Paul Hertelendy and Erik Rüttener discuss the challenge of rolling out a global catastrophe reporting standard in Japan.
The European windstorms of 1999 lent weight and immediacy to the issue of data quality; insurers had underestimated the potential for property loss, and many had not secured enough reinsurance coverage. We realised the only solution was a universal one; data reporting would have to improve across the industry, using a common standard accepted by all market participants.
We brought together a working group of modellers and insurance and reinsurance professionals to discuss improvements to the existing catastrophe reporting standard created by CRESTA, an independent organisation set up by the reinsurance industry in 1977 to develop a global system for controlling the accumulation of natural hazard risks. The result, CRESTAplus, was adopted by a roundtable of insurers, brokers and reinsurers and introduced to the European market of 2001.
Throughout the first and now the second phase of the development of the CRESTAplus cat reporting standards, we've kept a deceptively simple phrase close at hand: it's not an industry standard unless everyone uses it. For us this has meant three things. First, the standard must capture enough information to report in any market. Second, it must be easy to use. Third, each market participant - insurer, reinsurer and broker - must see a compelling reason to adopt it.
In the past, a compelling reason has not been hard to find; insurance markets are open to change after a year of loss. Before the shock of Hurricane Hugo in 1989, the industry did not believe that hurricane losses on the US east coast could top $10bn. After Hurricane Andrew's $20bn (2001 dollars) loss in 1992, reinsurers began to employ cat models to estimate losses, and the US market adopted the UNICEDE reporting data standard developed by AIR Worldwide in 1993.
In Europe in 1999, European winter storms Lothar and Martin anchored the second-worst year of catastrophe losses since records began. Again, a compelling argument: many insurers were not adequately covered for these events. A study from the modeling company Risk Management Solutions (RMS) shows that, at the time, more than 60% of cedents indicated exposures by reporting their own premium income; fewer than 20% reported at the level of detail we would eventually establish through the CRESTAplus standard. It was not difficult to get cedents, recovering from significant losses, to understand the need to improve reporting.
We initially conceived CRESTAplus as a European answer to the post-Andrew standards adopted in the US. But as we began discussions with a working group drawn from the entire industry, we realized there was no reason why our work could not yield a global standard, and very quickly began to place a high priority on Japan. If we could adapt the standard to the first loss policies and complex multi-location risks of the Japanese market, we reasoned, we could apply the standard to any portfolio, anywhere.
A reminder in Tokyo
After rolling out the standard in October 2001 in London and Paris in time for Europe's January 1 renewals, the reinsurance roundtable convened this May in Tokyo to present CRESTAplus to the Japanese market.
We suspected that Japan would present a different challenge. For Europe, already primed by loss to change habits, data quality was the most convincing argument; higher-quality data would yield more accurate exposure information and, in turn, more accurate model results. Japan, however, though experiencing tighter terms and conditions in the wake of the September 11 losses, had no immediate loss of its own to look back on. More important, data quality in Japan was already quite high, reflecting the demands of that market's complex policies. Quality does not limit modelling accuracy the way it has done in Europe.
Our suspicions were confirmed; Japanese companies were concerned chiefly with the additional internal workload of adapting to a new standard. Preparing our presentations for the Tokyo conference, we found ourselves bumping against our old truth about what makes a standard a standard. We believe, however, after the Tokyo conference, and given the positive response from visits with our own clients in Japan, that CRESTAplus gives the Japanese market a standard it can use easily, and a reason to use it. We thought it would be worthwhile to lay out that logic here, since if it's true now in Japan, it holds for any market at any time.
We launched the CRESTAplus discussions in August 2000 with two specific reforms in mind: we wanted the industry to report aggregate sums insured, rather than premium volume or number of risks and we wanted to break down these sums at a high geographic resolution, for example, at post code level. At the time, reporting had not caught up with the computing power available to everyone in our market. We were aware, though, that with raw computing power often comes needless complexity. Without a careful selection of the most telling information, we ran the danger of creating a form with hundreds of fields, one that made the modellers happy but asked too much of the data provider. Again our axiom: if it's not easy to use, not everyone is going to use it. CRESTAplus would have to be both lean and descriptive (Figure 1).
We started by looking at the available standards (Figure 2), developed a hierarchical list of the information we found important and looked for ways to make that information more specific or more flexible. Traditionally, the reinsurance underwriter has deduced the aggregate sum insured from the reported premium volume. The aggregate sum insured came high on our list, so we dedicated two fields to describing it: number of risks; and the values of those risks. For the European market, we organised the locations of risks (another high priority) by postal code, but the standard allows this to vary by market. US ceding companies, for example, generally produce data by county.
The working group came up with some novel ways of simplifying data management. Axel Middelmann of Gerling Global Re suggested that we code perils and risk types by powers of two, assigning the code 2 to earthquake shock, 4 to an earthquake fire following, down the list until we reach landslide/mudslide, which was given a code of a whopping 8,192. The utility of this approach was as immediately apparent to us as it probably is to you: we had no idea what Axel was talking about. He explained that this convention automatically produces a unique code for each combination of individual peril codes. Thus, 104 can only be the sum of the peril codes for wind (8), flood peril (32) and storm surge (64). We eventually decided to use the same system to code risk types.
The chief concern from our Japanese cedents is not ease of use or flexibility, but cost. In a market with consistently high data quality, why invest in the transition to a new standard? The answer seems counter-intuitive: CRESTAplus saves time and increases efficiency for the reinsurer.
Reinsurers underwriting portfolios in Japan routinely spend 50% to 60% of the total time for each deal on data formatting. We believe that standard reporting can reduce this to below 5%. Since the advent of UNICEDE, data formatting hasn't even been an issue in the US. Additionally, one of the more crucial tasks we set ourselves in developing the standard was to nail down the definitions of each term and each category of data. In the past, for example, there has been confusion between the phrases `insurable value' and `total sum insured,' both of which seem essential to our trade. Common understanding minimises the questions during the underwriting process, and as with standard data formatting, for the reinsurer it reduces the time spent underwriting.
This brings us to an important equation: lower the cost to the reinsurer of processing a treaty and reducing modelling uncertainty, and you shorten response time, which over the long run should be reflected in the cost of catastrophe protection. This holds true in any market. In general, we tried to make sure through the broad composition of the working group that CRESTAplus would offer benefits for all sides of the reinsurance agreement.
We and other reinsurers in the Japanese market have been working with cedents to implement CRESTAplus for the 1 April 2003 renewals. Field trials during the 1 April 2002 renewals revealed that only a few minor adjustments would be necessary. Elsewhere, the standard showed some success during the 1January 2002 renewals in Europe. Several large composites, for example, responded favourably after testing CRESTAplus for use in their own internal systems. We have also seen interest from rating agencies, curious about whether the standard will make the assessment of cat exposure more accurate.
In addition to expanding the geographic reach of the existing standard, we are working on ways to make it more useful. The working group is developing the standard to include policy-specific location information, necessary to underwrite facultative business, and plans to introduce this refinement by the end of 2002.
We are also investigating the possibility of introducing standardised loss reporting this year. We realised that to include it from the outset might be a little too ambitious, but more accurate loss reporting would help the industry test the accuracy of its models. As with exposure reporting, loss reporting accuracy is in the long-term interest of all market participants. The reinsurance market cycle is in part subject to the influence of imperfect information; as industry losses revise expectations, premiums rise - often sharply - after an event. Accurate modeling helps match expectations to actual losses. As more reinsurers adopt a strictly technical underwriting approach, it may even help flatten the market cycle.
By Paul Hertelendy and Erik Rüttener
Paul Hertelendy and Erik Rüttener are respectively the Underwriting Manager Property Catastrophe and Manager of Global Natural Hazards for Converium.