorologist may well argue
that the origin of catastrophe modeling lies in the modern science of
understanding the nature and impact of natural hazards. In particular, the
common practice of measuring an earthquakes magnitude and a hurricanes
intensity is one of the key ingredients in catastrophe modeling. A standard set
24
of metrics for a given hazard must be established so that risks can be assessed
and managed. This measurement began in the 1800s, when the first modern
seismograph (measuring earthquake ground motion) was invented and
modern versions of the anemometer (measuring wind speed) gained
widespread usage.
In the first part of the twentieth century, scientific measures of natural
hazards advanced rapidly. By the 1970s, studies theorizing on the source and
frequency of events were published. Significant analyses include the U.S.
Water Resources Council publication on flood hazard (USWRC, 1967), the
Algermissen study on earthquake risk (Algermissen, 1969) and National
Oceanic and Atmospheric Administration (NOAA) hurricane forecasts
(Neumann, 1972). These developments led U.S. researchers to compile hazard
and loss studies, estimating the impact of earthquakes, hurricanes, floods, and
other natural disasters. Notable compilations include Brinkmanns summary
of hurricane hazards in the United States (1975) and Steinbrugges anthology
of losses from earthquakes, volcanoes, and tsunamis (1982).
These two separate developments mapping risk and measuring
hazard came together in a definitive way in the late 1980s and early 1990s,
through catastrophe modeling as shown in Figure 2.1. Computer-based
models for measuring catastrophe loss potential were developed by linking
scientific studies of natural hazards measures and historical occurrences with
advances in information technology and geographic information systems
(GIS). The models provided estimates of catastrophe losses by overlaying the
properties at risk with the potential natural hazard(s) sources in the geographic
area. With the ability to store and manage vast amounts of spatially referenced
information, GIS became an ideal environment for conducting easier and
more cost-effective hazard and loss studies.
Around the same time, several new modeling firms developed
computer software for analyzing the implications of natural hazard risk. Three
major firms emerged: AIR Worldwide was founded in 1987 in Boston; Risk
Management Solutions (RMS) was formed in 1988 at Stanford University;
and EQECAT began in San Francisco in 1994 as a subsidiary of EQE
International. In 2001, EQE International became a part of ABS Consulting.
When introduced, the use of catastrophe models was not widespread.
In 1989, two large-scale disasters occurred that instigated a flurry of activity
in the advancement and use of these models. On September 21, 1989,
Hurricane Hugo hit the coast of South Carolina, devastating the towns of
Charleston and Myrtle Beach. Insured loss estimates totaled $4 billion before
the storm moved through North Carolina the next day (Insurance Information
Institute, 2000). Less than a month later, on October 17, 1989, the Loma
Prieta Earthquake occurred at the southern end of the San Francisco
peninsula. Property damage to the surrounding Bay Area was estimated at $6
billion (Stover and Coffman, 1993).
 25
These two disasters sent a warning signal to the insurance industry.
On the heels of these two events, Hurricane Andrew made landfall in
Southern Florida in August of 1992. Within hours of landfall, AIR Worldwide
issued a fax to its clients to the effect that losses, as estimated in real time by
the AIR Worldwide hurricane model, might reach the astonishing amount of
$13 billion. It was not until months later that the final tally, $15.5 billion, was
issued by the Property Claim Services Office.
Nine insurers became insolvent as a result of their losses from
Hurricane Andrew. Insurers and reinsurers realized that, in order to remain in
business, they needed to estimate and manage their natural hazard risk more
precisely. Many companies turned to the modelers of catastrophe risk for
decision support. The modeling companies grew and catastrophe models
increased in number, availability, and capability. By 2001, other organizations
joined these front-runners in developing catastrophe models for assisting
insurers and reinsurers in pricing their insurance policies and determining
how much coverage to offer in hazard-prone areas of the country.
The series of natural disasters in 1989 and 1992 also sent a warning
signal to the public sector of the United States. The government recognized
the need for an accurate assessment of the impact of disasters for mitigation
and emergency planning purposes. In 1992, the Federal Emergency
Management Agency (FEMA) funded a study to assess the latest loss
estimation methodologies for earthquakes. The agency issued a report in 1994
on the results of this study entitled: Assessment of the State of the Art
Earthquake Loss Estimation Methodologies (FEMA 249, 1994).

Figure 2.1. Development of catastrophe modeling.

26
This study convinced FEMA to fund the development of Hazards
U.S. (HAZUS), a catastrophe model in the public domain. HAZUS is
labeled as an open source model in Figure 2.1. From the outset, one of
FEMAs goals was to create a methodology that was the standard national
loss methodology for assessing losses from natural hazards (FEMA, 2002).
The first version of HAZUS was developed with a combination of public and
private resources to estimate earthquake losses and was released in 1997
(NIBS, 1997). Updates to the HAZUS earthquake model have been in the
form of data and software integration; methodologically, the software remains
the same. In 2004, the latest HAZUS multi-hazard methodology, relabeled
HAZUS-MH, integrates the earthquake module with two new modules for
estimating potential losses from wind and flood (riverine and coastal) hazards.

2.2 Structure of Catastrophe Models

The four basic components of a catastrophe model are: hazard,
inventory, vulnerability, and loss as depicted in Figure 2.2. First, the model
characterizes the risk of natural hazard phenomena. For example, an
earthquake hazard is characterized by its epicenter location and moment
magnitude, along with other relevant parameters. A hurricane is characterized
by its projected path and wind speed. The frequency of certain magnitudes or
frequencies of events also describes the hazard in question.

Figure 2.2. Structure of catastrophe models.

Next, the model characterizes the inventory or portfolio of properties

at risk as accurately as possible. Arguably, the most important parameter used
to characterize the inventory is the location of each property at risk. A process
called geocoding is normally used to assign geographic coordinates such as
latitude and longitude to a property based on its street address, ZIP code or
another location descriptor. With a propertys location in spatial terms, other
factors that could aid in estimating the vulnerability of a property are added to
its characterization. For a building, these parameters include such features as
its construction type, the number of stories in the structure, and its age. If the
 27
property is insured, information on the nature of the policy, such as the
deductible and coverage limit, is also recorded.
The hazard and inventory modules enable the calculation of the
vulnerability or susceptibility to damage of the structures at risk. In essence,
this step in the model quantifies the physical impact of the natural hazard
phenomenon on the property at risk. How this vulnerability is quantified
differs from model to model. For example, the HAZUS model classifies a
structure as being in a Slight, Moderate, Extensive, or Complete damage state.
Other models construct damage curves and relate structural damage to a
severity parameter, such as peak gust wind speed or spectral acceleration. In
all models, damage curves are constructed for the building, its contents and
time element losses, such as business interruption loss or relocation expenses.
From this measure of vulnerability, the loss to the inventory is
evaluated. In a catastrophe model, loss is characterized as direct or indirect in
nature. Direct losses include the cost to repair and/or replace a structure.
Indirect losses include business interruption impacts and relocation costs of
residents forced to evacuate their homes. Proprietary models include the
ability to analyze insurance policies, so that the loss can be properly allocated.
More details on these elements of a catastrophe model are provided in
Chapter 3.

2.3 Uses of a Catastrophe Model for Risk Management

A catastrophe model is employed to assess catastrophe risk and
improve risk management decisions. But how is this accomplished? Briefly,
the model output is quantified and presented in a way that is useful to the
stakeholder. Once these metrics are in hand, alternate risk management
strategies, such as mitigation, insurance, reinsurance and catastrophe bonds,
can be assessed. Currently, insurers and reinsurers are the stakeholders with
the most widespread interest and integrated use of catastrophe models.
Reinsurance brokers in particular have enhanced the use of catastrophe
models. It is fairly common for a broker to collect data for potential clients,
run the models on that data, and provide the output to interested reinsurers.
The capital markets have also been eager users of this technology in
order to more accurately price catastrophe bonds. In fact, their recent interest
and involvement in natural hazards have been made possible by the
quantification afforded by catastrophe modeling. Property owners are less
likely to use catastrophe models themselves, but their decision processes are
directly or indirectly influenced by the outcomes. At the governmental level,
catastrophe modeling presents both a positive opportunity and a political
dilemma for regulators and emergency management agencies.
As an example of a positive use of the models, consider the use of
HAZUS to measure the impact of an earthquake. One model output option is
28
to create a GIS map of the potential loss. Given the definition of the hazard,
including the earthquakes epicenter location, and the concentration of the
properties at risk, Figure 2.3 depicts a map of the displaced households for the
Charleston, South Carolina region subject to an M 7.3 earthquake. The largest
concentration of loss, measured by the number of individuals seeking shelter
following the disaster, is near the scenarios epicenter. This map is potentially
useful to emergency response and recovery officials responding to a disaster.

Figure 2.3. Catastrophe model output: Map of shelter requirements predicted by

HAZUS for M 7.3 events in Charleston, South Carolina region.

Another output option is the exceedance probability (EP) curve. For a

given portfolio of structures at risk, an EP curve is a graphical representation
of the probability that a certain level of loss will be surpassed in a given time
period. Special attention is given to the right-hand tail of this curve where the
largest losses are situated. Figure 2.4 depicts an EP curve for an insurer with a
portfolio of residential earthquake policies in Long Beach, California. In
contrast to a GIS map of loss, which presents loss in a spatial manner, an
exceedance probability curve portrays loss in a temporal manner.
An EP curve is particularly valuable for insurers and reinsurers to
determine the size and distribution of their portfolios potential losses. Based
on the EP curve, they can determine the types and locations of buildings they
would like to insure, what coverage to offer, and what price to charge. To
 29
keep the probability of insolvency at an acceptable level, insurers can also use
an EP curve to determine what proportion of their risk needs to be transferred
to either a reinsurer and/or the capital markets.
For example, suppose an insurer in Long Beach offers residential
earthquake coverage and the insurers exceedance probability curve for its
portfolio is as depicted in Figure 2.4. Further suppose the insurer specifies
$10 million as an acceptable level of loss at a 1% (1-in-100) probability of
exceedance. Based on the graph, it can be seen that loss profile of the current
portfolio would be unacceptable since the 1-in-100 loss for the portfolio is
$ 15 million. The insurer would need to look for ways to reduce its portfolio,
transfer $5 million of loss to a reinsurer, or purchase a catastrophe bond to
cover it.

Figure 2.4. Catastrophe model output: Right-hand tail of exceedance probability

curve predicted by EQECAT for all possible events.

2.4 Derivation and Use of an Exceedance Probability

Curve
Given the importance of how insurers use catastrophe modeling and
the EP curve to manage risk, it is essential to understand how the EP curve
can be created from the loss output.

2.4.1 Generating an Exceedance Probability Curve

For the purposes of illustration, some simplifying assumptions are
made to generate an EP curve. Suppose there is a set of natural disaster
30
events, which could damage a portfolio of structures. Each event has an
annual probability of occurrence, and an associated loss, The number of
events per year is not limited to one; numerous events can occur in the given
year. A list of 15 such events is listed in Table 2.1, ranked in descending order
of the amount of loss. In order to keep the example simple and calculations
straightforward, these events were chosen so the set is exhaustive (i.e., sum of
the probabilities for all of the events equals one).
The events listed in Table 2.1 are assumed to be independent
Bernoulli random variables, each with a probability mass function defined as:

If an event does not occur, the loss is zero. The Expected Loss for a given
event, in a given year, is simply:

The overall expected loss for the entire set of events, denoted as the average
annual loss (AAL) in Table 2.1, is the sum of the expected losses of each of
the individual events for a given year and is given by:

Assuming that during a given year, only one disaster occurs, the exceedance
probability for a given level of loss, can be determined by calculating:

The resulting exceedance probability is the annual probability that the

loss exceeds a given value. As seen in the equation above, this translates into
one minus the probability that all the other events below this value have not
occurred. The exceedance probability curve for the events in Table 2.1 is
shown in Figure 2.5. Sidebar 1 explains how the EP curve can be used to
determine probable maximum loss (PML).
 31

SIDEBAR 1: PML as a function of the EP Curve

The exceedance probability curve illustrated in Figure 2.5 enables an
insurer to determine his PML or Probable Maximum Loss for a portfolio of
structures in a given time period. The term PML is a subjective risk metric
and is associated with a given probability of exceedance specified by the
insurer. For example, suppose that an insurer specifies its acceptable risk level
as the 0.4% probability of exceedance. The insurer can use the EP curve to
determine how large a loss will occur at this probability level. Often, PML
limits are framed in terms of a return period. The return period is simply the
inverse of the annual probability of exceedance. In this example, a 1-in-250
year PML is the lower limit on the loss at a 0.4% probability of exceedance
on the EP curve. From the inset of Figure 2.5, it can be seen that the PML is
approximately $21 million.
32

Figure 2.5. Exceedance probability curve

2.4.2 Stakeholders and the Exceedance Probability Curve

The exceedance probability curve can also be used to distribute the
losses between stakeholders. Suppose there are three stakeholders who share
the losses from a particular disaster. The owner retains the first part of the
loss, a second party covers the middle portion and a third party covers the
extreme portion. This scenario could represent a portfolio of homes with the
homeowners having deductibles on their insurance policies such that they
cover the first portion of the loss, an insurer covers the middle portion and a
reinsurer handles the losses above a certain amount. Figure 2.6 shows a
simple illustrative example. The potential loss for a portfolio with a total
value of $100 million is split between three participants: P1, P2, and P3. The
first $5 million of loss (L1) would be borne by P1 (homeowners), losses
between $5M and $30M (L2) by P2 (insurer), and losses in excess of $30M
(L3) by P3 (reinsurer). If the events facing the three parties were those given
in Table 2.1, then the reinsurer would never experience any claim payments
because the maximum loss would be $25 million.
Now suppose the three parties face the set of events in Table 2.1, but
there is some uncertainty associated with the losses from each of the first 14
events has a loss of zero). In other words, the losses in Table 2.1
represent the mean estimates of loss; each event has a distribution of loss
associated with it. There is now a range of possible outcomes for each event,
and some of these will penetrate the higher layer L3 (Figure 2.7). By
 33
combining the loss distributions for all the events, the probability of
exceeding a specific loss level can be calculated. This then becomes the basis
for developing EP curves for each of the parties with resources at risk.

Figure 2.6. Layering for hypothetical portfolio, total value $100 million.

Figure 2.7 shows a set of loss-causing events with a high level of

uncertainty in the loss distributions where the coefficient of variation (CV) on
the event losses is 1.0. 1 By definition, the coefficient of variation is the ratio
of the standard deviation to the mean. The effect of this high uncertainty is
clearest on L3. If there were no variability in the losses, L3 would not be
affected because no event is capable of reaching a $30 million loss, as
previously stated. Based on the assumption (CV = 1.0), there is an annual
probability of 0.28% that an event would cause some loss to L3.
This illustrative example shows how catastrophe modeling provides a
means of both quantifying risks and allocating them among stakeholders.
Using these metrics, it is possible to make rational, informed decisions on
how to price risks and determine how much coverage is needed based on an

1
Note that the assumption of a constant coefficient of variation for all events is not realistic and
is used only for ease of illustration. The CV on the event loss generally decreases as the size of
the loss increases; a portfolio CV of 1.0 for the most damaging event in this example is highly
unlikely.
34
acceptable level of risk. However, there are uncertainties inherent in the
catastrophe modeling process that can have a large impact on the distribution
of risk among stakeholders. The quantification and disaggregation of
uncertainty provides opportunities for stakeholders to reduce risk. As will be
discussed in Part II, some of this uncertainty can be reduced by better data,
but a significant component is an intrinsic part of the physical process.

Figure 2.7. Exceedance probability curves for total portfolio and individual
participants.

2.5 Insurability of Catastrophe Risks

In most developed countries, insurance is one of the principal
mechanisms used by individuals and organizations to manage risk. Insurance
allows the payment of a relatively small premium for protection against a
potentially large loss in the future. In the United States, some property
insurance coverage is required by law or by the lending institution. For
example, homeowners normally have to purchase fire coverage as a condition
for a mortgage. Automobile liability insurance is also required in most states
 35
as a condition for licensing a car. However, earthquake insurance is usually
not required by lenders on single-family residences.
Insurance pricing can be a signal of how risky certain activities are
for a particular individual. To illustrate, consider automobile insurance. For
cars that are the same price, younger, inexperienced drivers of sporty vehicles
pay more in premiums than older drivers of more conservative cars. For life
and health insurance, smokers pay more for coverage than nonsmokers. This
allocation of risk seems appropriate since it is tied to the likelihood of
outcomes resulting from the nature of an individuals lifestyle. If one
individual is more susceptible to a specific risk, then the cost for coverage
against a loss from that risk is greater. Of course, since insurance rates are
subject to regulation, the price of the policy may not fully reflect the
underlying risk.
The key challenge is how to allocate catastrophe risk among
stakeholders in a manner similar to what is done for more frequent, non-
extreme events. For automobile coverage, considerable historical data are
available and utilized to estimate insurance premiums for individuals with
different risk characteristics. The large number of data points and the absence
of correlation between accidents allow the use of actuarial-based models to
estimate risk (Panjer and Willmot, 1992). With respect to natural disasters,
there are limited data available to determine the probabilities of events
occurring and their likely outcomes. In the absence of past data, there is a
need for insurers to model the risk. Catastrophe models serve this purpose by
maximizing the use of available information on the risk (hazard and
inventory) to estimate the potential losses from natural hazards.

2.5.1 Conditions for Insurability of a Risk

Consider a standard insurance policy whereby premiums are paid at
the start of a given time period to cover losses during this interval. Two
conditions must be met before insurance providers are willing to offer
coverage against an uncertain event. The first condition is the ability to
identify and quantify, or estimate at least partially, the chances of the event
occurring and the extent of losses likely to be incurred. The second condition
is the ability to set premiums for each potential customer or class of
customers.
If both conditions are satisfied, a risk is considered to be insurable.
But it still may not be profitable. In other words, it may be impossible to
specify a rate for which there is sufficient demand and incoming revenue to
cover the development, marketing, operating, and claims processing costs of
the insurance and yield a net positive profit over a prespecified time horizon.
In such cases, the insurer will opt not to offer coverage against this risk.
To satisfy the first condition, estimates must be made of the
frequency of specific events and the likely extent of losses. Such estimates
36
can be based on past data or catastrophe modeling, coupled with data on what
experts know about a particular risk. The insurer can then construct an
exceedance probability (EP) curve that depicts the probability that a certain
level of loss will be exceeded on an annual basis.
With respect to the second condition, if there is considerable
ambiguity or uncertainty associated with the risk, insurers may wish to charge
a much higher premium than if they had more precise estimates of the risk
(Kunreuther, Hogarth and Meszaros, 1995). Moreover, if the capacity of the
insurance industry is reduced due to recent large losses, then premiums will
rise due to a shortage in supply. The situation will be exacerbated if the recent
losses trigger an increase in demand for coverage, as was the case after
Hurricane Andrew in 1992 and the Northridge earthquake in 1994
(Kunreuther and Roth, Sr. 1998).
Once the risk is estimated, the insurer needs to determine a premium
rate that yields a profit and avoids an unacceptable level of loss. There are a
number of factors that influence an insurers decision on what premium to set.
State regulations often limit insurers in their rate-setting process, and
competition can play a role in what may be charged in a given marketplace.
Even in the absence of these influences, there are a number of issues that an
insurer must consider in setting premiums: uncertainty of losses, highly
correlated losses, adverse selection, and moral hazard. Neither adverse
selection nor moral hazard appears to be a major problem with respect to
natural hazard risks. Adverse selection occurs when the insurer cannot
distinguish (or does not discriminate through price) between the expected
losses for different categories of risk, while the insured, possessing
information unknown to the insurer, selects a price/coverage option more
favorable to the insured. Moral hazard refers to an increase in the expected
loss caused by the behavior of the policyholder. One example of moral hazard
is moving unwanted furniture into the basement so an impending flood can
destroy it, but this behavior occurs very infrequently. Given the difficulty
uncertainty of losses and highly correlated losses pose in setting premiums,
they are discussed below.

2.5.2 Uncertainty of Losses

Natural disasters pose a set of challenging problems for insurers
because they involve potentially high losses that are extremely uncertain.
Figure 2.8 illustrates the total number of loss events from 1950 to 2000 in the
United States for three prevalent hazards: earthquakes, floods, and hurricanes.
Events were selected that had at least $ 1 billion of economic damage and/or
over 50 deaths (American Re, 2002).
Looking across all the disasters of a particular type (earthquake,
hurricane or flood), for this 50-year period, the median loss is low while the
maximum loss is very high. Given this wide variation in loss distribution, it is
 37
not surprising that there is a need for catastrophe models to aid insurers and
reinsurers in estimating the potential loss from events that have not yet
occurred but are scientifically credible.

Figure 2.8. Historical economic losses in $ billions versus type of significant U.S.
natural disaster. 1950-2000 (Source: American Re)

2.5.3 Highly Correlated Losses

Natural disasters involve spatially correlated losses or the
simultaneous occurrence of many losses from a single event. If insurers sell a
block of residential policies in a neighborhood, they could potentially
experience a large (spatially correlated) loss should a disaster occur in the
region. For example, due to their high concentration of homeowners policies
in the Miami/Dade County area of Florida, State Farm and Allstate Insurance
paid $3.6 billion and $2.3 billion in claims respectively in the wake of
Hurricane Andrew in 1992. Given this unexpectedly high loss, both
companies began to reassess their strategies of providing coverage against
wind damage in hurricane-prone areas (Lecomte and Gahagan, 1998).
In general, insurance markets flourish when companies can issue a
large number of policies whose losses are spatially and otherwise
independent. The portfolio follows the law of large numbers, and is thus
predictable. This law states that for a series of independent and identically
distributed random variables, the variance around the mean of the random
variables decreases as the number of variables increases. Losses from natural
hazards do not follow the law of large numbers, as they are not independent.
38
2.5.4 Determining Whether to Provide Coverage
In his study, James Stone (1973) sheds light on insurers decision
rules as to when they would market coverage for a specific risk. Stone
indicates that firms are interested in maximizing expected profits subject to
satisfying a constraint related to the survival of the firm. He also introduces a
constraint regarding the stability of the insurers operation. However, insurers
have traditionally not focused on this constraint in dealing with catastrophic
risks.
Following the disasters of 1989, insurers focused on the survival
constraint in determining the amount of catastrophe coverage they wanted to
provide. Moreover, insurers were caught off guard with respect to the
magnitude of the losses from Hurricane Andrew in 1992 and the Northridge
earthquake in 1994. In conjunction with the insolvencies that resulted from
these disasters, the demand for coverage increased. Insurers only marketed
coverage against wind damage in Florida because they were required to do so
and state insurance pools were formed to limit their risk. Similarly, the
California Earthquake Authority enabled the market to continue to offer
earthquake coverage in California.
An insurer satisfies the survival constraint by choosing a portfolio of
risks with an overall expected probability of insolvency less than some
threshold, A simple example illustrates how an insurer would utilize the
survival constraint to determine whether the earthquake risk is insurable.
Assume that all homes in an earthquake-prone area are equally resistant to
damage such that the insurance premium, z, is the same for each structure.
Further assume that an insurer has $A dollars in current surplus and wants to
determine the number of policies it can write and still satisfy its survival
constraint. Then, the maximum number of policies, n, satisfying the survival
constraint is:

Whether the company will view the earthquake risk as insurable

depends on whether the fixed cost of marketing and issuing policies is
sufficiently low to make a positive expected profit. This, in turn, depends on
how large the value of n is for any given premium, z. Note that the company
also has some freedom to change its premium. A larger z will increase the
values of n but will lower the demand for coverage. The insurer will decide
not to offer earthquake coverage if it believes it cannot attract enough demand
at any premium structure to make a positive expected profit. The company
will use the survival constraint to determine the maximum number of policies
it is willing to offer.
The EP curve is a useful tool for insurers to utilize in order to
examine the conditions for meeting their survival constraint. Suppose that an
 39
insurer wants to determine whether its current portfolio of properties in Long
Beach is meeting the survival constraint for the earthquake hazard. Based on
its current surplus and total earthquake premiums, the insurer is declared
insolvent if it suffers a loss greater than $ 15 million. The insurer can construct
an EP curve such as Figure 2.4 and examine the probability that losses exceed
certain amounts. From this figure, the probability of insolvency is 1.0%. If
the acceptable risk level, then the insurer can either decrease the
amount of coverage, raise the premium and/or transfer some of the risk to
others.

2.6 Framework to Integrate Risk Assessment with Risk

Management
Figure 2.9 depicts a framework for integrating risk assessment with
risk management and serves as a guide to the concepts and analyses presented
in this book. The risk is first assessed through catastrophe modeling.
Catastrophe modeling combines the four components (hazard, inventory,
vulnerability, and loss) to aid insurers in making their decisions on what type
of protection they can offer against a particular risk.
The key link between assessing risk via catastrophe models and
implementing risk management strategies is the stakeholders decision
processes. The types of information stakeholders collect and the nature of
their decision processes are essential in developing risk management
strategies. With respect to insurers, catastrophe models are the primary
sources of information on the risk. Their decision rule for developing risk
management strategies is to maximize expected profits subject to meeting the
survival constraint. Property owners in hazard prone areas utilize simplified
decision rules in determining whether or not to adopt mitigation measures to
reduce future losses to their property and/or to purchase insurance.
For purposes of this book, risk management strategies are broadly
classified as either risk reduction measures, such as mitigation, or risk transfer
measures, such as reinsurance. For example, strategies for residential property
owners often involve a combination of measures, including mitigation,
insurance, well-enforced building codes, and land-use regulations. In California
and Florida, all these initiatives exist in some form. Strategies for insurers could
involve charging higher rates to reflect the uncertainty of the risk, changing their
portfolio so they can spread the risk across many areas, or reassigning the risk
using risk transfer instruments such as reinsurance and/or catastrophe bonds.
40

Figure 2.9. Framework for linking risk assessment with risk management.

2.7 Summary and Relationship to Parts II-IV

This chapter examined the history of catastrophe modeling and the
role catastrophe models play in making a risk insurable. Part II provides a
more detailed discussion of catastrophe modeling for earthquakes and
hurricanes. The output from catastrophe models provides important
information for insurers to manage their risk. By modeling the risk, insurers
can more accurately estimate the premiums to charge for insurance coverage
from natural disasters. In addition, insurers and reinsurers are able to tailor
their coverage to reduce the chances of insolvency. They can develop new
strategies for managing their portfolios so as to avoid losses that might
otherwise cause an unacceptable reduction in surplus. These strategies are
discussed in Part III of the book.
The impact of insurers risk management strategies on profitability
and probability of insolvency are explored further in Part IV of the book.
Exceedance probability curves are constructed using real market data for
insurers in Oakland, California, Long Beach, California and Miami/Dade
County, Florida and alternative strategies are examined, including requiring
mitigation to homes in these disaster-prone areas and using risk transfer
instruments to satisfy an insurers survival constraint. The book concludes
with a chapter on the future role of catastrophe models in dealing with the
risks associated with terrorism as an extreme event.
 41
2.8 References

American Re (2002). Topics: Annual Review of North American Natural

Catastrophes 2001.

Algermissen, S.T. (1969). Seismic risk studies in the United States, 4th World
Conference on Earthquake Engineering Proceedings, Chilean Association for
Seismology and Earthquake Engineering, Santiago, Chile.

Brinkmann, W. (1975). Hurricane Hazard in the United States: A Research

Assessment. Monograph #NSF-RA-E-75-007, Program on Technology, Environment
and Man, Institute of Behavioral Sciences, University of Colorado, Boulder,
Colorado.

FEMA 249 (1994). Assessment of the State of the Art Earthquake Loss Estimation
Methodologies, June.

FEMA (2002). HAZUS99 SR2 Users Manual, Federal Emergency Management

Agency.

Insurance Information Institute (2000). Catastrophes [online]. The Insurance

Information Institute 10 June 2000 <http://www.iii.org/media/issues/catastrophes>.

Kozlowski, R. T. and Mathewson, S. B. (1995). Measuring and Managing

Catastrophe Risk,1995 Discussion Papers on Dynamic Financial Analysis, Casualty
Actuarial Society, Arlington, Virginia.

Kunreuther, H., R. Hogarth, J. Meszaros and M. Spranca (1995). Ambiguity and

underwriter decision processes, Journal of Economic Behavior and Organization,
26: 337-352.

Kunreuther, H. and Roth, R. (1998). Paying the Price: The Status and Role of
Insurance Against Natural Disasters in the United States. Washington, D.C: Joseph
Henry Press.

Neumann, C.J. (1972). An alternate to the HURRAN tropical cyclone forecast system.
NOAA Tech. Memo. NWS SR-62, 24 pp.

NIBS (1997). HAZUS: Hazards U.S.: Earthquake Loss Estimation Methodology. NIBS
Document Number 5200, National Institute of Building Sciences, Washington, D.C.

Panjer, H. H. and Willmot, G.E. (1992). Insurance Risk Models. Illinois: Society of
Actuaries.

Steinbrugge, K. V. (1982). Earthquakes, Volcanoes, and Tsunamis: An Anatomy of

Hazards. Skandia America Group: New York, New York.
42

Stone, J. (1973). A theory of capacity and the insurance of catastrophe risks: Part I
and Part II, Journal of Risk and Insurance, 40: 231-243 (Part I) and 40: 339-355
(Part II).

Stover, C.W. and Coffman, J.L (1993). Seismicity of the United States, 1568-1989.
U.S. Geological Survey Professional Paper 1527, United States Government Printing
Office, Washington, D.C.

USWRC (1967). A Uniform Technique for Determining Flood Flow Frequencies,

U.S. Water Resource Council, Hydrology Committee, Bulletin 15, Washington, D.C.
http://www.springer.com/978-0-387-23082-5