A "CLIPER" Risk Model for Insured Losses From US Hurricane Landfalls and the Need for an Open-Source Risk Model

To plan for the consequences of hurricanes, earthquakes, and other natural hazards the public and private sectors use a variety of risk models. Model output is tailored for specific users and includes a range of parameters including: damage to structures, insured losses, and estimates of shelter requirements to care for people displaced by the catastrophe. Extensive efforts are made to tune risk models to past events. However, model "forecasts" of losses are rarely verified through a comparison with new events. Instead, new events generally are used to further tune a new version of the model. In addition, there has been no public attempt to determine which model has the most predictive skill, in part because there is no agreed upon reference forecast, and in part because most risk models are proprietary. Here I describe a simple risk model that can be used to provide deterministic and probabilistic exceedance probabilities for insured losses caused by hurricanes striking the US coastline. I propose that loss estimates based on the approach used in this simple model can be used as a reference forecast for assessing the skill of more complex commercial models. I also suggest that an effort be initiated to promote the development of an open-source risk model. The simple risk model combines wind speed exceedance probabilities estimated using the historical record of maximum sustained winds for hurricanes at landfall, and a set of normalized insured losses produced by landfalling hurricanes. The approach is analogous to weather, or climate, forecasts based on a combination of CLImatology and PERsistence (CLIPER). The climatological component accounts for low frequency variability in weather due to factors such as seasonality. The analog to climatology in the simple risk model is the historical record of hurricane wind speeds and insured losses. The insured losses have been corrected for the effects of inflation, population increases, and wealth, and other factors. The persistence component of CLIPER forecast accounts for higher frequency variability induced by phenomenon such as storms that have time scales of a few days. The analog to persistence in the simple risk model is the ability to account for climate variability associated with changes in the El Nino-Southern Oscillation and the North Atlantic Oscillation. Over the past three years model estimates of insured losses from landfalling hurricanes compare well with actual insured losses. However, it is difficult to assess how the model compares with commercial models because of proprietary concerns and nondisclosure agreements. The wide range of risk model users is an indication of the value of risk models. However, the proprietary nature of most risk models can produce problem. For example, Florida is developing its own Hurricane Loss Projection Model for regulatory purposes in part because the proprietary risk models are not open to public scrutiny. An open-source risk model could benefit any such effort to develop a custom risk model and provide a standard that could be used for regulatory and financial purposes. In addition to the ongoing work by Florida, there are other efforts to develop parts of an open-source risk model. For example, the USGS and the Southern California Earthquake Center (SCEC) are developing open-source modules for seismic hazard analysis. A more formal community effort to develop components of an open-source risk model for a variety of hazards would benefit both the public and private sector, minimize duplication of effort, and maximize the use of limited resources.