Athena and Artemis

An analysis of claims data for the Rhône River floods in 2003, the most expensive event to hit the French scheme for compensating victims of natural catastrophes, has demonstrated the value of a geographical information system in estimating catastrophe exposures.

Insurers use complex data incorporating time elements, such as the contract period and claims development during the duration of the policy, and geographical descriptions of the location of insured properties throughout the portfolio. They collect this information essentially at the time of underwriting and subsequent changes in the risk and when claims are paid.

This information can also be used in the risk management process, both in terms of the exposure to natural hazards and their probable manifestation in terms of claims. Geographical information systems (GIS) allow insurers to integrate and manage the temporal and spatial data found in their portfolios, but individual company results give an incomplete picture of market exposure. Achieving a global, market wide perspective, independent of the underwriting policies and geographical zones of activity of individual insurers, requires the collection and amalgamation of a representative sample of market data.

The Caisse Centrale de Réassurance (CCR) is a public reinsurance company offering reinsurance for risks covered under France's 1982 law on compensation for victims of natural catastrophes. It collects information on primary insurers' portfolios and losses from natural catastrophes which it holds in a dedicated database, called Athena. Athena provides the raw data for the modelling project known as Artemis, which simulates natural catastrophes covered by the 1982 law and the costs of the claims that they generate.

The analysis using Artemis of the 2003 Rhône floods, which resulted in claims totalling 738 million euros, including motor, shows that the model correctly reproduces the characteristics of the event at local level and gives an accurate indication of the level of insured damages.

We believe this demonstrates that geographical information systems (GIS) are suitable to perform time and spatial analyses of insurance portfolios and analyse the risks at the level of individual property addresses. With a reliable market base, it is possible to generate experience reports that will contribute to the understanding of risk.

Underlying principles

Four fundamental principles underlie the Athena database:

• Conservation of the memory of natural catastrophes

• Analysis of the cause of events and their consequences

• Dissemination of the results of the analyses

• Heightened awareness of the need for mitigation

Athena's objective is to make a complete inventory of the risks and losses generated by natural catastrophes and model them. The data it holds come from insurance companies, who collect the information in the normal conduct of their business the Athena data consist of two types:

1 Insured risks (described as vulnerability or property covered), notably

- The commune where the insured property is sited

- The type of risk: individual, agricultural, commercial, industrial, house or apartment, and whether the assured is the owner, tenant or a non-resident owner

- The contract period and the relevant premium

- The economic values at risk

2 Claims history.

- The date of occurrence

- The cost

- The peril: flood, subsidence, earthquake etc.

Currently, the information held in Athena represents 70% of the market, and it can be extrapolated to create a global vision of risk. This collection of data makes it possible to produce experience reports on recent events.

In this way, Athena has a double objective: contributing a better appreciation of the exposure of insurance companies, the CCR and the state, and supporting catastrophe modelling.

One of the difficulties encountered in the exploitation of claims records concerns the location of individual risks in a policy covering multiple sites. The map above represents the four sites on the same policy. If the claim is associated with the whole policy, the loss cannot be linked to a specific site. Furthermore, if all the sites covered are situated in communes within a natural flood disaster area, the declaration orders give no additional information which would allow greater precision in locating the loss.

Statistical simulation of flood

The Artemis project makes use of the Athena database with the aim of modelling, initially, the most expensive events, which are floods. They represent 60% of the total cost of natural catastrophes in continental France since the beginning of the 1982 law regime. The objectives are:

• To estimate, within a few days of occurrence, the cost of an event so as to improve the calculation of the necessary technical provisions

• To evaluate the exposure to the state, the CCR and insurance companies, notably by simulation of historic events, such as the Paris flood of 1910, on contemporary portfolios

The estimation of the cost of a given event on a portfolio is made up of three parts:

1 The hazard model, which simulates the physical phenomena that occur during the event

2 The vulnerability model, which integrates the characteristics of the portfolio taking account of vulnerability of the insured property, independent of the event

3 The damage model, which estimates the probable claims according to the intensity of the event and the vulnerability of the portfolio

We have analysed the Rhône floods in December 2003 with the help of the information from Athena and simulated the property damage, excluding motor vehicles, with Artemis. The Rhône event affected 24 departments from the Loire to the mouth of the Rhône between 1 and 10 December 2003. The state of natural catastrophe comprised 1,490 communes, representing 33,944 km2. Total claims were 738 million euros.

Ten communes (meaning less than 1% of the communes included in the declared state of catastrophe) accounted for more than half of the total cost of the event and over 60% of the non-motor claims. The commune of Arles in the Department of the Gard alone was responsible for nearly 25% of the losses. This high cost was the result of a breach in the gates of a passageway under the railway embankment upstream of the commune, which resulted in significant damage in the industrial zone in the north of Arles.

The simulation by Artemis allowed an analysis at the local level, for example, the commune of Saint-Gilles. During the Rhône 2003 event, the commune of Saint-Gilles did not suffer flooding directly from excess rainfall. The total rainfall of 160 mm in three days which fell on Saint-Gilles does not constitute enough to result in significant run-off; by contrast, in September 2002, a total of 558 mm of rain fell on the city of Arles.

Instead, a substantial accumulation of rainfall drained into the whole Rhône basin and resulted in the overflowing of the river at Saint-Gilles, which is close to its mouth. A fact illustrates the significance of this phenomenon; the return period for the discharge observed at Saint-Gilles rose to 500 years.

Saint-Gilles is situated downstream of the river, thus it is especially vulnerable to such over-flowing. It had a higher proportion of exposed property, both private and business, than the whole of the area damaged by the event. This can be seen when we examine the breakdown of the number of policies in the zones affected by run-off and over-flowing compared to the breakdown of the costs in these affected zones.

It follows from the hazard model and the vulnerability model, that damage to property from over-topping of the river would be more costly than damage to property from run-off of excess rainfall, and this was the case.

For all policies, the total losses from over-flow were 82% and 18% from run-off, although 85% of the policies for private property and 58% for commercial property were located in the zone affected by run-off. This breakdown compares with an average for the whole Rhône flood of 65% for over-flowing and 35% for run-off where 97% of policies were located in zones affected by run-off.

The total cost estimated by Artemis in euros is in keeping with the actual losses within the limits of the confidence intervals.

In conclusion, the sharing of data makes it possible to get a market view and to appreciate more precisely the risk. CCR continues to contribute to the understanding of the effect of natural catastrophes in setting up the Athena database. Athena is an indispensable primary source for experience reports. It is essential that data included in Athena is of high enough quality and sufficient quantity for us to perform these analyses.

The use of geographical information systems to perform the time-space analysis of exposures has become essential in dealing with insurance issues. Artemis, an example of such a tool, allows us to draw out the characteristics of floods on a local scale and provides an order of magnitude for the cost of insured damage.

Suzanne Corona is senior vice president, director of the French natural catastrophe division for the Caisse Central de Réassurance.

www.ccr.fr