Modeling Typhoon Risk in South Korea

In June 2002 the Red Devils of South Korea conquered the world of football in a flash and sparked the enthusiasm of millions of fans. Barely three months later, typhoon Rusa destroyed the largest football stadium in the country. Between August 31 and September 1, 2002, South Korea was subjected to violent torrents of rain, and thousands of houses, bridges, and farmlands were destroyed. One hundred and nineteen people lost their lives to the storm.

South Korea is the seventh largest insurance market in the world. Total premium income is approximately $57 million, of which 25 percent comes from nonlife insurance companies. The insurance penetration rate is also high. Premiums as a percentage of the gross domestic product reached 10 percent for life and 3.2 percent for nonlife insurance in 2000. Statistics predict that at least one typhoon will strike the Korean peninsula each year during August.

Because of the high market penetration in South Korea, more sophisticated risk analysis is needed to determine typhoon risk. Unfortunately, the typhoon history only reveals data back to the 1950s. For an accurate risk analysis it is essential to include events with low frequencies (e.g., 100-, 250-, 1,000-year return periods) to examine the possible maximum loss and to establish a risk premium.

Typhoon data was downloaded from the Joint Typhoon Warning Center including measurements from 1950 to 2001. For each track, observations include date, time (6-hour intervals), and the main parameters of a typhoon, which are sustained wind speed, longitude and latitude, and pressure (not available for most tracks). In addition, data from 76 weather stations helped to determine the approximate wind field pattern. The wind field is used to examine the vortex of each stochastic event. To create a stochastic event set, ArcGIS Geostatistical Analyst examined the probabilities of wind speed, forward speed, and direction of each track.

The random sampling technique determined wind speed, forward speed, and direction of each track point. All probabilities of the necessary parameters, which were determined by kriging, were defined as subgroups. Finally, parameters were selected randomly from each subgroup.