Catastrophe Modeling and Natural Disaster Risk
Understand catastrophe modeling frameworks for natural disaster risk assessment on Exam MAS-II.
Catastrophe Model Components
Catastrophe models have four modules that work together to estimate the financial impact of natural disasters on insured portfolios. The hazard module simulates the physical characteristics of events (frequency, location, and intensity). The vulnerability module translates physical intensity to building damage. The exposure module maps the insured portfolio to geographic locations and building characteristics. The financial module applies insurance contract terms to compute the insured loss.
These four modules represent a significant advance over purely statistical approaches to catastrophe risk. Historical insurance loss data alone is insufficient for estimating extreme event losses because the observation period is short relative to the return periods of interest (100 to 500 years), and the insured exposure has changed dramatically over time. Catastrophe models combine physical science, engineering, and financial modeling to simulate a much wider range of plausible events than history has recorded.
The Hazard Module
The hazard module simulates event occurrence and intensity using historical data and scientific models. For hurricanes, this includes modeling storm genesis (where storms form), track (the path the storm follows), and intensity (wind speed and storm surge at each location along the track). For earthquakes, seismic source models define fault locations, rupture magnitudes, and recurrence rates, while ground motion attenuation models translate rupture characteristics to shaking intensity at specific sites.
The output of the hazard module is a stochastic event catalog: a large set (typically 10,000 to 100,000 simulated years) of events with their physical characteristics at every relevant geographic location. Each event has an associated annual frequency (or probability of occurrence). The catalog represents the full range of plausible events, including events more severe than any observed in the historical record. The quality of the hazard module depends on the underlying scientific models and the historical data used to calibrate them.
The Vulnerability Module
The vulnerability module translates hazard intensity (such as peak wind speed or ground shaking) to physical damage using damage functions, also called vulnerability functions. A damage function specifies the mean damage ratio (repair cost as a fraction of replacement value) and its variability for a given building type at a given hazard intensity. Different building types (wood frame, masonry, steel frame, reinforced concrete) have different damage functions reflecting their structural characteristics.
Damage functions are developed from engineering analysis, post-event damage surveys, and claims data. They account for factors such as construction material, building age, number of stories, roof type, and occupancy (residential versus commercial). The variability in damage at a given intensity reflects the heterogeneity in construction quality, building orientation, and other factors that cannot be individually observed but affect damage outcomes.
The Exposure and Financial Modules
The exposure module maps the insured portfolio to specific geographic locations (typically geocoded to latitude/longitude) and associates each location with building characteristics needed by the vulnerability module. Data quality is critical: an error in geocoding can place a coastal building inland (underestimating hurricane risk) or vice versa. Missing or inaccurate building characteristics (such as assuming wood frame when the building is actually masonry) directly affect the damage estimate.
The financial module applies insurance contract terms to the estimated physical damage to compute the insured loss. This includes applying per-location deductibles (including percentage deductibles common in hurricane and earthquake coverage), site limits, blanket limits, and reinsurance structures. The financial module handles complex contract structures such as occurrence limits, aggregate limits, reinstatements, and coinsurance. The output is the insured loss for each simulated event, after all contract terms are applied.
Key Outputs
The exceedance probability (EP) curve is the primary output of a catastrophe model. It shows the probability of exceeding various loss levels in a given time period. Two versions are commonly produced. The occurrence exceedance probability (OEP) curve shows the probability that the largest single event in a year exceeds a given loss level. The aggregate exceedance probability (AEP) curve shows the probability that the total annual losses from all events exceed a given loss level. The OEP is relevant for per-occurrence reinsurance purchasing, while the AEP is relevant for aggregate reinsurance and capital adequacy.
Average annual loss (AAL) is the mean of the annual aggregate loss distribution. It represents the long-run average catastrophe loss per year and is the starting point for catastrophe reinsurance pricing. The probable maximum loss (PML) at a given return period (for example, the 250-year PML) is the loss level exceeded with probability 1/250 in a single year. PML estimates guide reinsurance purchasing decisions, capital requirements, and rating agency assessments.
Model Uncertainty and Limitations
Model uncertainty arises from each of the four modules and should be explicitly quantified. Hazard uncertainty reflects imperfect knowledge of event frequencies and intensities (especially for rare events beyond the historical record). Vulnerability uncertainty reflects the imprecision of damage functions. Exposure uncertainty reflects data quality issues in the insured portfolio. Financial uncertainty is generally smaller but can arise from complex or ambiguous contract terms.
Running multiple catastrophe models (from different vendors) and comparing results provides a practical assessment of model uncertainty. Material differences between models should be investigated and understood rather than simply averaged. Actuaries should also recognize that catastrophe models are calibrated primarily to past physical and engineering data, and they may not fully capture emerging trends such as climate change, urbanization, or changes in building codes. Supplementary analysis and expert judgment may be needed to address these factors.
Common Pitfalls
A common error is treating catastrophe model output as precise. All model outputs carry significant uncertainty, and point estimates (such as a single AAL or PML figure) should always be accompanied by a range or confidence interval. Another pitfall is ignoring demand surge: after a major catastrophe, the cost of materials and labor increases, inflating actual losses above model predictions based on pre-event costs.
Students sometimes confuse occurrence EP and aggregate EP. The occurrence EP asks "what is the probability that any single event exceeds this loss level?" while the aggregate EP asks "what is the probability that total annual losses exceed this level?" For a portfolio with many possible events, the aggregate EP curve will show higher loss levels at the same probability than the occurrence EP, because multiple moderate events can combine to produce a large annual aggregate even if no single event is extreme.