Sensitivity analysis for computer model projections of hurricane losses

Area of Science:

• Meteorology and Atmospheric Science
• Risk Management and Actuarial Science
• Structural and Wind Engineering

Background:

• Accurate hurricane loss projection is critical for insurance and risk assessment but remains highly uncertain.
• Hurricane Charley (2004) highlighted extreme variability in damage estimates due to track shifts and rapid intensification.

Purpose of the Study:

• To conduct an uncertainty and sensitivity analysis for a sophisticated computer model projecting hurricane-related losses.
• To quantify the influence of input variables on wind speed predictions within the Holland wind field model.

Main Methods:

• Utilized a sophisticated computer model based on the Holland wind field for loss projections.
• Performed uncertainty and sensitivity analyses as part of the model auditing process.
• Employed standardized regression coefficients to measure the impact of input variables on wind speed.

Main Results:

• Demonstrated the significant uncertainty inherent in hurricane loss projection models.
• Quantified the contribution of various input variables to the projected wind speed magnitude.
• Provided insights for refining future sensitivity and uncertainty analyses in hurricane modeling.

Conclusions:

• Sensitivity analysis is crucial for understanding and improving the reliability of hurricane loss projection models.
• Standardized regression coefficients offer a robust method for assessing variable influence on model outputs.
• The findings support the Florida Commission on Hurricane Loss Projection Methodology's auditing process for insurance rate filings.