The following observations
summarise where Risk Frontiers has been in the last ten years and offer views
of the future.
A prophecy: More than 50 years ago Robert Frost, American
poet and soothsayer, noted in his classic PML study: Some say the world will
end in fire. Some say in ice.
For most of the last decade the industry has had a strong conviction that
earthquakes and tropical cyclones are the Probable Maximum Loss events for
Australia. We now believe that hail is the PML peril at least for Sydney for
the sorts of average recurrence intervals that the industry usually deals
with – say less than 500 years. Neither Risk Frontiers nor others have
done enough homework on potential bushfire losses but extreme bushfire seasons
would seem to have more serious bottom line implications for direct insurers
than for reinsurers. The industry needs to pay more attention to fire and
2. The Devil is in the Detail:
As we learn more about loss modelling, it becomes clearer that finer resolution
models provide more accurate estimates of potential portfolio losses. Representing
model geology as homogeneous across areas as large as postcodes, let alone
CRESTA Zones, in earthquake PML models is no longer satisfactory.
Loss estimation needs to focus on areas where assets at risk are located.
For Sydney, for example, only 22% of greater Sydney is occupied by residential
dwellings. Postcodes include large areas of open space relatively free of
insured assets, particularly on the urban fringes. Our recent study of potential
hailstorm losses in Sydney suggested that when portfolio data and losses are
aggregated as 255 postcodes more than one third of the 8900km2 considered
produced loss estimates differing by more than ±50% from estimates
based on 6674 Census Collection Districts. All of this implies that model
results are likely to be more accurate where postcodes and CCDs are smaller.
We should move to models that run on geocoded locations, particularly for
commercial and industrial portfolios where significant proportions of insured
assets reside in a handful of buildings.
3. Better Damage: While we now
know a fair bit about the damage produced by natural perils, we are still
short on detail. Only rarely can we determine damage rates – for example,
we don’t know how many double-brick dwellings in Modified Mercalli 7
and 8 areas in the Newcastle earthquake were undamaged? Similarly, it is not
easy to estimate the number of buildings for which there was no insurance
claim in Duffy and Chapman following the January 2003 fires.
In the 1998 Katherine flood 24% of the insured damage to residential buildings
(excluding contents) concerned replacement of built-in furniture – kitchen
cupboards, built-in wardrobes etc. Would the use of more resilient materials
be economic? Or sustainable? Has betterment a place in a triple bottom line
4. The after life: Improved
understanding of natural hazards, the new generation of spatial data, and
more sophisticated modelling allow natural perils risk assessment at increasingly
fine resolutions. FloodAUS already provides flood risk ratings for 1.3 million
addresses in eastern Australia. These new capabilities provide powerful opportunities
for insurers, mortgage providers and others to engineer portfolio risk at
a variety of scales and with a range of resolutions. Does reduced risk for
those insurers who embrace this future imply increased natural perils risk
and eternal damnation for those who don’t?
If a majority of insurers and mortgage providers adopt portfolio engineering
there are likely to be some additional environmental and social consequences.
Will residential insurance become impossible for those in the most risky locations?
Is this not a bad thing if it leads to improved land use planning and better
building codes? Does natural perils risk assessment offer a path to a more
For further information
Telephone: +61-2-9850 9683
Facsimile: +61-2-9850 9394
Damage on the Radar
Hail losses amount to
34% of the total losses in the 1967-2003 Insurance Disaster Response organisation
database, the largest proportion for any natural peril.
The main variables responsible for hail damage to property are:
- Frequency of hailstorms
- Intensity of hailstorms,
dependant on number, size (especially maximum hailstone size) and velocity
- Wind speed, which
increases the velocity of hailstones
- Exposure and vulnerability
of buildings to hailstorms
As ground measurements of hail are rather scarce and incomplete, the advantages
of remote sensing tools, such as radar, are apparent. Additionally, real time
weather radar information can be used as an early warning tool, provided that
suitable algorithms can be developed and validated by insurance claims information.
The Bureau of Meteorology operates the Weather Watch Radar “Letterbox”,
located 60km south-west of Sydney. It operates with 10-minute time steps and
a 1 km range resol-ution. The radar emits signals that are reflected by raindrops
and hailstones within thunderstorm clouds. This radar reflectivity or radar
echo provides information about location and intensity of the hail (depending
on size and concentration). Radar reflectivities greater than 55 dBZ are assumed
to contain significant hail. The radar reflectivity provides a snapshot of
the thunderstorm before the hail hits the ground. This reflectivity is then
projected onto a low storm level and close to the surface where hailstones
occur. An empirical relationship is used for further calculation as well as
integration over space and time. The resulting hail intensity represents the
entire volume of hail received per surface unit over time, a quantity that
is related to hail damage.
Figure1: Map of coastal NSW with Radar and storm
locations, dates, times, maximum hailstone sizes and insured losses. Radar
and hail data were provided by the Bureau of Meteorology, Insured losses by
Insurance Disaster Response Organisation and Emergency Management Australia.
Initial Radar investigations have been conducted on 12 recent severe thunderstorms
and their damaging hail cells (Figure1). Focus of the investigations so far
has been on the Sydney storms of October 1995, April 1999 and December 2001.
Hailstone size data for the April 1999 storm was available from a Risk Frontiers’
survey in the aftermath. Ground damage data in the form of insurance claims
have been provided for the December 2001 storm, which unfortunately turned
out to be more a destructive windstorm than a hailstorm. Nevertheless, initial
results indicate a preferred area for hail damage to the left side of the
storm path. Merging hail cells tend to cause substantially more damage than
single cells (Figure 2).
Figure 2: December 2001 severe thunderstorm with
two distinct damage paths and merging hailstorm cells. Blue dots indicate
damaged property. The background map shows the street pattern with occupied
Those results are preliminary as this work is in progress. To back up these
new findings, further ground damage data will be needed from other storms.
However, early results suggest that near real-time estimates of hail damage
can be made for buildings and possibly automobiles using the radar methodology.
These estimates would assist insurers in managing claims, in estimating resources
required and in initial reserving. Near real-time estimates would also assist
State Emergency Services in locating the worst affected areas and assessing
manpower and material requirements to protect damaged roofs and house contents.
For further information
Telephone: +61-2-9850 6378
Facsimile: +61-2-9850 9394
on the Catastrophe
capital market has been utilised as a source of additional capacity for insurance
risk. The need for additional capacity was perpetuated by several large events
in the late 1980’s and early 1990’s - Hurricanes Hugo and Andrew,
and the Northridge earthquake. One response has been the securitisation of
insurance risk, and the instrument used to transfer insurance risk to the
capital markets became known as a catastrophe (CAT) bond.
Similar to other bonds, a CAT bond has a periodic interest (coupon) payment
and return of principal at maturity. The innovation underlying this type of
instrument, however, is that principal and interest payments are at risk,
with full repayment conditional on the non-occurrence of a specified natural
The major incentive for investors to purchase a CAT bond is that they are
arguably uncorrelated with other stocks and bonds traded on the capital markets.
Investing a small proportion of a diversified portfolio in CAT bonds that
have low probability of loss will reduce portfolio risk by almost as much
as the purchase of a risk-free security. In other words, in order to improve
the risk-return profile of a portfolio, a CAT bond need only earn an expected
return slightly above the risk-free rate.
If an Australian peril were to be securitised, what would it look like? The
most likely candidates are: Sydney hailstorm, Sydney earthquake, or tropical
cyclone in Queensland.
We have been exploring some of the issues pertaining to a hypothetical hail
CAT bond for the Sydney region using Risk Frontiers’ probabilistic model
- HailAUS. Cash flow analysis and financial modelling determine an
approximate value for the premium paid by the sponsor into the Special Purpose
Vehicle (SPV). This enables the rate on line (ROL) to be calculated for the
reinsurance contract between the sponsor and the SPV.
In order to estimate the premium, we need to know the approximate return on
the bond - usually expressed as the spread above LIBOR. On the basis of previous
transactions, the return can be determined from the expected loss of the layer,
i.e., the calculated risk. The overall risk-return profile is shown in Figure
1. Demand for a new risk, such as an Australian CAT bond, is likely to lower
the required return because of diversification opportunities for investors.
In structuring a CAT bond, an appropriate trigger needs to be selected. To
date, most CAT bonds have focused on earthquake and wind events, and so the
most appropriate trigger for a hail bond has yet to be established. Various
options for triggers include indemnity, parametric, and industry-loss indices.
From a potential sponsor’s perspective, the relative attraction of this
hypothetical hail CAT bond would be determined by considering the projected
ROL along with associated advantages such as the elimination of credit risk.
The industry must draw its own conclusions as to whether the cost of the bond
makes it a credible option with which to complement conventional reinsurance
Figure 1. Spreads at issuance against expected
loss for CAT securities outstanding as at August 2003. Figure (1.b) is the
same as Figure (1.a) emphasising the lower risk-return portion of the data.
For further information
Telephone: +61-2-9850 6378. Facsimile: +61-2-9850 9394
Canberra Bushfires: What can we learn?
Saturday, 18 January 2003, bushfires that had been burning for more than a
week to the west and southwest of Canberra penetrated the suburbs. Heavy fuel
loads, difficult terrain, elevated temperatures and high windspeeds, all made
control difficult. When the inferno finally reached the outer perimeters of
the city, it fell upon a community ill prepared. When the smoke cleared, there
had been four deaths, some 500 properties destroyed and damage estimated at
around A$300 million.
Soon after the event,
Risk Frontiers undertook an informal survey of the area. The following qualitative
observations stood out.
As the fire swept through pine forests adjoining the suburbs, it consumed
all of the foliage and undergrowth, leaving only a charred landscape of
smouldering trunks. At a soil depth of only 2 cm, however, there were still
live roots suggestive of a fire propagation of high intensity and speed.
In the suburb of Chapman,
dwellings were destroyed despite having a ‘buffer’ of 500 m
of near-bare pasture land between them and the fire front. This damage could
not be attributed to radiant heating.
In terms of the nature
of the destruction, homes appeared to have been either completely laid waste
(Figure 1) or to have escaped largely unscathed. This binary outcome appeared
almost independent of what happened to the dwelling’s immediate neighbours.
what can we learn about bushfire penetration at the urban-bush interface under
extreme conditions and without community preparation? Using a fine-resolution
satellite image of Duffy, the worst affected suburb (Figure 2), we determined
the probability of house destruction as a function of distance from the edge
of the pine forests (Figure 3). Within the first 100 m, about 50% of all dwellings
were burnt down; the limit of house destruction was 674 m.
2: A 0.6-m resolution QuickBird image showing damaged houses in the
west and north of Duffy, Canberra.
The linearity of Figure
3 is intriguing, especially when one considers all the possible complicating
factors. The simplest model consistent with this linearity and the observations
given earlier follows. Imagine the fire moving towards the suburb and throwing
burning embers up to about one kilometre ahead of the fire front. As the fire
crosses the perimeter of the suburb, it runs out of fuel producing embers
capable of travelling significant distances.
3: Probability of home destruction with distance from forest edge.
Assume now that for each
ember that hits a house, there is a fixed and low probability of ignition.
Then the likelihood of a fire destroying a house, in the absence of efforts
by either the fire services or the occupants to put it out, will be simply
proportional to the number of ‘missiles’ that hit the house. The
number of hits per house is a linear function of distance from the forest
edge, as observed. If we think of the missiles as flaming arrows fired by
a band of marauding Indians galloping towards the stockade of a frontier settlement,
then we can call this the Sitting Bull Theory.
For further information
John McAneney or Keping Chen
Telephone: +61-2-9850 9683. Facsimile: +61-2-9850 9394