Frequently Asked Questions

1. How does the transparent FPHLM offer advantages when compared to industry black box models?[TOP]

a. Transparency

Commercial models (such as those from Verisk or Moody's RMS) act as "black boxes" where algorithms and methodologies are closely guarded trade secrets. The FPHLM is open to the scientific community, allowing for peer-reviewed validation and an understandable foundation to check proprietary model assumptions. Unlike proprietary industry models, it allows scientific scrutiny and provides regulators like the Florida Office of Insurance Regulation (OIR) with an unbiased baseline to review insurance company rate filings.

b. Public Scrutiny

We make available all our methodology, assumptions, equations, algorithms, flowchart etc. However, we do not share the source code to ensure its integrity.

2. What information is available? Who benefits?[TOP]

a. Independent Regulatory Tool

It gives state regulators an objective, non-political metric to accept or reject insurer requests for rate hikes. This prevents rate manipulation and ensures premium pricing is actuarially sound for the state's millions of homeowners.

b. Stress Testing

The OIR uses the model to perform vital stress tests, ensuring private property insurers possess the financial capacity to withstand major catastrophic events.

c. Hyper-Focused Florida-Centric Variables

Because the FPHLM is solely focused on Florida, it aggressively accounts for state-specific vulnerabilities.

d. Workforce Development

The project has trained over a hundred graduate and undergraduate students in fields such as computer science, engineering, meteorology, hydrology, coastal science, and insurance. These professionals are future leaders to benefit the state of Florida and the nation regarding risk modeling and mitigation for enhancing community resilience.

3. How does the model help with mitigation analysis?[TOP]

a. Objective Mitigation Analysis

The model reliably quantifies how much money property owners save by retrofitting homes and enforcing building codes, making it an excellent policy tool for promoting hurricane resilience. The model provides the scientific basis for premium discounts for mitigation. Such discounts, negotiated by FL-OIR, can amount to over 50% in South Florida. Tests done in the Wall of Wind Experimental Facility helped produce improved vulnerability functions, including those with mitigation features.

b. Examples of Actual Discounts Offered for Mitigation

Homeowner median annual insurance premium for a $300,000 masonry home, 2023 (2% deductible):

4. How do FPHLM outputs compare with those from other models?[TOP]

2025 ranking of cat models for statewide modeled losses, ranked from highest to lowest losses:

1. Karen Clark & Company
2. Verisk
3. Florida Public Hurricane Loss Model
4. Moody's Corporation (previously Risk Management Solutions)
5. Impact Forecasting
6. Corelogic

FPHLM loss estimate is in the middle. This provides evidence that the FPHLM results fall somewhere in the middle of the range of other proprietary models.

5. How is the accuracy of FPHLM validated and continuous improvement implemented?[TOP]

The model is backed by very good statistical validation. Statistical tests indicate a strong agreement between the actual and modeled losses. Team members conduct cutting edge peer reviewed research that help improve the catastrophe models. Innovative testing at the NSF-supported NHERI Wall of Wind Experimental Facility helps to improve modeling accuracy.

Below are some evidences on FPHLM comparison of modeled vs actual losses based on claims data (66 observations across hurricanes and companies). The comparison indicates a reasonable agreement between the actual and modeled losses.

• The correlation between actual and modeled losses is found to be 0.965, which shows a strong positive linear relationship between actual and modeled losses.
• We tested whether the difference in paired mean values equals zero using the paired t test (t = 1.33, df = 75, p-value = 0.1872) and Wilcoxon signed rank test (V = 1627, p-value = 0.397). Based on these tests, we failed to reject the null hypothesis of equality of paired means and concluded that there is insufficient evidence to suggest a difference between actual and modeled losses.
• We also observed that 51% of the actual losses are more than the corresponding modeled losses, and 49% of the modeled losses are more than the corresponding actual losses. This shows that our modeling process is not biased.
• Following Lin (1989), the bias correction factor (measure of accuracy) is obtained as 0.949, and the sample concordance correlation coefficient is found to be 0.916, which again shows a strong agreement between actual and modeled losses.

Our sample of observation is significantly larger than other modelers.