After an earthquake caused extensive, unexpected damage to high-tech manufacturing facilities, Erik Rüttener and Linus Phoon call for better risk reporting in Taiwan.

Chi-Chi, the earthquake that devastated Taiwan in the autumn of 1999, highlighted the island's niche as supplier of a significant portion of the world's computer components by halting production for as long as two weeks at many of the island's semiconductor plants. The event left behind not only loss, but also a sense of economic vulnerability; semiconductor plants underpin the economy of the entire island.

Chi-Chi was not a blind fault. Taiwan's seismic profile, although quite complex, is well understood and reasonably well modelled. However, the earthquake caused far greater loss, insured and uninsured, than had been expected of an event of that magnitude. Underwriters had underestimated the degree of engineering damage and business interruption losses to the semiconductor industry, and the quality and resolution of the data available for Taiwan failed to provide a clear picture of the distribution of insured value on the island.

Manufacturers and the government of Taiwan have worked since the Chi-Chi earthquake to mitigate risk and improve infrastructure, and insurers have raised retentions. Many lines are assessed and priced more accurately than they were before October 1999, as can be expected after an industry loss.

Yet Taiwan's insurance market, three years later, is still at a crucial stage of development. Reinsurers no longer provide the capital support of proportional capacity for natural hazards, while some of the factors that limited the effectiveness of the island's earthquake models are still in place, and can only be addressed on a national or industry-wide basis. It is helpful, then, when trying to figure out the best way to provide capacity now, to look back three years and see what the models failed to capture.

Loss of utilities

At the time, much of Taiwan's electronics production capacity was concentrated at Hsinchu Science Park, located on the island's west coast, comparatively far from the epicentre of the Chi-Chi event. The ground motion at Hsinchu was minimal, and short-term loss - damage to equipment and silicon wafers-in-progress - came to only $7.7m, 1.5% of the total loss for the event. However, the earthquake knocked out power and water supply to the area. The Taiwanese government was able to redirect power to Hsinchu, but the island's single utility provider was strapped, and production was halted or slowed, in some places for several weeks. By the time the electronics industry was producing at full power, the total loss had ballooned to $525m.

Producing a semiconductor chip is an extraordinarily resource-draining process. A fabrication plant runs twenty-four hours a day, consuming 150,000 litres of deionised water per hour. A single wafer of 300 chips draws 800 kilowatt-hours of power, enough to light a household for two months; some of the steps in the production process require temperatures in excess of 800 degrees Celsius. It can take several days to recalibrate instruments and reheat ovens after even a brief power outage. Modellers did not have a grasp of the water and power `lifeline' dependencies of the fabrication plants, and were not aware of the potential severity of a power loss. Utility supply continues to be a concern for underwriters.

Loss of supplies

The companies of Hsinchu make up a self-sufficient network of producers and suppliers. If any one of these companies is unable to supply, it can affect production throughout the entire complex. The Chi-Chi earthquake, with minimal ground motion, shattered Taiwan's stockpile of the quartz tubes critical for heating the wafers, which proved to be a bottleneck in the weeks that followed. As with the lifeline dependencies, models failed to account for the interdependence between suppliers on the island, and the insurance industry found itself paying for business interruption losses far downstream of the actual physical loss.

Some of Taiwan's industries and single producers have addressed utility supply on an individual basis. Most of the insureds in the semiconductor industry employ backup generators, and the manufacturers of Hsinchu have created an electricity pool that will provide emergency power for up to 48 hours. Despite these measures, it is still difficult to produce loss estimates from utility interruptions with any confidence. Many sectors outside of semiconductor manufacturing have yet to develop even a minimal redundancy. In any case, no single company can take the steps necessary to ensure continuous utility supply. Ultimately this has to be addressed at the national level. Though the Taiwanese government has floated proposals to do just this, a concrete plan has yet to emerge, and Taiwan Power continues to run at full power to meet public demand.

A second industrial park has been constructed on Taiwan since the Chi-Chi earthquake. Tai-nan, located in the southwest of the island, carries even less direct earthquake exposure than Hsinchu. More important, it provides a valuable source of backup production. Many companies have opened second facilities in the new park to spread their exposure. Additionally, many companies have stockpiled several weeks' worth of production material and spares, to avoid the shipping time from Japan or Europe in the case of an event. And what is good for one company can be positive for others: the smaller the likelihood that any one supplier is unable to meet demand, the smaller the likelihood of business interruption elsewhere in the chain.

Business interruption is notoriously difficult to model. Despite the measures taken since the Chi-Chi earthquake, Taiwan's industrial facilities are no exception. In the wake of the event, many insurers have been reluctant to write business interruption lines on a gross-profit basis, preferring instead to cover spending charges. This shift in attitude underlines an important point: it is difficult for the insurer to have as detailed an understanding as the insured does of supply and utility exposure. Models are not yet sophisticated enough to reflect these interdependencies, and may never be. Additionally, Taiwan still lacks a comprehensive plan to mitigate business interruption exposures.

Engineering risk

Semiconductor production is an incredibly delicate business. A single chip may contain 40 layers of circuit patterns, perfectly aligned and etched within a tolerance of a single micron - 1/50th of the width of a human hair. Chips are produced in `clean rooms' where particle density cannot exceed one particle per cubic foot of air, 1/1,000,000th of the density of typical rural air.

Fabrication plants are tremendous concentrations of capital, and as such the buildings themselves follow stringent construction standards. As mentioned before, ground movement at Hsinchu was minimal, yet many of the production facilities sustained damage to precisely those processes at the heart of semiconductor fabrication: etching and the clean rooms. Modellers were not aware of the intolerance of these facilities to even the slightest of vibrations. Even if the building itself has sustained no damage, repair and alignment (or replacement) of delicate instruments is both costly and time-consuming.

This risk, however, has turned out to be easier to mitigate. Most companies, for example, have complied with insurers' request to move clean rooms to the ground floor, where vibrations can be minimised.

Data quality

Though a spike in computing power has made models much more comprehensive over the last decade, it takes an industry loss event to bring about the market-wide changes necessary to feed those models with enough information to actually produce more accurate results. US insurers after hurricane Andrew in 1992 agreed to a far-reaching reform in loss reporting, adopting a consistent format, adding information and increasing resolution to the zip-code level. European insurers and reinsurers developed CRESTAplus, a similar set of standards, after the European windstorms of 1999. The standards have now been introduced to the Japanese market as well.

The lack of correct, high-resolution data in Taiwan before the Chi-Chi loss lead to wildly inaccurate loss estimates. These losses, however, have yet to bring about a similar reform in reporting standards. It is still difficult in Taiwan to come up with accurate concentrations of value, and a split by line of business and coverage type. This step can be taken only by the entire industry, and the responsibility lies with all participants, reinsurers included.

A soft market in the past has removed the incentive to provide high-quality data. Currently, insurers in the Taiwanese market, faced with a higher cost of coverage and potentially greater retentions, would do well to provide more focussed data, with a higher degree of resolution. This will allow reinsurers to better estimate losses, and help insurers better understand exposure. We would recommend an approach similar to the one we brought to Europe and Japan with the CRESTAplus standard:

  • all market participants use the same reporting format;

  • clearly define categories to eliminate confusion, for example between insurable and exposed value;

  • break down exposures by lines of business and amount insured; and

  • increase geographical resolution - zip code or possibly even finer - to match detailed understanding of fault structures.

    Many insurers in this market, however, do not yet have the systems necessary to capture and manage this data. For some natural hazards, for example flooding, the data stations have not yet been established to match this increased resolution. Standardised loss reporting could be of tremendous value to the entire market, perhaps more so than any other measure, but for Taiwan it may still be a long way off.

    Non-proportional moves

    Rates hardened everywhere in the world in 2001. However, for Taiwan the 1 December renewal last year marked a maturation of the island's insurance industry. Reinsurers are no longer offering catastrophe protection on a proportional basis, and have taken natural perils out of property treaties. Insurers are taking a bigger piece of the risk, and underwriting has improved considerably in the last year. Similarly, as deductibles move up, insureds are taking a more active hand in corporate risk management, which has accelerated some of the mitigation measures mentioned earlier. Look for the market to maintain this discipline during the upcoming renewal season.

    In the long run, earthquake exposure will continue to become a better write in Taiwan, if the industry can keep its focus on underwriting discipline. Models are well suited to understand the severity side of the island's tremendous exposure to natural hazards. Reinsurers are beginning to provide catastrophe covers on an excess-of-loss basis, providing cover where they best understand the risk, and insurers and insureds are getting a better handle on their own exposure and the relationships between construction, engineering and business interruption risks, allowing reinsurers in turn to judge each risk on its own merits.

    By Barry J Yetton

    Erik Rüttener (left) is manager of Converium's natural hazards group in Zurich. Linus Phoon (below) is deputy general manager of Converium's Singapore office.