Model Uncertainty and the Pricing of Hurricane Risk in Florida

Figure 1: Pairwise comparisons of hurricane risk model projected loss cost per dollar exposure for frame homes in 2021. Each dot represents one ZIP code’s projected loss cost for two models.

Pairwise comparisons of hurricane risk model projected loss cost per dollar exposure for frame homes in 2021. Each dot represents one ZIP code’s projected loss cost for two models. The figure shows a grid of pairwise comparisons of loss predictions for seven catastrophe risk models labelled A through G. For each pair of models, scatter plots show predicted loss rates for 560 Forida ZIP codes. In each case the scatter of points is upward sloping but the cluster of points may lie either above or below a 45 degree line which would indicate that the forecasts match. Separate cells report pairwise correlations, which range from 0.805 to 0.911.

Return to text.

Figure 2: Projected loss costs per dollar exposure for frame homes in 2021. To aid comparisons across models ZIP codes are colored according to to their percentile rank after pooling all projections for all models.

Projected loss costs per dollar exposure for frame homes in 2021. To aid comparisons across models ZIP codes are colored according to their percentile rank after pooling all projections for all models. The figure shows seven small maps of Florida, with ZIP codes colored to reflect loss predictions for each of seven catastrophe risk models. The maps show broadly similar pattern, with red areas indicating relatively higher losses in the south of the state and near the costs, and blue areas indicating relatively lower losses in the north and in inland areas. Though the maps are similar, some show deeper red in coastal areas and a more pronounced color gradient from north to south.

Return to text.

Figure 3: Cumulative density of ZIP code average frame home projected loss costs in 2021. Gray bars show ranges from lowest to highest loss cost projections for each ZIP code.

Cumulative density of ZIP code average frame home projected loss costs in 2021. Gray bars show range from lowest to highest loss cost projections for a ZIP code. The figure is a cumulative distribution chart with “Frame Loss Cost” in the x axis and “Cumulative Probability” on the y axis. The chart shows a rising black line that increases quickly for losses from 0.00 to 0.01 and then levels off for losses above 0.01. Overlayed on the black line are horizontal gray lines which show the range of losses predicted for each ZIP code. Each gray line is centered on the black line. The range of losses reported by the gray lines are smaller for small values of the black line, and larger for large values of the black line.

Return to text.

Figure 4: Architecture of a typical hurricane risk model.

Architecture of a typical hurricane risk model. The figure is a flow chart showing the structure of a typical catastrophe risk model. Three gray boxes show “Hazard”, “Vulnerability”, and “Financial” modules. Arrows show that information flows from the hazard box to the vulnerability box, and then from the vulnerability box to the financial box. An ellipse with a connecting arrow shows that “Stochastic Event Set” information flows into the hazard box. A rounded box shows that “Portfolio Data” flows into each of the modules. “Structure Location” flows into the hazard box. “Structure Features” flows into the vulnerability box. “Policy Terms” and “Ceded Risk” both flow into the financial box.

Return to text.

Figure 5: Florida ZIP code tabulation areas covered by the Federal Insurance Office public data release, shaded according to population density. Darker regions have higher population density. Monochrome regions are not covered by the FIO data release or do not correspond to ZIP code tabulation areas.

Florida ZIP code tabulation areas covered by the Federal Insurance Office public data release, shaded according to population density. Darker regions have higher population density. Monochrome regions are not covered by the FIO data release or do not correspond to ZIP code tabulation areas. The figure shows a map of Florida with ZIP codes colored according to whether they are included in the Federal Insurance Office public data release. Red colored areas are located along the coast and in central and northeastern Florida.

Return to text.

Figure 6: Distribution of ZIP code average premiums per policy, by year.

Distribution of ZIP code average premiums per policy, by year. The figure shows two probability distribution functions: a red shaded distribution labelled 2019, and a blue shaded distribution labelled 2021. Both distributions are highly skewed, with s peak near $4,000 and a long right tail. The peak of the blue distribution is slightly higher and to the left of that of the red distribution.

Return to text.

Figure 7: Distribution of ZIP code loss cost projection coefficient of variation across vendor models, by year.

Distribution of ZIP code loss cost projection coefficient of variation across vendor models, by year. The figure shows two probability distribution functions: a red shaded distribution labelled 2019, and a blue shaded distribution labelled 2021. The blue distribution has a large peak near 0.25 and a smaller peak near 0.5. The red distribution has a single peak near 0.38.

Return to text.