One way to improve catastrophe models would be to open the black box

The eminent statistician George Box first coined an adage that some catastrophe modelers might find apt: "All models are wrong; some models are useful." There is no question that catastrophe models are useful, and there is probably agreement that the "less wrong" we can make the models the better it will be for business. But, how can we make the models "more right"?

One way is to develop open source risk models. Open source software is computer code that is freely available for use and adaptation by others.

Open source risk models could be used as a testing ground for new scientific and engineering knowledge. Freely available advances made possible by the open source effort would further the field of catastrophe risk modeling and benefit insurers, regulators, emergency managers, model developers and others involved in dealing with catastrophes.

Early open source efforts focused on developing software to support computer operating systems and communication. Today open source software acts as the operating system for many computers, runs more than half the world's web servers, directs most of the email transport on the internet and provides the standard for secure communication over the internet. There are open source software versions of desktop programs such as word processors, spreadsheets and presentation software and there are open source compilers and programming languages.

The surge of interest in open source software extends to modelling efforts in the earthquake hazard community. The US Geological Survey (USGS) and the Southern California Earthquake Center started an effort to develop open source software for seismic hazard analysis (the Open Seismic Hazard Analysis, or OpenSHA). The Pacific Earthquake Engineering Research Center sponsors a project to develop a software framework for earthquake engineering studies that simulate the seismic response of structural and geotechnical systems (the Open System for Earthquake Engineering Simulation, or OpenSees).

These initiatives could form a foundation for model improvements that will lead to better assessments of seismic risk and the response of structures to ground motions.

The hurricane wind community lacks an extensive open source development effort. One potential exception is the public hurricane loss projection model funded by Florida's Department of Insurance. It remains to be determined if the code for this model will be open or closed source. With closed source software only the object code, which only a computer can use, is available; access to the source code is restricted. However, it seems likely that at least parts of the hurricane loss projection model's source code will be in the public domain, because US federal employees participate in its development.

Modelling companies

One might wonder what the companies that license proprietary risk models think of open source modelling efforts. They would probably benefit from technical advances and new business opportunities produced by an ambitious open source modelling effort. In addition, there could be important synergies between the work associated with developing open source risk models and other initiatives, ranging from homeland security to zoning codes. For example, enhanced data sets on infrastructure would aid risk modellers, emergency managers and planners. Furthermore, improvements in the collection of loss data would benefit anyone interested in modelling, regulating and underwriting re/insurance. The development of an open source risk model would provide an additional, and important, benefit: a deeper understanding of what we don't know. This is because an open source risk model would give a user the ability to assess the sensitivity of a model to the probability and character of a hazard and the vulnerability of a structure.

The timing is right to initiate a coordinated effort to develop open source risk models. Accordingly, members of the Risk Prediction Initiative (RPI), Southern California Earthquake Center and other interested people met at a workshop in March 2005 entitled "Open source catastrophe risk modeling: How can we do it better?" At the workshop scientists, engineers, commercial risk modellers and model users discussed how to best advance efforts at developing an open source risk model. It will be a challenge to assemble the detailed data on economic and insured losses for past hurricanes, earthquakes and floods needed to develop realistic estimates for loss exceedance probabilities.

The workshop was one step on a path towards developing George Box's "useful model". It is likely to be a demanding path, but given the potential benefits, it is one that should not go unexplored.

Dr Richard Murnane is senior programme manager for the Risk Prediction Initiative (RPI), which is based at the Bermuda Biological Station for Research.

Email: rmurnane@bbsr.edu.

Web site: www.bbsr.edu/rpi

Topics