Impact Forecasting

Risk. Reinsurance. Human Resources.

Aon Benfield

Impact Forecasting FCHLPM 2019 Hurricane Standards

Model Overview Presentation (Page 3)Standards Review Presentation (Page 51)June 2021

2© 2021 Impact Forecasting L.L.C. Proprietary

IF/Aon Attendees

Kopal Arora, Senior Scientist, R&DKatie Carter, Managing Director, Aon Reinsurance SolutionsSiamak Daneshvaran, Ph.D., P.E., A.Re., A.R.M., Senior Managing Director, Global Head of R&DXian He, Ph.D., Senior Scientist, R&DSteven Jakubowski, President, R&DAshwin Joseph, Associate Director, Business Development & Client SupportYujin Liang, Ph.D., P.E., Director, R&DMaria Lomelo, Managing Director, Global Program DirectorChris Long, Director, Software and AnalyticsMinchong Mao, FCAS, CCRMP, MAAA, Managing Director, Actuary, Aon Reinsurance SolutionsNehal Naik, Managing Director, Software DevelopmentSami Pant, Ph.D., P.E., Senior Scientist, R&DBin Pei, Ph.D., Associate Director, R&DAdam Podlaha, Ph.D., Head of Impact ForecastingSri Harshitha Polamuri, Ph.D., Senior Scientist, R&DVenkatesh Ramaiah, Associate Director, Software DevelopmentRoozbeh Raoufi, Ph.D., Senior Scientist, R&DElham Sharifineyestani, Ph.D., Senior Scientist, R&DWill Skinner, Managing Director, Global Head of Business DevelopmentVipin Unnikrishnan, Ph.D., Associate Director, R&DKarthik Yarasuri, Senior Scientist, R&D


Aon Benfield

Impact Forecasting Florida Hurricane Model Version 1.0ELEMENTS Version 15.0

FCHLPM 2019 Hurricane StandardsModel Overview PresentationJune 2021


Agenda

Section 1 IntroductionSection 2 Hazard ComponentSection 3 Vulnerability ComponentSection 4 Financial ComponentSection 5 Software Component


Aon Benfield

Introduction


1Introduction


Introduction

This is the first Impact Forecasting (IF) hurricane model submission to the FCHLPM

Submission details– Standards: 2019– Model: Impact Forecasting Florida Hurricane Model Version 1.0– Software: ELEMENTS Version 15.0

The model has been verified by the Professional Team as meeting all the Commission’s model standards

The model is respectfully submitted to the Commission for acceptability approval

Introduction


About Impact Forecasting

Impact Forecasting LLC is a catastrophe-risk modeling firm, wholly owned by Aon– Founded in 1995– Global team of 100+ research, software, and support professionals

IF maintains a suite of independent, transparent, and modular catastrophe models– Transparent views of risk for peak perils and gap perils– Open model and software architecture

ELEMENTS is IF’s catastrophe loss modeling software platform– Runs Impact Forecasting models and 3rd party models

IF’s models are used by Aon catastrophe modeling teams, insurers, reinsurers, and regulators to quantify and manage catastrophe risk

Introduction


IF Hurricane Model Development History

Introduction


Determine wind speeds at risk

locations

$0 $100

0.01%

0.02%

0.05%

0.10%

0.20%

0.40%

1.00%

2.00%

5.00%

10.00%

20.00%

Insured Loss

Exce

edan

ce P

roba

bilit

y

Model Overview

Introduction

Simulate hurricane events

Assign damage ratios

Calculate insured loss


Aon Benfield

Hazard Component


2Hazard

Component


Hazard Component Overview

Historical Years: 1900–2018 Historical Scenarios: ~ 90 Stochastic Simulation Years: 200,000 Stochastic Events: > 41,000 Hazard Resolution: 1 km / ZIP Code

Hazard Component

StochasticEvents

HistoricalScenarios


Stochastic Event Modeling Approach

Hazard Component

Step 1:Genesis modelTC initialization

Step 2: Track, intensity and Rmax modelsTC movement, intensification and size change

Step 3: Decay modelCentral pressure filling

Step 4: Wind field and terrain modelsWind speed calculation

* Rmax – Radius of maximum winds


Step 1: Genesis Model

Annual number of tropical cyclones– Negative binomial distribution Initial storm parameters

– Randomly sampled from historical genesis data

Hazard Component


Step 2: Track Propagation, Intensity and Rmax Models

Track propagation model– Forward speed and heading angle (Vickery et al. 2000) Intensity model

– Central pressure (Vickery et al. 2000)– Maximum sustained wind speed Vmax (Knaff & Zehr 2007) Rmax model (Vickery & Wadhera 2008)

Hazard Component


Step 3: Inland Decay Model

Central pressure decay after landfall Inland decay rate

– Vickery (2005), varies by region

Hazard Component

Gulf

FloridaPeninsula Mid-Atlantic

New England

MexicoCaribbeanIslands

Remaining coastsin the Caribbean


Step 4: Windfield Model

Hazard Component

Willoughby wind profile

Provides spatial distribution of wind

– Symmetric windfield• Willoughby’s model Parametric wind profile obtained after

analyzing flight-level wind measurements

– Inflow angle model• Models the deviation of hurricane

winds from tangential direction

– Asymmetry model• Models the asymmetrical hurricane

windfield behavior


Step 4: Terrain Model

Hazard Component

Mean Terrain Factor Snapshot

Low Terrain Factor

HighTerrain Factor

Assess actual wind speeds considering influence of ground surface roughness

– Land Use Land Cover (LULC)• Indicates the different land types (e.g., forests, wetlands, impervious surface)

– Roughness length• LULC converted to a quantitative

parameter which is the basis of roughness coefficient equations

– Terrain factor• Modifies the open terrain wind speeds

to account for actual terrain conditions• Based on ESDU; evaluated in 8

directions• Lower the terrain factor, higher is the

reduction in wind speed


Historical Scenario Modeling and Validation

Hazard Component

IF Florida Hurricane Model suite includes a set of historical hurricanes landfalling in or by-passing Florida

– Generated at 1 km resolution– Maximum gust wind speeds at 10m height

Terrain model

Hurricane wind speeds

Hurricane parameters

Windfieldmodel

Validation– Modeled wind speeds compared with

station measurements

IRMA 2017


Stochastic Model Validation – Occurrence Rate

Hazard Component

1 23

4

5

6 7

8

91011


Stochastic Model Validation – Hurricane Parameters

Hazard Component

Forward Speed Central Pressure Rmax

Heading Angle Vmax

Modeled hurricane parameters are validated individually against historical observations at landfall


Stochastic Event Sampling

Hazard Component

Direct Monte Carlo simulation approach to estimate hurricane loss costs and hurricane output ranges

200,000 years of stochastic simulation – Maximum standard error at county level is less than 2.5% of the loss cost for every

county in Florida

166,000 loss-generating events in Florida; Importance sampling reduced the final event set to ~41,000 events– Based on the 2017 FHCF, the difference in statewide average annual loss based on

all events and the sampled set of events is less than 0.01%


Event Set Catalog Construction

Historical catalog– 89 historical hurricanes that made landfall in Florida and its neighboring states, as

well as by-passing hurricanes that produced loss in Florida

Stochastic catalog– 41,470 loss-generating hurricanes that by-pass or make a direct landfall in Florida

Cell-level wind speeds– 1 km resolution– Used when properties are provided with latitude/longitude information

ZIP Code averaged wind speeds– Population-weighted average– Used when risks are geocoded to ZIP-Code level

Hazard Component


Aon Benfield

Vulnerability Component


3VulnerabilityComponent


Overview of Impact Forecasting Vulnerability Model

The vulnerability component has been developed over the course of more than 25 years at Impact Forecasting

The original damage functions were developed using a component-based Monte Carlo simulation tool– Uses engineering principles and reliability methods in the form of load and capacity

characteristics of various building components to estimate failure probability and wind damage characteristics

The vulnerability functions are calibrated and validated using historical insurance claims data, post-event reconnaissance reports and peer reviewed research publications



IF Building Stock Classifications

The building stock classifications are based on a review of building stock inventory using information from:– U.S. Census– Tax Appraiser Data– Public Domain Reference Information such as

• Federal Emergency Management Agency (FEMA)• Florida Hurricane Catastrophe Fund (FHCF)• Post-Event Reconnaissance Surveys

IF building stock classification was compared with exposure data from historical insurance claims

Impact Forecasting and Aon have supported clients for more than 20 years through hurricane modeling using IF models



IF Building Stock Classifications


Occupancy TypesResidential single familyResidential multi-familyResidential renterResidential condo unit ownerCommercial apartment condo associationUnknown occupancy type

Construction TypesMasonryMobile home double wideMobile home not tied downMobile home tied downMobile home unknown tie down stateReinforced concreteSteel frameUnknown construction typeWood

Number of StoriesUnknown stories1 story2 stories3 stories4 to 10 stories11 to 20 storiesGreater than 20 stories

Year Built Band (non-MH)Unknown year builtPre-19951995 to 20012002 to 2011Post-2011

Year Built Band (MH only)Unknown year builtPre-19761976 to 19931994 to 19981999 to 2007Post-2007

Tier (Vulnerability Region)1 (Broward County, Miami-Dade County and the Florida Keys)2 (Areas where the ultimate design wind speed is 140 mph or greater, areas within 1 mile of the coastal mean high-water line, and areas not in Tier 1)

3 (All areas not in Tier 1 and 2)


Building Vulnerability Function Development

Developed using a building component-based wind damage simulator The Monte Carlo simulation technique is used to perform statistical sampling

for various wind loading conditions on different structural components in order to estimate damage outcomes for each component

Different classes of components in the building have their relevant capacities and the associated probability distributions– Failure of a component, given a wind speed, occurs when the load is larger than the

capacity Pressures acting on different building components, the number of openings

damaged due to windborne debris impacts are simulated for each wind speed Component-based losses are aggregated into overall building outcomes which

represent both the mean and the standard deviation of the damage ratio



Building Component-Based Damage Simulator


Wind

Wind

InternalPressure

Gable

Roof CoveringDamage

Window Damage

Define building and component information

Assign component capacities

Calculate wind loads on components based on

their position

Compute damage statistics

Wind speed in m/s

Mea

n D

amag

e R

atio


Final ELEMENTS Building Vulnerability Functions


An example building vulnerability function with mean and standard deviation used in ELEMENTS is provided below


Contents Vulnerability Function Development

Contents vulnerability functions were initially established in 1998 using engineering judgement and calibrated with historical data from Hurricane Andrew– These data were embedded into the simulation tool and used to establish the initial

contents vulnerability curves

Vulnerability curves have been further calibrated over time with historical damage data – Following the 2004, 2005, and 2008 storm seasons – Following Superstorm Sandy in 2011

The current set of contents vulnerability functions were updated in 2020 – New claims from four homeowners' portfolios

As contents vulnerability is a function of building damage, any update to building loss expectations is reflected in contents response



Time Element Vulnerability Function Development

The time element vulnerability functions were developed as a function of building damage based on engineering analysis and judgement and published peer-reviewed research

The time element vulnerability functions were initially developed using a component-based model combined with restoration process modeling using the estimation of expected restoration times (Program Evaluation and Review Technique/Critical Path Method) to determine time element losses

The cost & daily output data for repair activities were obtained from RSMeans(Waier, 1995; Mewis, 2020)

The time element vulnerability functions were calibrated and validated using claims data from multiple historical events



Secondary Modifier Overview

The impact of mitigation measures and secondary characteristics on a building’s expected vulnerability and its associated uncertainty distribution is captured by secondary modifiers which may increase, decrease, or maintain the primary building vulnerability functions

The secondary modifiers were developed through wind engineering principles and calibrated based on post-disaster damage survey reports, insurance claims data, and engineering judgement

The effects of multiple hurricane mitigation measures and secondary characteristics are combined using a multiplicative methodology

The combined impact on the building damage ratio is capped with an upper and lower bound to control the maximum change when many secondary modifiers are simultaneously applied

The change in building damage ratio also affects the contents and time element vulnerabilities since the contents and time element vulnerabilities are derived and modeled as a function of the building vulnerability

In the absence of any mitigation or secondary risk characteristics, the model uses the primary vulnerability curve without any modification



Secondary Modifier Categories

Key secondary modifiers were identified based on structural engineering expertise and post-event reconnaissance of historical hurricane damage


Roof AgeUnknown Roof AgeRoof Age 0-5 yearsRoof Age 5-10 yearsRoof Age >10 years

Roof-to-Wall ConnectionsUnknown ConnectionNo Strap and No ToenailOnly Toenail100% Straps100% Clips

Roof CoveringUnknown Roof CoveringShingles (90 mph)TilesRated Shingles (150 mph)Rated Shingles w/ Second water BarrierMetal Roof Covering

Roof SheathingUnknown Roof sheathingRoof Sheathing(6d @ 6"/12")Roof Sheathing(8d @ 6"/12")Roof Sheathing(8d @ 6"/6")Metal DeckConcrete Deck

Roof TypeUnknown Roof TypeHip Roof TypeFlat Roof TypeUnbraced Gable Roof TypeBraced Gable Roof Type

Floor SizeUnknown BuildingBig Building (>=5000 ft2)Average Building (2000-5000 ft2)Small Building (<2000 ft2)

Window ProtectionUnknown Window ProtectionNo Window protectionOrdinary Non-ImpactBasic ImpactHurricane Impact

Garage DoorsGarage Door UnknownGarage Door BracedGarage Door UnbracedNo Garage


Historical Loss Validation

IF’s Florida Hurricane Model reasonably replicates incurred hurricane losses for 12 historical events from 2004 through 2018 for four insurance companies


Modeled loss vs. claims by hurricane event


Aon Benfield

Financial Component


4Financial

Component


Exposure Classes

Financial Component

Location Risk Characteristics

Insured Limits

Account / Policy

Values

Risk CharacteristicData: Detailedinformation on riskcharacteristics, such as occupancy, construction, stories, and yearbuilt

Limits & Deductible Data: Data on application of insured limits and deductibles (coverage limits, policy limits, etc.)

Account and Policy Data: Account/ Policy ID, layered & shared (i.e., $3M p/o $5M XS $3M, reinsurance -treaty or facultative)

Values Data: Structure, contents, and time element values (complete TIVs)

Location Data: Postal Code or Street Address information for geocoding


External Data: Import Process – Storage and Additional Data Fields

Financial Component

Data Import: Loads external data into Exposure Database– Address/ Geographic coordinates: Used to assign

various parameters such as wind tier– Geocoding: Data import into ELEMENTS assigns

geographic coordinates (latitude, longitude) from address data

– Hazard ID: Assigns the hazard lookup (cell ID field) based on geographic coordinates (used to tie exposure and hazard footprint together)

– Vulnerability curves: Determined from primary and secondary construction, occupancy classifications, and tier

– Summary Reporting: After import, a summary exposure report is available for review

Exposure Database

External Text Data Loads into Exposure Database


ELEMENTS Insurance Processing Features

Performs Monte Carlo simulation using beta distribution for loss calculation Simulated ground up losses are run through full financial engine Supports insurance and reinsurance structures Supports annual hurricane deductible Supports complex commercial policy structures, sub-limits Produces event loss tables at portfolio and various breakout levels such as

state, county, ZIP Code, etc. Input exposure includes both replacement values and insured limits

(Insurance-to-Value up to the user)

Financial Component


ELEMENTS Financial Module

Financial Component


ELEMENTS Financial Module

Financial Component

Step 1: For each event

Hazard database contains the event catalog for all wind peril events and footprints

Data table contains the 1-km gridded data points for each peril event in the event set

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

20 30 40 50 60 70 80 90 100

Dam

age

Ratio

(0 <

DR

<1.0

)

Wind Speed (gust m/s)

Example Damage Function

Example DF

Step 2: For each sample

Calculate damage ratio by coverage Apply coverage and location level terms Apply policy level terms Apply reinsurance terms Repeat for each sample

Vulnerability based on Ground Up Losses

( )µ,σ


ELEMENTS Results

Financial Component

Event Loss Tables

Portfolio

Line of Business

LOB by State

Country

Postal Code

Operational

Analysis Info

Location

Account

LOB by County

State

County

User Defined

Loss Case Master

Analysis Status

Results Database

Analysis Results Database

Analysis Process: Analysis process

results are loaded into the RESULTS database

Primarily two forms of information are stored:– Operational –

administrative data on runs and settings

– Event Loss Tables (ELTs) for various breakout data groupings

Processed

PMLs/AALs


Aon Benfield

Software Component


5Software

Component


ELEMENTS Catastrophe Modeling Platform

Software Component

Detailed modeling application platform designed for growth and retention, portfolio aggregation, underwriting, and risk transfer analyses

− Supports personal, commercial and industrial lines of business

Multi-tier secured architecture featuring customizable deployment for different team size and usage

SQL server-based architecture that coordinates analyses across multiple cores− Efficient input of exposure information and easy access to loss results

In-depth reporting beyond PML results

Exposure Import

Hazard Loss Financial Terms

Results


Scalable Deployment

Software Component

CloudClient/ServerPC

Laptop or DesktopModel developers, clients

with limited models

On-Premises DeploymentInsurance and reinsurance

companies

Cloud Deployment3rd party developers, insurance

and reinsurance companies


ELEMENTS Catastrophe Modeling Platform Progression

Software Component

ELEMENTS v1.0 - v4.0Running in-house within IF

Pre-2003 2006

ELEMENTS v5Desktop product

5 models

2010

ELEMENTS v7Client-Server15+ models

2015

ELEMENTS v9.5On Cloud

ELEMENTS v11100+ models

2017 2021

ELEMENTS v15100+ models

Florida Hurricane Model

2017

https://www.insuranceerm.com/content/awards/insuranceerm-annual-awards-uk-and-europe/


Impact Forecasting Model and Product Development

Software Component

Development methodology– Using Agile methodology– Iterative development and testing

Technology– Windows .Net Framework– C# and SQL– MATLAB– Python– R

Development process– Capture requirement specification– Scope, plan and schedule– Documentation (architecture, design)– Implementation and peer reviews – Testing– Product delivery and support

Documentation– Requirements– Model methodology– Architecture and design– Variable mapping table– All documents stored on SharePoint

Standards– Following ISO 5807 for flowcharts– BPMN for business process diagram– UML for class, sequence and use case diagram

Coding using best practice– All codes are stored on TFS (version control system)– Coding guidelines created for various languages used– Variable naming for better readability– Use of constants instead of hard coded values– Commenting for each class, method and logical blocks


ELEMENTS User Interface

Software Component

Informative and intuitive User Interface to streamline workflow

Dashboard displaying jobs and status Exposure reporting

Exposure mapping


Loss Results

Software Component

Loss (OEP/AEP) reports are available on the User Interface Reports can be exported to Excel for further

processing

OEP Curve


IF Florida Hurricane Model Summary

Hazard Component Vulnerability Component

Financial Component ELEMENTS Platform

Supports basic and complex insurance and reinsurance structures

Suitable for insurance rate filing and reinsurance portfolio analysis

Robust and modular platform for hosting IF and 3rd party models

Well-tested and verified following rigorous process and industry standards

200k years of simulation and 41k+ events Stochastic and historical events modeled

from inception to dissipation using historical data

Reflects recent scientific and engineering advancements

Thoroughly validated using historical observations offshore, at landfall, and inland

19k+ vulnerability functions 400+ secondary modifier combinations Component-based simulation approach Well-calibrated using extensive and

detailed claims data


Aon Benfield

Questions?


Aon Benfield


FCHLPM 2019 Hurricane StandardsReview PresentationJune 2021


Agenda

Section 1 Response to DeficienciesSection 2 General StandardsSection 3 Meteorological Standards Section 4 Statistical StandardsSection 5 Vulnerability StandardsSection 6 Actuarial StandardsSection 7 Computer/Information Standards


Aon Benfield

1Response to Deficiencies

Response to Deficiencies




G-2.B, page 30 and Form G-3, page 134: The signatory for Statistical Standards does not have the requisite advanced degree in statistics. – Addition of Radek Solnicky, M.S., Mathematical Statistics, as signatory for the

Statistical Standards. Consequently, the addition of Radek Solnicky to Table 1 (G-2, Disclosure 2, page 31) and Appendix B, page 249.

M-2, Disclosure 3, pages 42-43: Incomplete. The annual-frequency negative-binomial distribution should be listed here. – Listed the annual-frequency negative-binomial distribution.

M-4, Disclosure 10, page 52 and Form M-2, pages 143-149: Incorrect. The form states that the “open terrain” surface roughness only applies open terrain to land points, while water points are to be kept to the same as standard model version. However, the response to M-4 Disclosure 10 indicates that modeler applied open terrain everywhere, including water points. – Response to M-4 Disclosure 10 was modified to clarify that open terrain surface

roughness was applied to land points, and water points were kept the same as the standard model version for the open terrain maps in Form M-2.




M-6, Disclosure 4, pages 56-57: Incomplete. Missing comparison to 40 mph radius. The text says the 73 mph and 40 mph simulated and modeled radii “are compared below.” – Clarified response and added data for 40 mph radius to Table 2.

V-1, Disclosure 3, page 82: Incomplete. No response for number of insurers and amount of hurricane loss separated into personal residential, commercial residential, and manufactured homes provided. – Added required data (number of insurers and amount of hurricane loss) to Table 6.

A-1, Disclosure 3, page 98: Non-responsive. Methods need to be described. – Methods that describe how policy types are distinguished in the model have been

included.



Aon Benfield

2General

Standards

General Standards



G-1 Scope of the Hurricane Model and its Implementation (1/2)

General Standards

Model version number– Impact Forecasting Florida Hurricane Model Version 1.0

Platform– ELEMENTS Version 15.0

The model estimates loss costs and probable maximum losses to insured residential property caused by wind damage from hurricanes. The methods are actuarially sound as demonstrated in Standard A-4.

Impact Forecasting employs standard methods and procedures to ensure agreement and correct correspondence of databases, data files, and computer source code. Source control software and error tracking systems are used to maintain accuracy.

All software and data used to develop and validate the model, project loss outputs, and create forms comply with the Computer/Information Standards and are kept in centralized, model-level file areas.

Impact Forecasting uses automated procedures to generate submission forms when it is indicated to do so in the form instructions.


G-1 Scope of the Hurricane Model and its Implementation (2/2)

General Standards

Items reviewed by the Professional Team include but not limited to:– The processes used to assure agreement among databases, data files, and

computer source code to presentation slides, technical papers, equations, and internal documents

– The development and features of the ELEMENTS application platform– The process used in preparing client exposure data, importing the data into

ELEMENTS, and handling problematic entries– The loss convergence tests performed for different portfolio analyses– The methodology for developing surface roughness coefficients– The intensity threshold when accounting for the total basin annual count– The source of the sea-surface temperature data and its use in the model– The use of Vmax data in modeled storm genesis for years in HURDAT2 when

pressure data is unavailable– The methodology for calibrating Florida landfall rates– The script used to create Form A-8– The revised business workflow diagram

√ Standard Verified


G-2 Qualifications of Modeling Organization Personnel and Consultants Engaged in Development of the Hurricane Model

General Standards

Impact Forecasting staff and consultants involved in the development of the hurricane model possess a wide range of multi-disciplinary skills in meteorology, structural engineering, actuarial science, statistics, computer science, and risk analysis.

Model developers have advanced degrees and the majority hold PhDs in their fields of expertise.

The model and submission documentation have been thoroughly reviewed by individuals holding suitable qualifications within the professional disciplines listed above and are outlined in Disclosure G-2.2.

The resumes of the personnel engaged in the development of the hurricane model and responsible for the current hurricane model and the submission are reviewed by the Professional Team.



G-3 Insured Exposure Location (1/2)

General Standards

The model uses United States Postal Service ZIP Code data that is post processed by third-party vendors, including Zip-Codes.com and GreatData. The issue date of the current iteration is October 2019.

ZIP Code centroids are population weighted. ZIP Code boundary and centroid data is examined by Impact Forecasting for

consistency and quality. All ZIP Code-dependent components, such as vulnerability tiers, ZIP Code

Events, ZIP Code Terrain Factors, and ZIP Code Gust Factors (see Disclosure G-3.4) are recreated using the latest update of the ZIP Code data in the model.

Impact Forecasting uses industry-proven geocoding API (Pitney Bowes) to convert street addresses to location coordinates that are routinely quality-checked to ensure accuracy.


G-3 Insured Exposure Location (2/2)

General Standards

Items reviewed by the Professional Team include but not limited to:– The geographic displays of ZIP Code boundaries and centroids for the entire state– The methodology for population-weighted windspeeds to represent ZIP Code

windspeeds for both stochastic and historical storms– The number of ZIP Codes used in completion of the submission forms– The development and implementation of the ZIP Code Events database, the ZIP

Code Terrain Factor database, the ZIP Code Gust Factor database, and the Vulnerability Tiers database

– The process for validating ZIP Code data from the third-party providers, ZIP boundaries from ZIP-Codes.com and population-weighted centroids from GreatData

– Examples of ZIP Code data quality assurance testing– The treatment of ZIP Code centroids in uninhabitable terrain or over water– Examples of geocoding for complete and incomplete street addresses– The process and examples of assigning ZIP Codes to latitude-longitude points



G-4 Independence of Hurricane Model Components

General Standards

The meteorological, vulnerability, and actuarial components of the Impact Forecasting Florida Hurricane Model are theoretically sound and developed independently.

Each component is validated individually without consideration of possible biases in other components.



G-5 Editorial Compliance

General Standards

The Impact Forecasting submission has been compiled, edited, and reviewed by a person with experience in authoring and reviewing technical documents for consistency, grammar, and editorial accuracy.

The web-based collaboration platform, Microsoft SharePoint, is used to manage document control of the submission.

Each signatory on Expert Certification Forms G-1 through G-7 was responsible for the final review of their section, in addition to preliminary reviews throughout the submission production process.

Items reviewed by the Professional Team include but not limited to:– The documentation process for compiling and reviewing the submission document,

including review by each section author and signatory– A snapshot of the initial submission modification history– The flowchart defining the process for creating submission forms and integration into

the submission document– The expert certification in updated Form G-7



Aon Benfield

3Meteorological

Standards

Meteorological Standards



M-1 Base Hurricane Storm Set


The IF Florida Hurricane Model development and validation used the National Hurricane Center HURDAT2 data as of November 25, 2019, incorporating the period 1900-2018.

The complete Base Hurricane Storm Set has been used without trending, weighting, or partitioning.

The modeled hurricane frequencies agree well with the historical observations for each region and each category defined by the Saffir-Simpson Hurricane Wind Scale (see Form M-1).

Items reviewed by the Professional Team include but not limited to:– The external data sources used in development of the hazard model– The methodology for stochastic event modeling– The histograms of observed annual frequency to modeled using a negative binomial

distribution and a Poisson distribution– The methodology for computing by-passing hurricane frequencies– The landfall frequency goodness-of-fit chi-square tests by region for FL and neighboring states– The annual occurrence rates in Form M-1 compared to Form S-1– The comparisons of historical to modeled distributions of forward speed, heading angle, and

central pressure at 2x2-degree cells and the goodness-of-fit in each cell



M-2 Hurricane Parameters and Characteristics (1/2)


All hurricane parameters and characteristics are modeled using well-accepted methods from peer-reviewed scientific and technical literature. The methods are implemented and validated by IF scientists using historical data. A complete list of references is provided in Disclosure G-1.6.

The dependencies among variables in the windfield component and how they are represented are discussed in Disclosure M-2.2

All hurricane parameters are treated the same in the historical and stochastic storm sets. The stochastic events are modeled using data and formulas utilized to create the historical storms.

The IF Florida Hurricane Model models the 3s gust wind speed at 10m height. The conversion of sustained wind speed to gust wind speed considers the wind averaging time as well as the terrain conditions.


M-2 Hurricane Parameters and Characteristics (2/2)


Items reviewed by the Professional Team include but not limited to:– The use of Vmax data in modeled storm genesis for years in HURDAT2 when pressure data is

unavailable– The model domain and 5x5-degree cell assignments for storm track propagation and intensity– The process used for calibrating the stochastic model– The calculations and distributions for forward speed, heading angle, central pressure, and

Vmax. The comparisons of modeled and historical distribution fits– The decay of winds over land. The scatter plot of modeled to observed Vmax– The regression equation for Rmax. The graphical comparison of modeled to observed Rmax,

including the model mean as a function of intensity.– The comparison of historical to modeled annual landfall occurrence rates by coastal segment

for Category 1-2 hurricanes and for Category 3-5 hurricanes– The autocorrelation of the error for forward speed– The simulated central pressure time series for FL landfalling and by-passing storms– The Extended Best Track Rmax frequency distribution– The maximum number of time steps allowed in the model and how stochastic tracks are

terminated– The calculation for ZIP Code averaged population-weighted windspeeds for the 1 km by 1 km

grid points– The external data sources used to generate the model windfields



M-3 Hurricane Probability Distributions (1/2)


The probability distributions of hurricane parameters and characteristics are modeled using historical data. Validation including statistical tests is performed to ensure that the modeled statistics agree well with the observations. Details of the statistical tests can be found in Disclosure S-1.6.

The modeled hurricane frequencies at landfall match the observed frequencies obtained from the Base Hurricane Storm Set for different hurricane categories (1 to 5 based on the Saffir-Simpson Hurricane Wind Scale) and different geographic regions (coastal segments). The comparisons between modeled and observed hurricane frequencies are provided in Form M-1.

The landfall intensity of each hurricane found in either the Base Hurricane Storm Set or the model simulations is defined using maximum 1-minute sustained wind speed. The classification of hurricane categories uses wind speeds in statute miles per hour and follows the Saffir-Simpson Hurricane Wind Scale.


M-3 Hurricane Probability Distributions (2/2)


Items reviewed by the Professional Team include but not limited to:– The validation and goodness-of-fits for track direction, heading angle, Vmax, and

Rmax distributions– The forward speed, heading angle, central pressure, and Vmax goodness-of-fit tests

for Alabama/Mississippi and Georgia landfalls– The probability distributions and data sources provided in Form S-3– The code for simulating time series of relative intensity



M-4 Hurricane Windfield Structure (1/2)


Windfields generated by the IF Florida Hurricane Model are consistent with observed historical storms affecting Florida. Comparisons of the model-generated winds to the observed historical hurricane winds show good agreement. The comparisons are provided in Disclosure M-4.8.

The land use and land cover database used in the IF model is National Land Cover Database (NLCD) 2011. The land use and land cover data are converted into surface roughness coefficients using methodologies available in published literature (Hansen, 1993; ESDU, 2002a, 2002b).

The effects of the vertical variation of winds for multi-story buildings are considered in the vulnerability curves.


M-4 Hurricane Windfield Structure (2/2)


Items reviewed by the Professional Team include but not limited to:– The Willoughby et al. (2006) parametric windfield and the equations for its

parameters– The wind profile adjustments for inflow angle and translation-induced asymmetry– The use of the National Land Cover Database (NLCD, 2011) as the source of land

use land cover data– The process for converting surface roughness lengths to roughness coefficients– The contour maps of peak gust windspeeds for Hurricane Charley (2004), Hurricane

Wilma (2005), Hurricane Irma (2017), and Hurricane Michael (2018). The scatter plots of modeled versus historical windspeeds for each storm.

– The comparisons of modeled to observed time series of peak gust windspeeds for Hurricane Charley (2004), Hurricane Dennis (2005), Hurricane Irma (2017), and Hurricane Michael (2018)

– The map depicting the spatial distribution of model surface roughness and coefficients

– The maps of the spatial distribution of winds for the LaborDay03 (1935) and NoName09 (1945) storms



M-5 Hurricane Landfall and Over-Land Weakening Methodologies (1/2)


The hurricane inland decay rates used by the IF Florida Hurricane Model are developed using historical data based on a methodology published in recent scientific literature (Vickery, 2005). Details of the model along with the validation analysis that was performed are described in Disclosures M-5.1 and M-5.2.

The IF Florida Hurricane Model uses roughness coefficients to capture the effect of terrain on wind speed, including wind speed transition due to terrain changes from over-water to over-land. The roughness coefficients are developed based on the methodology provided in ESDU (2002a, 2002b).

Hurricane intensity (i.e., central pressure and maximum sustained wind speed) changes as the storm transitions from over-water to over-land. Some hurricane parameters, such as radius of maximum winds and wind profile parameters, which are dependent on hurricane intensity, also change accordingly.

Decay models were developed separately for non-continental U.S. land masses. The decay rates for stochastic and historical hurricanes are treated the same.


M-5 Hurricane Landfall and Over-Land Weakening Methodologies (2/2)


Items reviewed by the Professional Team include but not limited to:– The inland decay model– The regression fits for each region using historical data– The plots comparing modeled to historical over-land decay rates– The comparisons of the modeled windfield with historical observed windspeeds for

Hurricane Andrew (1992), Hurricane Jeanne (2004), and Hurricane Irma (2017). Reviewed contour maps of the windfield footprints at and following landfall.



M-6 Logical Relationships of Hurricane Characteristics


The magnitude of asymmetry increases with an increase in translation speed, all other factors held constant, as described in Disclosure M-6.1.

The mean wind speed decreases with increasing surface roughness, all other factors held constant, as described in Disclosure M-6.2.

Items reviewed by the Professional Team include but not limited to:– The methodology for calculating surface roughness and the impact on modeled

windspeeds– The Form M-3– The equation for Rmax– The histogram of HURDAT2 central pressure by hurricane categories– The modeled relationship between central pressure and Rmax– The windfield asymmetry factor– The animations and snapshots of hurricane winds demonstrating the role of storm

translation in windfield asymmetry and the impact of surface roughness on windspeed



Aon Benfield

4Statistical Standards

Statistical Standards



S-1 Modeled Results and Goodness-of-Fit


Hurricane parameters and characteristics are modeled using regression equations and probability distributions fitted to historical data. These equations and distributions are developed and validated based on methods documented in current scientific and technical literature.

Good agreement is observed between modeled and measured statistics of various hurricane parameters and characteristics in different geographic regions. Details of the goodness-of-fit tests performed at landfall can be found in Disclosure S-1.6.

Items reviewed by the Professional Team include but not limited to:– The annual storm count probability distribution– The probability distributions associated with Rmax, Vmax, inland decay, heading

angle, forward speed, and central pressure– The supporting evidence for the adequacy of the match between historical and

modeled results– The use of goodness-of-fit tests to provide metrics in the calibration process– The handling of far-field pressure in the model– Forms S-1, S-2, S-3, and S-4



S-2 Sensitivity Analysis for Hurricane Model Output


Impact Forecasting has assessed the sensitivity of temporal and spatial outputs with respect to the simultaneous variation of input variables using currently-accepted scientific and statistical methods.

The sensitivity analysis results for hurricane loss costs can be found in Form S-6. The most sensitive aspects of the model include the far field pressure (FFP), inland decay rate (α) and central pressure (Cp).

The results of the temporal sensitivity analysis for modeled wind speeds are shown in Disclosure S-2.1.

Items reviewed by the Professional Team include but not limited to:– The Sensitivity Analysis distribution choices, numerical and graphical results and

conclusions as given in Form S-6– The animations of sensitivity analysis results



S-3 Uncertainty Analysis for Hurricane Model Output


Impact Forecasting has performed an uncertainty analysis on the temporal and spatial outputs of the hurricane model using currently accepted scientific and statistical methods. The analysis has identified and quantified the extent to which input variables impact the uncertainty in model output as input variables are simultaneously varied.

The uncertainty analysis results for hurricane loss costs can be found in Form S-6. The far field pressure (FFP) is the major contributor to the uncertainty in hurricane loss costs for Category 1 and Category 3 storms. For Category 5 hurricanes, Rmax has the greatest impact.

The temporal uncertainties of the modeled wind speeds are presented in Disclosure S-3.1.

Items reviewed by the Professional Team include but not limited to:– The Uncertainty Analysis distribution choices, numerical and graphical results and

conclusions as given in Form S-6– The animations of uncertainty analysis results



S-4 County Level Aggregation



The IF Florida Hurricane Model stochastic event set is constructed using 200,000 years’ worth of synthetic hurricanes. IF increased the sample size (number of simulation years) until the error in hurricane loss cost associated with the sampling process was negligible.

The contribution error in hurricane loss cost is lower than 2.5% of the estimated loss cost value for every county in FL.

Items reviewed by the Professional Team include but not limited to:– The convergence results at the county level attributable to the sampling process with

the 200,000 years of simulation– The process to reduce the event set size while maintaining estimation performance


S-5 Replication of Known Hurricane Losses


The model reasonably replicates incurred hurricane losses for 12 historical events from 2004 through 2018 from four insurance companies without significant bias. This is the case for both personal and commercial residential lines and by county, coverage type, and construction class.

Form S-4 was reviewed by the Professional Team.



S-6 Comparison of Projected Hurricane Loss Costs


The historical and modeled annual average statewide hurricane loss costs based on the 2017 FHCF personal and commercial residential zero deductible exposure data are $2.81 and $3.37 billion, respectively. Such difference exists due to different sample sizes of historical and stochastic storm sets and various uncertainties within.

The 95% confidence interval on the statewide average annual loss cost produced using the list of historical hurricanes in the Base Hurricane Storm Set is $1.55 billion to $4.08 billion. Since the modeled personal and commercial residential hurricane loss cost ($3.37 billion) is within the range, the difference between the historical and modeled hurricane loss costs is not statistically significant.

Form S-5 was reviewed by the Professional Team.



Aon Benfield

5Vulnerability Standards

Vulnerability Standards



V-1 Derivation of Building Hurricane Vulnerability Functions (1/3)


The vulnerability functions for the Impact Forecasting Florida Hurricane Model are developed based on sound engineering principles and are calibrated and validated using claims data from multiple historical events and peer reviewed research publications. – The engineering-based vulnerability functions are developed using a component-

based vulnerability approach. The uncertainties in both loading and resistance for each component are accounted for using probability distributions. The model methodology is described in detailed in Disclosures V-1.5, V-2.3, and V-3.3.

– The vulnerability functions were calibrated and validated using insurance claims data totaling USD $11.8 billion. These data are described in Disclosure V-1.3.

The residential building stock classifications are identified based on a review of construction inventories using information from the U.S. Census, tax appraiser’s data, public domain reference information such as the Federal Emergency Management Agency (FEMA), Florida Hurricane Catastrophe Fund (FHCF), and post-event reconnaissance surveys. In addition, the residential building stock classification was compared with exposure data from historical claims. These resources were used to compile standard building types for personal and commercial residential buildings.




The Impact Forecasting exposure module classifies buildings by their primary and secondary risk characteristics. – Primary: occupancy type, construction type, year built, tier (vulnerability region),

number of stories.– Secondary: roof age, roof-to-wall connections, roof covering, roof sheathing, roof

type, floor size, opening protection, garage doors. Vulnerability functions for personal residential, commercial residential, and

manufactured homes are derived separately and are included in the model. The vulnerability functions for appurtenant structures are modeled separately

using the same vulnerability functions as for buildings. – The Impact Forecasting Florida Hurricane model has the capability of choosing

different vulnerability functions for appurtenant structures and primary structures based on the actual building characteristics.

The model begins to estimate damage when the 1-minute sustained wind speed at 10-meter height is greater than 35.2 mph (3-second peak gust wind speed at 10-meter height is greater than 20 m/s). – This is consistent with fundamental engineering principals and has been validated

using claims data from historical events.




The vulnerability functions used in the Impact Forecasting Florida Hurricane Model account for damage to buildings from wind pressure, water infiltration, and missile impacts associated with hurricanes.

The damage due to flood (including hurricane storm surge and wave action) are not included in the IF Florida Hurricane Model.

Items reviewed by the Professional Team include but not limited to:– The building stock classifications– The probability distribution for capacity for a sample of building components– The development and implementation of the windborne debris model– The calibration of the model vulnerability functions– Examples of post-event site investigations– Vulnerability documentations– Year-built bands based on code enforcement and construction practices for non-

manufactured homes and for manufactured homes– The development and the basis for the vulnerability tiers– Example building and associated appurtenant structure vulnerability functions– Form V-1 and the process to complete the form



V-2 Derivation of Contents Hurricane Vulnerability Functions


The preliminary contents vulnerability functions were developed based on the Impact Forecasting damage simulator using engineering analysis and judgement. – These contents vulnerability functions provide loss behavior as a function of the

building damage ratio. – For every building vulnerability function, there is a corresponding contents

vulnerability function. The contents vulnerability functions were calibrated and validated using

detailed coverage-level insurance claims data based on the Impact Forecasting historical event catalog.

Items reviewed by the Professional Team include but not limited to:– The development and validation of the contents vulnerability functions– A scatter plot of modeled versus claims mean content damage ratios– Samples of contents vulnerability functions– The calibration and validation of contents vulnerability functions using claims data– The contents to building relationship in Form V-1



V-3 Derivation of Time Element Hurricane Vulnerability Functions (1/2)


The preliminary time element vulnerability functions were developed using a component-based model combined with a restoration process modeling using the estimation of expected restoration times (Program Evaluation and Review Technique/Critical Path Method) to determine time element losses (Seal, 2001). – The cost and daily output data for each repair activity were obtained from RSMeans

(Mewis, 2020). – The time element vulnerability functions describe loss behavior as a function of the

building damage ratio. The time element vulnerability functions were calibrated and validated using

historical insurance claims data. The time element vulnerability functions do not explicitly distinguish between

direct loss (i.e., losses related to building repair time) and indirect loss (e.g., losses caused by infrastructure service disruption under hurricane wind, storm surge, or flood). However, since insurance claims data from historical events were used to calibrate the time element vulnerability functions, the effects of local and regional infrastructure disruptions are implicitly accounted for within the time element vulnerability functions to the extent that such losses are included in the claims.


V-3 Derivation of Time Element Hurricane Vulnerability Functions (2/2)


Items reviewed by the Professional Team include but not limited to:– The scatter plot of modeled versus claims mean time-element damage ratios– The calibration and validation methodology– The relationship of time-element to building damage ratios in Form V-1



V-4 Hurricane Mitigation Measures and Secondary Characteristics (1/2)


The secondary risk characteristics (secondary modifiers) are used to account for the effects of hurricane mitigation measures. They were developed through wind engineering analysis and then calibrated based on post-disaster damage survey reports, insurance claims data, and engineering judgement.

Key secondary modifiers were identified based on structural engineering expertise and post-event reconnaissance of historical hurricane damage.

The set of secondary modifiers available for model users to select are provided in Disclosure V-4.4.

The secondary modifiers vary by primary vulnerability function and wind speed. They are multiplied to the primary vulnerability functions and can increase, decrease, or maintain the primary building vulnerability functions.


V-4 Hurricane Mitigation Measures and Secondary Characteristics (2/2)


The effects of multiple secondary modifiers on the building damage ratio are combined with a multiplicative methodology.– Limits (both lower and upper bounds) are implemented to control the maximum

change when many secondary characteristics are simultaneously applied. In the absence of any mitigation or secondary risk characteristics, the model

uses the primary vulnerability curve without any modification. Items reviewed by the Professional Team include but not limited to:

– The hurricane mitigation measures and secondary characteristics and example plots – The development, calibration, and implementation of secondary modifiers– The methodology and impact of applying single or multiple secondary modifiers– The range of loss costs by construction type relative to residential single-family

occupancy, unknown construction, and unknown year-built– The range of loss costs by opening protection and by roof covering relative to

unknown secondary modifiers– The probability distributions for roof covering, sheathing, and roof-to-wall connection – Forms V-2 and V-3



Aon Benfield

6Actuarial

Standards

Actuarial Standards



A-1 Hurricane Model Input Data and Output Reports (1/3)

Actuarial Standards

Any adjustments, edits, inclusions, or deletions made to insurance company or other input data are based upon accepted actuarial, underwriting, and statistical procedures. Sensitivity analyses were conducted to assess how certain assumptions affect the estimated losses to ensure the reasonableness of assumptions and data mappings.– The insured value and insurance limits are provided as separate inputs by the user.

No adjustments are made to these values within the model.– The IF Florida Hurricane Model does not make assumptions of depreciation and

does not reduce insured hurricane losses on account of depreciation. Users have the option to code the actual cash values instead of the replacement cost values into the hurricane model input file.

– The IF Florida Hurricane Model can distinguish among various personal and commercial residential policy forms based on occupancy type and construction class (for manufactured homes).

– Sensitivity analyses results were disclosed in Form A-6.



Actuarial Standards

All modifications, adjustments, assumptions, inputs and input file identification, and defaults necessary to use the hurricane model are actuarially sound and are included with the hurricane model output report and in the Impact Forecasting documentation. Treatment of missing values required to run the hurricane model is actuarially sound.– Treatment of missing values is discussed in A-1 Disclosure– The input forms used by the hurricane model are provided in Appendix E: Hurricane Model

Input Form.– Analysis options for Florida rate making are in compliance with Florida Statute FS 627.0628.

– IF model generates a suite of import, analysis and output reports for users to conduct quality control and for regulators to confirm assumptions.



Actuarial Standards

Items reviewed by the Professional Team include but not limited to:– Data format requirements and the process for importing data into the ELEMENTS

framework– Sample model output reports disclosing assumptions, post-import summaries, and

model settings



A-2 Hurricane Events Resulting in Modeled Hurricane Losses (1/2)

Actuarial Standards

Modeled hurricane losses and hurricane probable maximum loss levels reflect all insured wind-related damages from storms classified as landfalling or by-passing hurricanes that produce minimum damaging wind speeds or greater on land in Florida, as described in Disclosure A-2.1.– The calculation of hurricane loss costs and hurricane probable maximum loss levels

for Florida includes damage from landfalling and by-passing hurricanes. Damage is included in the calculation of hurricane loss costs and probable maximum losses from the time when the hurricane first reaches damaging wind speeds on land in Florida.

The Impact Forecasting Florida Hurricane Model only provides users the option to model wind-only losses. Storm surge losses are excluded from any Florida hurricane analysis by the Impact Forecasting Florida Hurricane Model.– Appendix F shows only wind loss can be selected in the IF Florida Hurricane Model.


A-2 Hurricane Events Resulting in Modeled Hurricane Losses (2/2)

Actuarial Standards

Items reviewed by the Professional Team include but not limited to:– The process for combining exposure, hazard, and footprint data used to generate

losses – The methodology for determining ground-up losses. Reviewed examples of ground-

up and gross loss calculations.– Discussed that loss convergence tests were performed to address the effects of

varying portfolio sizes. Reviewed table of the minimum number of samples necessary for convergence of portfolios of different sizes.

– The criteria for identifying by-passing hurricanes and selected tracks– Discussed that the model computes wind and storm surge losses separately.– Discussed that the Florida Hurricane model only provides users the option to model

wind losses.



A-3 Hurricane Coverages (1/2)

Actuarial Standards

Building: The vulnerability functions are developed based on well-established structural and wind engineering science in literature and calibrated and validated by analyzing historical hurricane claims data. In addition, Aon is a major sponsor of the Insurance Institute for Business and Home Safety (IBHS) and serves on the IBHS Research Advisory Board. Aon and Impact Forecasting personnel have participated in several post-hurricane site surveys. The knowledge acquired during these independent studies has been used to enhance the calibration and validation of the vulnerability functions.

Appurtenant structure: The Impact Forecasting Florida Hurricane Model calculates appurtenant structure hurricane loss costs separately from building, contents, and time element hurricane loss costs. The Impact Forecasting Florida Hurricane Model estimates hurricane losses for appurtenant structure coverage associated with personal and commercial residential properties using the same method as for building coverage, as described in Disclosure A-3.1, but using the input appurtenant structure coverage values. The IF model can assign a different vulnerability function for appurtenant structures if the appurtenant structure is separately coded with different characteristics in the input file.

Contents: The Impact Forecasting Florida Hurricane Model estimates hurricane losses for contents coverage associated with personal and commercial residential properties using the method described in Disclosure A-3.1, but using the input contents coverage values. The contents damage functions are based on building damage ratio.


A-3 Hurricane Coverages (2/2)

Actuarial Standards

Time Element: The Impact Forecasting Florida Hurricane Model estimates hurricane losses for time element coverage associated with personal and commercial residential properties using the method described in Disclosure A-3.1, but using the input time element coverage values. The time element damage functions are based on building damage ratio.

Items reviewed by the Professional Team include but not limited to:– The methodology for producing building, appurtenant structure, contents, and time-

element loss costs– A calculation of frame-owners loss costs in Form A-1 and PMLs for ZIP Code 32202

in Duval County– Discussed that law and ordinance coverage and loss assessment coverage for

condo-unit owners are not explicitly considered in the model.– Discussed the Actuarial Standards signatory’s review of the model submission under

the Actuarial Standards. Discussed how IF actuary attested the model results to be actuarially sound.



A-4 Modeled Hurricane Loss Cost and Hurricane Probable Maximum Loss Level Considerations (1/2)

Actuarial Standards

Hurricane losses and probable maximum loss levels generated by the IF Florida Hurricane Model do not include expenses, risk load, investment income, premium reserves, taxes, assessments, or profit margins.

The Impact Forecasting Florida Hurricane Model does not make a prospective provision for economic inflation.

The Impact Forecasting Florida Hurricane Model only provides users the option to model wind-only losses. Storm surge losses are excluded from any Florida hurricane analysis by the Impact Forecasting Florida Hurricane Model.

In the Impact Forecasting Florida Hurricane Model, hurricane loss costs and hurricane probable maximum loss can be produced at the location or site level. Hurricane loss costs and hurricane probable maximum loss levels can then be calculated at any geographic level, such as ZIP Code, county, state, etc.


A-4 Modeled Hurricane Loss Cost and Hurricane Probable Maximum Loss Level Considerations (2/2)

Actuarial Standards

Demand surge is included in the IF model’s calculation of hurricane loss costs and probable maximum loss levels. The demand surge function is developed based on analysis of the pricing information for the construction sector from Xactimate and XactContents (Olsen & Porter, 2011a; Xactware, 2020) before and after hurricanes occurred between 1990 and 2019. The methods and assumptions underlying the demand surge factors are actuarially sound and described in Disclosure A-4.3.

Items reviewed by the Professional Team include but not limited to:• The hurricanes event losses corresponding to Form A-8• The demand surge methodology• The calculation of demand surge factors and implementation in the model• The methodology for determining probable maximum loss levels and the

methodology for computing the Aggregate Loss Distribution• The calculation of the mean and standard deviation of the annual average loss

based on the per-occurrence stochastic event set using 200,000 years of simulation

• The treatment of deductibles in the claims data



A-5 Hurricane Policy Conditions (1/2)

Actuarial Standards

Policy deductibles and policy limits are developed based on well-established insurance theory as discussed in Disclosure A-5.1.

The relationship among the modeled deductible hurricane loss costs is reasonable. When other variables are held constant, loss costs decrease as deductibles increase, as shown in Form A-6.

The Impact Forecasting Florida Hurricane Model is capable of handling annual deductibles. It complies with s. 627.701(5)(a), F.S.

Regarding policy exclusions, if the user removes a coverage amount from the input file, such as wind exclusion policies, the model would not generate losses for the excluded coverage. Loss settlement and other exclusions that are not explicitly reflected in the exposure data cannot be quantified by the model. However, loss settlement and other exclusions may be implicitly included in the model results to the extent that the claims data used to calibrate the model reflected these conditions.


A-5 Hurricane Policy Conditions (2/2)

Actuarial Standards

Items reviewed by the Professional Team include but not limited to:– Example use-case scenarios for applying policy limits and deductibles and examples

of gross loss calculations and application of policy limits and deductibles– Application of annual hurricane deductibles when multiple events occur in a given

year – The methodology for processing insurer claims data used for model validation



A-6 Hurricane Loss Outputs and Logical Relationship to Risk (1/2)

Actuarial Standards

IF’s methods, data and assumptions used to estimate losses are actuarially sound.

Loss costs exhibit logical relationships to risk in all cases, regarding– Construction type– Policy forms– Presence of mitigation features (covered in Vulnerability Standards)– Building code – Deductibles– Coverages– Geographic

Form A-6 demonstrates all of these logical relationships to risk.


A-6 Hurricane Loss Outputs and Logical Relationship to Risk (2/2)

Actuarial Standards

Items reviewed by the Professional Team include but not limited to:– Form A-1 losses for ZIP Code 33921 by construction type and coverages– Discussed that the three uncertainty intervals in Part B and Part C of Form A-8 were incorrectly

calculated. Discussed the reasons for the error and actions taken to correct and prevent the problem from recurring.

– The methodology for calculating the return periods for each range in Part A of Form A-8 as well as the frequency and severity distributions

– Example of modeling a coinsurance policy– The variation in loss costs for ZIP Codes 33001 and 33045 in Monroe County and for ZIP

Codes 32329 and 32323 in Franklin County– Maps of loss costs by county for the different construction and policy types in Form A-4– The loss costs in Form A-4 where frame loss costs are less than masonry loss costs and the

underlying reasons for the results– Loss costs in Form A-4 for Glades and Lafayette counties– Form A-6 and the reasonableness checks of the loss costs performed by the modeler– Revised Form A-8 for reasonableness– Maps of loss costs by ZIP Code showing the effects of land friction– The correction of Form A-6, the reasons for the error and actions taken to correct and prevent

the problem from recurring



Aon Benfield

7Computer/Information Standards

Computer/Information Standards



CI-1 Hurricane Model Documentation (1/2)


IF maintains the following three sets of documentation related to hurricane model development and implementation:– Model development documentation created by the Research and Development (R&D)

team detailing all components, formulas, and test data for the hurricane model;– Technical notes produced by the Software Team based on model methodology

described in the R&D document. The technical notes describe architectural and design strategies to implement the hurricane model in the ELEMENTS platform;

– High-level documentation created for end users which describes the components of the hurricane model and their usage in the ELEMENTS platform.

All documents are maintained in Microsoft Team Foundation Server (TFS) and are version controlled.

Documents describing model requirements, system architecture, design strategy, implementation details, and the user interface are produced by the Software Team. In addition, a document related to testing plans and test results is produced by the QA (Quality Assurance) Team.

The IF Software Team produces requirements, architecture, and design documents separately from the source code. These documents are used by the QA Team to prepare test cases and perform tests.


CI-1 Hurricane Model Documentation (2/2)


This is the first IF Florida Hurricane Model submission. IF will maintain a table of all substantive changes after the initial submission. Any changes to the accepted model will be tracked going forward for all subsequent submissions.

Items reviewed by the Professional Team include but not limited to:– The version control branching strategy in TFS used to support parallel development

of multiple models– The list of all externally-acquired hurricane model-specific software and data sources– Multiple documentations

• Model Development Process• Model Requirements for the US Hurricane Model• Hurricane Model Change Policy and Process• R&D Team Foundation Server Integration Process and Best Practices• Wind Vulnerability Simulator• Hazard Component Equation Variable Mapping and Logic Flows• Flowchart Standard Reference Documentation• R&D Code Internal Test Documentation• Statistical Standard Equation Variable Mapping and Logic Flows• Vulnerability Component Equation Variable Mapping and Process Flow Diagrams



CI-2 Hurricane Model Requirements


IF maintains documentation for major software components along with a database schema, technical notes, a user guide, an installation guide, and a deployment guide for successful deployment and usage of the models on the ELEMENTS platform.

IF produces various types of internal (i.e., used only by IF) and external (i.e., client-facing) documents at different phases of the model development process. Documents produced and maintained by the Software Team include: – Software Development Process– Feature Requirement Specification– Architecture/Design Document– Technical Notes– Test Plans and Test Cases– Database Schema– Product Release Notes (Internal and External)– ELEMENTS User Guide– Input Data Format– ELEMENTS Installation and Deployment Guide– Additional Tools User Guide– Coding Standard

The software requirements documentation is reviewed by the Professional Team.



CI-3 Hurricane Model Organization and Component Design (1/2)


Impact Forecasting develops and maintains documents which describe the database schema and relationships between data, including flowcharts to show the flow of information and links between data from various components of the software system. This includes schema of the Exposure/Results database and Model database.

Interaction between software components and sub-components is captured in architecture and design documentation, including the interactions between hazard and vulnerability modules.

A separate document describing interactions between multiple IF groups and team members is maintained using a diagram.

All flowcharts developed and maintained by Impact Forecasting follow industry standards based on ISO 5807, Unified Modeling Language (UML) and Business Process Model and Notation (BPMN).


CI-3 Hurricane Model Organization and Component Design (2/2)


Items reviewed by the Professional Team include but not limited to:– The control and data flowcharts and verified the compliance of the flowcharts with

the Impact Forecasting Flowchart Standards– An example of database schema– A revised diagram illustrating the flow of information among teams– Flowcharts in the submission and supporting document



CI-4 Hurricane Model Implementation (1/2)


IF maintains the following:– Coding guidelines and a handbook of coding best practices– A network organization diagram involving all the servers and hardware components

involved in running the hurricane model on the ELEMENTS platform– A list of all components affecting hurricane loss cost and probable maximum loss

calculations along with implementation details, such as number of lines of code and number of lines of comments for each component and sub-component

All model files and databases provided by the R&D group are stored and backed up regularly. The Software Team compares production-ready files and databases with original copies using MD5 Hash comparison.

Descriptions of all static components and interactions between components are documented at various levels of detail using flowcharts and other depictions. This documentation can be used to track implementation details in actual code.


CI-4 Hurricane Model Implementation (2/2)


All components including Classes and Methods are sufficiently commented in the code as per industry guidelines. All Classes are commented with each class’s purpose and linkage to other Classes. All methods are commented with Input and Output parameters along with inner workings of the method, exceptions, and logic. An important line of code is commented in detail with linkage to work items in the source control system (TFS) when applicable.

A list of all components referred to in Standard G-1 and used in software implementation is maintained and tracked. A cross-referenced list of equations and variables used in implementation is documented and maintained.

Items reviewed by the Professional Team include but not limited to:– The script responsible for the generation of Form A-8– The analyzer report containing the number of lines of code with and without

comments by project– The ELEMENTS network organization diagram– The implementation of demand surge factors– Multiple documents, including the coding guidelines for languages used by the

modeler and software developer



CI-5 Hurricane Model Verification (1/3)


The model testing and verification methods include:– General verification

• Primary developers from the R&D team perform and document the changes.• Non-primary developers from the R&D team peer review and approve the changes.• Results and documentation of changes are shared with the software team. • The software team independently implements these changes in the production-ready

software following industry standard processes. • The QA team performs various tests in which results (hazard and losses) are compared

with original runs performed by model developers.• Upon successful comparison, changes are approved and released in the final product.

– Component testing• The software team uses the Team Foundation Server (TFS) Unit testing framework to

perform unit tests on all components when applicable. All unit test cases are documented and stored in the TFS repository.

• Unit tests for all components are performed where applicable. For all eligible components and sub-components, an input, along with the expected output, is defined and tests are run to match expected results. Only upon passing the unit tests is the code allowed to be checked in as a part of the final build.

• In addition to the unit tests, automated tests, which go through sets of sub-components and check the correctness of the results, are also performed.




The model testing and verification methods include:– Component testing (cont’d)

• Sets of regression tests are documented and performed with each iteration of internal and production releases. New regression tests are added with newly added features to complete the coverage of all cases.

• End-to-end tests involving all hurricane model components are performed regularly by the QA team with each internal and production release. Multiple test cases are created to make sure all possible interactions among sub-components and components are covered.

– Data testing• Specific test cases are developed for components which access model data from a

database or files and are compared against expected results. • These tests are run periodically with every internal and production release cycle. • The Unified Function Test (UFT) suite by MicroFocus is used to automate test cases in

addition to manual testing to verify communications between various components. • The combination of log files, intermediate files, and SQL databases are also used to verify

data flow and inter-component communication. • Microsoft Excel and Word are used to document the testing process and its outcome.• A data hash mechanism is used when copying data from one server to another to ensure

that no data are corrupted which can corrupt results. All data used by the hurricane model is regularly backed up.




Items reviewed by the Professional Team include but not limited to:– An example summary report that is produced for each testing cycle– Examples of ZIP Code data quality assurance testing– The series of logical tests performed on the loss cost relationships in Form A-6– The testing software used, and additional manual test cases performed– An example of a unit test and the test results– The geocoder testing and verification documentation



CI-6 Hurricane Model Maintenance and Revision (1/2)


Impact Forecasting created policies and processes to be followed when changing model components in any form. This includes the production of a business case which includes the driving factor(s) that necessitate the change and the objective and impact of the change.

The IF Florida Hurricane Model periodically updates to reflect new learning and advancement in understanding of the model components, a new version identifier will be assigned when the updates result in changes to the FL residential hurricane loss costs or PMLs.

IF uses Team Foundation Server (TFS) as the code, documentation (including feature and change requests and requirements), and requirements repository. – Any changes to the model components are tracked using TFS source control system

which records the change, author, and date/time of the change. Impact Forecasting will maintain a list of hurricane model changes after the

initial submission. A unique version identifier will be assigned according to the versioning scheme established by IF (see Disclosure CI-6.2).


CI-6 Hurricane Model Maintenance and Revision (2/2)


Items reviewed by the Professional Team include but not limited to:– The software development process– The policy for Model and ELEMENTS platform versioning– Examples of Model and ELEMENTS platform release numbering schemes– The version history for the model and the ELEMENTS platform– The model and platform revision management. Reviewed an example of the models

hosted under the ELEMENTS platform.



CI-7 Hurricane Model Security


Impact Forecasting documents and follows security processes to secure access to the code, data, and documentation as prescribed by company’s policy and industry standard. The security processes include:– Secured Computer Access

• Office Security• Data Center Access

– Secured Hurricane Model• Firewall and Network Security• User Account Access• Anti-virus

– Secured Documentation, Software, and Data• Code• Data Security• Documentation

Items reviewed by the Professional Team include but not limited to:– The data and network security procedures– The data retention and recovery protocols



Aon Benfield

Questions?


DisclaimerLegal DisclaimerAon’s Reinsurance Solutions business, part of Aon UK Limited (for itself and on behalf of each subsidiary company of Aon plc) (“Aon”) reserves all rights to the content of this report (“Report”). This Report is for distribution to Aon and the organisation to which it was originally delivered only. Copies may be made by that organisation for its own internal purposes but this Report may not be distributed in whole or in part to any third party without both (i) the prior written consent of Aon. and (ii) the third party having first signed a “recipient of report” letter in a form acceptable to Aon. Aon cannot accept any liability to any third party to whom this Report is disclosed, whether disclosed in compliance with the preceding sentence of otherwise. To the extent this Report expresses any recommendation or assessment on any aspect of risk, the recipient acknowledges that any such recommendation or assessment is an expression of Aon opinion only, and is not a statement of fact. Any decision to rely on any such recommendation or assessment of risk is entirely the responsibility of the recipient. Aon will not in any event be responsible for any losses that may be incurred by any party as a result of any reliance placed on any such opinion. The recipient acknowledges that this Report does not replace the need for the recipient to undertake its own assessment. The recipient acknowledges that in preparing this Report Aon may have based analysis on data provided by the recipient and/or from third party sources. This data may have been subjected to mathematical and/or empirical analysis and modelling. Aon has not verified, and accepts no responsibility for, the accuracy or completeness of any such data. In addition, the recipient acknowledges that any form of mathematical and/or empirical analysis and modelling (including that used in the preparation of this Report) may produce results which differ from actual events or losses. The Aon analysis has been undertaken from the perspective of a reinsurance broker. Consequently this Report does not constitute an opinion of reserving levels or accounting treatment. This Report does not constitute any form of legal, accounting, taxation, regulatory or actuarial advice.

Limitations of Catastrophe ModelsThis report includes information that is output from catastrophe models of Impact Forecasting, LLC (IF). The information from the models is provided by Aon Benfield Services, Inc. (Aon) under the terms of its license agreements with IF. The results in this report from IF are the products of the exposures modelled, the financial assumptions made concerning deductibles and limits, and the risk models that project the pounds of damage that may be caused by defined catastrophe perils. Aon recommends that the results from these models in this report not be relied upon in isolation when making decisions that may affect the underwriting appetite, rate adequacy or solvency of the company. The IF models are based on scientific data, mathematical and empirical models, and the experience of engineering, geological and meteorological experts. Calibration of the models using actual loss experience is based on very sparse data, and material inaccuracies in these models are possible. The loss probabilities generated by the models are not predictive of future hurricanes, other windstorms, or earthquakes or other natural catastrophes, but provide estimates of the magnitude of losses that may occur in the event of such natural catastrophes. Aon makes no warranty about the accuracy of the IF models and has made no attempt to independently verify them. Aon will not be liable for any special, indirect or consequential damages, including, without limitation, losses or damages arising from or related to any use of or decisions based upon data developed using the models of IF.

Additional Limitations of Impact Forecasting, LLCThe results listed in this report are based on engineering / scientific analysis and data, information provided by the client, and mathematical and empirical models. The accuracy of the results depends on the uncertainty associated with each of these areas. In particular, as with any model, actual losses may differ from the results of simulations. It is only possible to provide plausible results based on complete and accurate information provided by the client and other reputable data sources. Furthermore, this information may only be used for the business application specified by Impact Forecasting, LLC and for no other purpose. It may not be used to support development of or calibration of a product or service offering that competes with Impact Forecasting, LLC. The information in this report may not be used as a part of or as a source for any insurance rate filing documentation.THIS INFORMATION IS PROVIDED “AS IS” AND IMPACT FORECASTING, LLC HAS NOT MADE AND DOES NOT MAKE ANY WARRANTY OF ANY KIND WHATSOEVER, EXPRESS OR IMPLIED, WITH RESPECT TO THIS REPORT; AND ALL WARRANTIES INCLUDING WARRANTIES OF MERCHANTABILITY AND FITNESS FOR APARTICULAR PURPOSE ARE HEREBY DISCLAIMED BY IMPACT FORECASTING, LLC. IMPACT FORECASTING, LLC WILL NOT BE LIABLE TO ANYONE WITH RESPECT TO ANY DAMAGES, LOSS OR CLAIM WHATSOEVER, NO MATTER HOW OCCASIONED, IN CONNECTION WITH THE PREPARATION OR USE OF THIS REPORT.

Impact Forecasting

Documents

Transcript of Impact Forecasting