Until now reinsurers had been unable to quantify with any confidence the risk of loss in multiple territories from a single hurricane. By developing a stochastic event set of many hundreds of thousands of tracks, RMS is able to not only assess the hurricane hazard at any individual location in the Caribbean islands, the US and Central American coasts, or indeed any point offshore, but also to compute the correlation of hurricane hazard, or clash, between multiple locations. The full severity/frequency relationship for loss can be determined for any geographic combination of insured exposure. Furthermore, during a real event, tracks from the stochastic population can be conditionally sampled based on the track and intensity of the real event to date. The full tracks of the selected stochastic events can also be used to provide a useful statistical forecast for losses from the real event up to several days before the event makes landfall.
The basin-wide model uses a random-walk technique frequently applied to turbulent dispersion by considering each hurricane to be advected through a 2D mean translational velocity field on which a turbulent translational velocity field has been superimposed. Both mean and turbulent velocity fields are inhomogeneous in two directions so the translation equations have been formulated to incorporate the interaction of these inhomogeneities. The tracks are modelled first and the intensity histories are added as a separate step. Model inputs are computed from the tracks of historical events in the Hurdat catalogue on a regular array of grid cells covering the whole Atlantic basin.
The model has been extensively calibrated. The rates of storms crossing each model cell have been computed and compared against the historical record. Since the historical record exists for only a little over 100 years, the historical rates for individual cells tend to be somewhat noisy, especially around the fringes of hurricane affected areas. The model rates are much smoother than the historical rates as they are based on a much larger event set. Similarly, the historical record is often too short to estimate the clash rate for many pairs of cells, although the clash can be computed from the stochastic model. Larger cells are necessary for clash comparisons to capture a significant number of historical events. Case studies of hurricanes Georges and Floyd over the last two seasons show that many tracks similar to the paths of these storms occur within the stochastic population.
The model is PC based, quick to run and allows users to explore hurricane hazard in the Atlantic basin in great detail. The model was developed for quantifying insurance risk but RMS believes the model has a large number of applications in other research areas.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner