The Rms Hurricane Model

THE RMS HURRICANE MODEL

Submitted in Compliance with the 2002 Standards of the Florida Commission on Hurricane Loss Projection Methodology

February 2003

Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 1 FLORIDA COMMISSION ON HURRICANE LOSS PROJECTION METHODOLOGY

Model Identification

Name of Model and Version: RiskLink Version 4.3a 4.2 SP1a

Name of Modeling Company: Risk Management Solutions, Inc.

Street Address: 7015 Gateway Boulevard

City, State, Zip: Newark, CA 94560

Mailing Address, if different from above: Risk Management Solutions Limited 10 Eastcheap London EC3M 1AJ U.K.

Contact Person: Brian Owens

Phone Number: +44 (0)20 7256 3807 Fax Number: +44 (0)20 7256 3838

E-mail Address: [email protected]

Date: February 2003

Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc FLORIDA COMMISSION ON HURRICANE LOSS PROJECTION METHODOLOGY

Table of Contents

TABLE OF CONTENTS I

LIST OF FIGURES IV

LIST OF TABLES V

2002 STANDARDS 1

5.1 General Standards 1 5.1.1 Scope of the Computer Model and Its Implementation 1 5.1.2 Qualifications of Modeler Personnel and Independent Experts 1 5.1.3 Model Revision Policy 2 5.1.4 Independence of Model Components 2 5.1.5 Risk Location 3 5.1.6 Identification of Units of Measure and Conversion Factors 3 5.1.7 Visual Presentation of Data 3

5.2 Meteorological Standards 4 5.2.1 Units of Measure for Model Output 4 5.2.2 Damage Function Wind Inputs 4 5.2.3 Official Hurricane Set or Suitable Approved Alternatives 5 5.2.4 Hurricane Characteristics 5 5.2.5 Landfall Intensity 6 5.2.6 Hurricane Probabilities 7 5.2.7 Hurricane Probability Distributions 9 5.2.8 Land Friction 10 5.2.9 Hurricane Overland Weakening Rate 12 5.2.10 Temporal and Spatial Wind Field Characteristics 13

5.3 Vulnerability Standards 13 5.3.1 Derivation of Vulnerability Functions 13 5.3.2 Required Vulnerability Functions 15 5.3.3 Wind Speeds Causing Damage 15 5.3.4 Construction and Codes 16 5.3.5 Mitigation Measures 16 5.3.6 Additional Living Expenses (ALE) 18

5.4 Actuarial Standards 18 5.4.1 Underwriting Assumptions 18 5.4.2 Actuarial Modifications 19 5.4.3 Loss Cost Projections 19

Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc i 5.4.4 Insurer Inputs 19 5.4.5 Demand Surge 20 5.4.6 Logical Relation to Risk 21 5.4.7 Deductibles and Policy Limits 23 5.4.8 Contents 24 5.4.9 Additional Living Expenses (ALE) 25 5.4.10 Replication of Known Hurricane Losses 26 5.4.11 Comparison of Estimated Hurricane Loss Costs 28 5.4.12 Output Ranges 29

5.5 Computer Standards 34 5.5.1 Primary Document Binder 34 5.5.2 Requirements 34 5.5.3 Model Architecture and Component Design 34 5.5.4 Implementation 35 5.5.5 Verification 35 5.5.6 Model Maintenance and Revision 36 5.5.7 User Documentation 37

5.6 Statistical Standards 37 5.6.1 Use of Historical Data 37 5.6.2 Comparison of Historical and Modeled Results 37 5.6.3 Uncertainty Characterization 38 5.6.4 Sensitivity Analysis for Model Output 38 5.6.5 Uncertainty Analysis for Model Output 38 5.6.6 County Level Aggregation 39

MODULES 40

MODULE 1 41

I. General Description of the Model 41 A. In General 41 B. Loss Costs 61 C. Other Considerations 63

II. Specific Description of the Model 68 A. Model Variables 68 B. Methodology 72 C. Validation Tests 77

MODULE 2 80

Background/Professionalism 80 1. Company Background 80 2. Professional Credentials 82 3. Multi-discipline Team 92 4. List of Clients 94

MODULE 3 99

Meteorology - Hurricane Set 99

Hurricane Wind Field 108

Vulnerability Functions 109

Insurance Functions 111

Average Annual Loss Functions 118

General 138

Data Flow Chart 139

TESTS 141

Form A 141

Form B 142

Form C 144

Form D 145

Form E 148

Form F 149

OUTPUT RANGES 154

ATTACHMENTS 219

ATTACHMENT A 220

ATTACHMENT B 221

ATTACHMENT C 224

Figure 5.1 Sample Visualization 4 Figure 5.2 Segments Used for Parameter and Rate-Smoothing 8 Figure 5.3 Comparison of Historical and Modeled Rates 9 Figure 5.4 Variation in Friction Coefficients for Florida ZIP Codes 12 Figure 5.5 Sample Event Claims Data - Wood Frame Construction 15 Figure 5.6 Relative Loss Costs by ZIP Code and Historical Hurricane Landfalls 22 Figure 5.7 Example Loss Distribution 24 Figure 5.8 Relative Structure and Contents Damage Ratios: Actual Claims Data 25 Figure 5.9 Relative Structure and ALE Damage Ratios: Actual Claims Data 26 Figure 5.10 Industry Loss Estimates (Residential and Commercial) for Recent Storms 27 Figure 5.11 Company Specific Loss Comparisons 28 Figure 5.12 Changes in Loss Costs by County 33 Figure 1.I.1 Mean Translational Velocities for ‘Type 2’ Hurricanes on a 2º x 2º Grid 45 Figure 1.I.2 150 Simulated ‘Type 2’ Tracks 45 Figure 1.I.3 Segments Used for Parameter and Rate-Smoothing 46 Figure 1.I.4 Sketch Showing Hurricane Model Parameters 47 Figure 1.I.5 Model Flow Chart 55 Figure 2.3.1 Business Workflow Diagram 94 Figure 3.I.1 Modeled Degradation Rates Compared to the Kaplan-DeMaria Curve 102 Figure 3.I.2 Modeled 102-year Historical and 100-year Stochastic One-Minute Windspeeds (mph) (thematic plots) 103 Figure 3.I.3 Modeled 102-year Historic and 100-year Stochastic One-Minute Windspeeds mph) (contour plots) 104 Figure 3.I.4 Comparison of Historical and Modeled Rates (category assigned based on 1-minute windspeeds) 105 Figure 3.V.1 Observed Damage (dots) vs. Modeled Damage (solid line) Curve 119 Figure 3.V.2 Relative Loss Costs (Woodframe – Ground Up) by ZIP Code and Historical Hurricane Landfalls (category based on 1-minute windspeed) 120 Figure 3.V.3 Ground-up loss costs for frame, masonry and mobile home 122 Figure 3.V.4 Percentage Change in County Loss Costs from Previous Year 126

Table 5.1: Differences in County Level Loss Costs and Principal Drivers of Change 30 Table 1.I.1 Model Variables 56 Table 1.I.2 Model Variables With Range of Possible Values 57 Table 1.I.3 Building Classification 59 Table 1.I.4 Impacts on Loss Costs Using Engineering Modifications 63 Table 1.II.1 Primary Databases Used by the Model 71 Table 1.II.2 Sample of Datasets Used for Development and Calibration of Vulnerability Functions 72 Table 1.II.3 Sample Client Loss Data Comparison 78 Table 1.II.4 Comparison of Actual and Estimated Industry Loss ($ million) 78 Table 2.2.1 Individuals Involved in Meteorological Aspects of the Model 90 Table 2.2.2 Individuals Involved in Vulnerability Aspects of the Model 91 Table 2.2.3 Individuals Involved in Actuarial Aspects of the Model 91 Table 2.2.4 Individuals Involved in Computer Science Aspects of the Model 91 Table 2.2.5 Individuals Involved in Statistical Aspects of the Model 92 Table 2.3.1 Sample List of Clients 95 Table 2.3.2 Mix of Company Clients Over the Last 5 Years 95 Table 2.3.3 Time Since Ratemaking Clients Became Clients 96 Table 3.I.1 Saffir-Simpson Hurricane Scale (for displayed parameters) 99 Table 3.I.2 Hurricane Parameters 103 Table 3.I.3 Historical and Modeled Hurricane Characteristics for Florida 106 Table 3.I.4 Model Results Probability of Hurricanes by Year 107 Table 3.IV.1 Example of Insurer Loss Calculation 111 Table 3.IV.2 Comparison 1: Mobile Home Losses - Various Events 115 Table 3.IV.3 Comparison 2: Hurricane Hugo - Losses by Construction Class 115 Table 3.IV.4 Comparison 3: Hurricane Andrew - Losses by Coverage Type 115 Table 3.IV.5 Comparison 4: Hurricane Andrew - Losses by County 116 Table 3.IV.6 Comparison 5: Various Companies - Total Losses 116 Table 3.V.1 Percentage change in weighted average loss costs from previous year statewide 124 Table 3.V.2 Percentage change in weighted average loss costs from previous year by region 124 Table 3.V.3 Percentage change in weighted average loss costs from previous year by coastal and inland counties. 125

Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc v Table 3.V.4 Monetary Contribution to the Average Annual Loss for each Storm in the Official Storm Set 126 Table 3.V.5 Contribution from Hurricane Andrew for Each Affected ZIP Code 128 Table 3.V.6 Model Results - Distribution of Hurricanes by Size 136 Table 3.VII.1 Summary of Form F Input and Output Files 149 Table 3.VII.2 Filling parameter values used for the first quantile input 150

FLORIDA COMMISSION ON HURRICANE LOSS PROJECTION METHODOLOGY

2002 Standards

5.1 General Standards

5.1.1 Scope of the Computer Model and Its Implementation

The computer model shall project loss costs for personal lines residential property from hurricane events, excluding flood and storm surge, except as flood and storm surge apply to Additional Living Expense (ALE). References to the model throughout the Standards shall include its implementation.

If the modeler uses historical data that include losses from flood and storm surge, then the modeler shall disclose the techniques employed to exclude such losses, and those techniques shall be based on accepted scientific methods.

If the modeler uses engineering or other data that include losses from flood and storm surge, then the modeler shall disclose the techniques employed to exclude such losses, and those techniques shall be based on justifiable methods.

RMS hurricane model loss costs for personal lines residential property include losses to the following coverages, as appropriate to the type and composition of the policy form in question: primary structures, appurtenant structures, contents, and additional living expenses. Output from the model can explicitly and separately define the losses to each of these coverages.

While the RMS hurricane loss model is able to estimate hurricane losses from flood and storm surge, this is an optional analytical feature that must be specifically “turned on” to include such losses. If this option is not activated, flood and storm surge losses are excluded from all loss cost projections. All loss cost projections provided herein exclude flood and storm surge. Loss data used in developing the RMS hurricane model loss does not include surge losses.

ALE loss functions have been calibrated/validated with actual hurricane event ALE coverage losses. In performing the vulnerability function development and calibration for ALE, the impact on ALE losses due to storm surge damage to infrastructure was not excluded, and was therefore included where applicable.

5.1.2 Qualifications of Modeler Personnel and Independent Experts

Model construction, testing, and evaluation shall be performed by modeler personnel or independent experts who possess the necessary skills, formal education, or experience to develop hurricane loss projection methodologies.

The model or any modifications to an accepted model shall be reviewed by modeler personnel or independent experts in the following professional Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 1 2002 Standards: 5.1 General Standards

disciplines, if relevant: structural/wind engineering (licensed Professional Engineer (PE)), statistics (advanced degree), actuarial science (Associate or Fellow of Casualty Actuarial Society or Member of the American Academy of Actuaries), meteorology (advanced degree), and computer science/engineering (advanced degree). These individuals shall abide by the standards of professional conduct adopted by their profession.

Overall, RMS employs approximately 6084 engineers and scientists who participate in various areas of model development (not all on U.S. hurricane). The team possesses a wide range of multi-disciplinary skills in engineering, the physical sciences, actuarial science, data development, data analysis and numerical modeling, computer science/engineering and quality assurance engineering. Of this staff, about 80% hold advanced degrees and approximately 3121 possess Ph.D. level qualifications in their fields of expertise. Module 2 “Background/Professionalism” documentation provides the background of those individuals specifically involved in the development of the RMS hurricane model. Those individuals possess the necessary skills, formal education, and experience, in all required disciplines, to develop hurricane loss projection methodologies and abide by the standards of professional conduct adopted by their profession.

5.1.3 Model Revision Policy

The modeler shall have developed and implemented a clearly written policy for model revision with respect to methodologies and data. The modeler shall clearly identify the model version under review. Any revision to any portion of the model that results in a change in any Florida residential hurricane loss cost must be accompanied by a new model version number.

The RMS hurricane model is periodically enhanced to reflect advances in RMS’ knowledge of hurricanes and the consequences of hurricanes. RMS scientists perform research on an ongoing basis to improve their understanding of the phenomena, and keep well apprised of developments in the general research and scientific community.

The process of model revision and release is rigorous and well documented. The RMS Quality Assurance Department maintains, archives, and documents the features of each release.

The current submitted model is RiskLink version 4.3a.4.2 SP1a

5.1.4 Independence of Model Components

The meteorology, vulnerability, and actuarial components of the model shall each be demonstrated to be theoretically sound without compensation for potential bias from the other two components. Relationships within the model among the meteorological, vulnerability, and actuarial components shall be demonstrated to be reasonable.

In the RMS hurricane model, vulnerability, meteorological, and actuarial functions are theoretically sound and are developed independently without compensation for potential bias from the other two components. For example, vulnerability functions relating damage ratios to Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 2 2002 Standards: 5.1 General Standards

windspeeds are fixed within the model and are not dependent on other aspects of the loss model. Relationships within the model among the meteorological, vulnerability, and actuarial components are reasonable.

5.1.5 Risk Location

Zip codes used in the model shall be updated at least every 24 months using information originating from the United States Postal Service. The United States Postal Service issue date of the updated information shall be disclosed.

Zip code centroids, when used in the model, shall be based on population data and shall be visually demonstrated to be reasonable.

Zip code information purchased by the modeler shall be verified by the modeler for accuracy and appropriateness.

It is RMS’ policy to update ZIP Codes using information ultimately derived from the United States Postal Service at least every 24 months. The vintage of the ZIP Code data used in the submitted model is April 2001. A proprietary database is being used for assigning a geographical coordinate to a location for which information is input at ZIP Code level. If the building location is input at a street level, then the proprietary geocoding software is used to determine the actual latitude and longitude based on the street address. If the building location is entered as a ZIP Code, then by default for each event, the model uses windspeeds which are exposure weighted averages of windspeeds across the ZIP. the location of the exposure is taken at the centroid of the ZIP Code. The ZIP Code centroids are weighted by population.

5.1.6 Identification of Units of Measure and Conversion Factors

All units of measure for model inputs and outputs shall be clearly identified. All conversion factors used by the model shall be disclosed.

All model output of length, wind speed, and pressure are in the units of statute miles, statute miles per hour, and millibars, respectively. All units of measure for model inputs and outputs are clearly identified. All conversion factors used in the development of the model are standard factors supported by scientific literature which can be shown to the Professional Team during their on-site visit.

5.1.7 Visual Presentation of Data

Visualizations shall be accompanied by legends and labels for all elements. Individual elements shall be clearly distinguishable, whether presented in original or copy form.

a. For data indexed by latitude and longitude, by county or by zip code, a color contour map and a continuous tone map with superimposed county and zip code boundaries shall be produced.

b. Florida Map Colors: Maps will use two colors, blue and red, along with shades of blue and red, with dark blue and dark red designating the lowest and highest quantities, respectively. The color legend and associated map shall be comprised of an appropriate number of intervals to provide readability.

RMS uses appropriate visualization techniques that enable the graphs, charts and maps to be clearly presented and understood. Sample visualization is shown in Figure 5.1. Additional visualizations are available throughout the documentation as well as will be available for on-site review by the Professional Team.

Figure 5.1 Sample Visualization

5.2 Meteorological Standards

5.2.1 Units of Measure for Model Output All model outputs of length, wind speed, and pressure shall be in units of statute miles, statute miles per hour, and millibars, respectively.

All model output of length, wind speed, and pressure are in the units of statute miles, statute miles per hour, and millibars, respectively. During the Professional Team visit, units used in output files can be presented and reviewed.

5.2.2 Damage Function Wind Inputs

Wind inputs to the damage function shall be in units consistent with currently used wind measurement units and/or shall be converted using standard meteorological/engineering conversion factors which are supported by literature and/or documented measurements available to the Commission.

Wind inputs to the damage function are in units consistent with currently used wind measurement units, and are converted into such form using standard conversion factors, which are supported by scientific literature.

5.2.3 Official Hurricane Set or Suitable Approved Alternatives

Modelers shall include in their base storm set all hurricanes, including by- passing hurricanes, which produce hurricane force winds in Florida. The storm set, derived from the Tropical Prediction Center/National Hurricane Center (TPC/NHC) document Tropical Cyclones of the North Atlantic Ocean, 1871- 1998, updated through the 2001 hurricane season and/or the HURDAT (HURricane DATa) data set, is found in the Report of Activities as of November 1, 2002 under Section VII, Compliance With Standards and Related Information, #4 (Base Storm Set). All proposed alternatives to the characteristics of specific storms in the storm set shall be subject to the approval of the Commission.

The storm set used to develop the RMS hurricane model for Florida includes both landfalling and non-falling hurricanes and matches the storm set provided by the Commission, including storms through 20010.

5.2.4 Hurricane Characteristics

Methods for depicting all modeled hurricane characteristics including but not limited to wind speed, radial distributions of wind and pressure, minimum central pressure, radius of maximum winds, strike probabilities, and tracks shall be based on information documented by scientific literature or modeler information accepted by the Commission.

Each of the methods for depicting hurricane characteristics in the RMS hurricane model is based on scientific information documented by scientific literature or RMS research. The major components of the hurricane hazard model are outlined below.

Landfall (“strike”) probabilities. Calibration of landfall probabilities is performed on a series of segments, approximately 50 nautical miles (nmi) in length that bound Florida the entire US coastline. The target historical probabilities are computed from the historical database by application of a smoothing algorithm that corrects for inappropriate variability in the historical record due to the limited length of record. The stochastic model is then calibrated to match the historical rates of landfall. For Florida the most important gates are those along the coast of Florida and its neighboring states, including Georgia, Alabama and Mississippi.

Storm track characteristics (parameter) probabilities. Calibration of forward speeds is performed by computing probability density functions (pdf) are measured for each of Florida’s 50 nmi segments for of forward speed from the historical record. velocity, and angle of landfall (azimuth). Due to the limited length of the historical record, the calibration is performed at a regional level by grouping neighboring gates together. the target, historical pdfs are computed from the historical database by application of a smoothing algorithm. Extrapolations, primarily

by including larger regional behavior, are also made to reflect the probability of occurrences beyond the range of the observed record at a gate.

Landfall pressure distributions Probability density functions (pdf) of landfall pressure are computed for each coastal segment around the US coast using historical landfall data for that segment supplemented by track data from nearby offshore regions. The pdfs are smoothed by normalizing landfall rates by Cat against the historical record at a regional level.

Modeling of storm tracks. Storm tracks are simulated using a random-walk Monte Carlo technique. This method creates realistic synthetic events, which preserve the statistical behavior of the historical events (mean and variance of translational velocity). Tracks are simulated in two steps. Firstly, the tracks are created and secondly pressure histories are added to the tracks. Each simulated event has a uniform annual rate of occurrence.

The track model is calibrated across the Atlantic basin by comparing the rates of storms crossing a grid of cells covering the basin. More detailed calibration is performed at coastlines by calculating the rate of crossing and pdf of forward speed and direction on linear gates along the coastlines. Pressure histories for each track are calibrated so landfall rates by Cat match the historical targets on each gate.

The original Monte Carlo track set is importance sampled to reduce the number of events used for loss calculation.

Windfield model. The windfield model calculations determine the localized wind speed associated with a storm event (historical or stochastic). The wind speed is calculated at a site identified by its latitude and longitude, taken either from a street address specific geocode or derived from the weighted centroid of a ZIP Code. The key storm parameters used in wind speed calculations include central pressure, radius-to-maximum wind and wind profile, wind direction, storm forward velocity and direction, landfall location, surface roughness, and track.

The windfield model used by RMS is based upon published scientific literature and RMS analysis of empirical windfield data. The theoretical and analytical formulations of the Wind Field Module are taken from a methodology originally developed at the Boundary Layer Wind Tunnel, University of Western Ontario, Canada (Georgiou 1983, 1985). See Module 1 for further details.

Inland dissipation of wind. RMS models the inland dissipation of wind as resulting from two distinct physical phenomena. The first is related to the filling (decay) of central pressure as a storm moves inland from water; the second is related to the degradation of surface wind due to the friction effects of the airflow over land.

5.2.5 Landfall Intensity

Models shall use maximum one-minute sustained 10-meter wind speed when defining hurricane landfall intensity. This applies both to the base storm set adopted in 5.2.3 used to develop landfall strike probabilities as a function of Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 6 2002 Standards: 5.2 Meteorological Standards

coastal location and to the modeled winds in each hurricane which causes damage. The associated maximum one-minute sustained 10-meter wind speed shall be within the range of wind speeds (in statute miles per hour) categorized by the Saffir-Simpson scale.

Saffir-Simpson Hurricane Scale (for displayed parameters): A scale from 1 to 5 that measures hurricane intensity.

Category Winds (mph) Central Pressure Damage (MB)

1 74 – 95 > 980 Minimal 2 96 – 110 965 - 979 Moderate 3 111 – 130 945 - 964 Extensive 4 131 – 155 920 - 944 Extreme 5 Over 155 < 920 Catastrophic

Saffir-Simpson Intensity is not a parameter of the RMS hurricane model, and is not used in any way for the calculation of wind speeds or associated loss costs. However, the stochastic (simulated) storms in the RMS model, defined by a series of hurricane characteristics, can be classified in terms of Saffir-Simpson Intensity measured either by the maximum one-minute sustained wind speed, or central pressure. In any case, however, the assignment of storm category by pressure for a simulated storm is consistent with the assignment based upon the modeled windspeed.

5.2.6 Hurricane Probabilities

Modeled hurricane probabilities shall reasonably match the historical record through 2001 for category 1 to 5 hurricanes, shall be consistent with those observed for each geographical area of Florida, and shall be displayed in vertical bar graphs. “Consistent” means: (1) spatial distributions of modeled hurricane probabilities shall accurately depict vulnerable coastlines in Florida and the states of Alabama, Georgia, and Mississippi; and (2) probabilities are compared with observed hurricane frequency using methods documented in accepted scientific literature or proposed by the modeler and accepted by the Commission.

Modeled landfall probabilities are consistent with what has been observed historically for each geographical area of Florida. The model is consistent both in terms of the total rate of landfalling hurricanes by region, and the rate of hurricanes of various intensities by region. The methodology for developing these landfall probabilities is soundly based on scientific methods.

In the RMS hurricane model, the probabilities of hurricane landfall are measured along a series of 50 nautical mile (nmi) coastal segments (each approximately 50 nmi in length) that bound Florida’s geography. Figure 5.2 shows the layout of the 50 nmi intervals along the coast of Florida, Georgia, Alabama and Mississippi.

NW NE

SE SW

Figure 5.2 Segments Used for Parameter and Rate-Smoothing

The probabilities are computed from the historical database by the application of a smoothing algorithm. The algorithm makes adjustments for non-representative variability in the historical record due to the limited length of record, and in doing so, properly considers the Florida Keys and the significant coastal discontinuity at the southern tip of the state. Figure 5.3 shows the results of the smoothing process by comparing the number of historical landfalling hurricanes to what is estimated by the model over a comparable period of time.

Historical 30 Entire State Modeled 25 20 15 10 5 0 12345

1.5 Northeast 1.0

0.5

16 0.0 Northwest 12 12345 8 10 4 Southeast 0 12345 5

0 10 12345 8 Southwest 6 3 By-Passing 4 2 2 0 1 12345 0 12345 Figure 5.3 Comparison of Historical and Modeled Rates (category assigned based on 1-minute windspeeds) Note: for storms with multiple landfalls, individual landfalls are counted separately

5.2.7 Hurricane Probability Distributions

Modeled probability distributions for hurricane intensity, eye diameter, forward speed, radii for maximum winds, and radii for hurricane force winds shall be consistent with historical hurricanes in the Atlantic basin as documented in accepted scientific literature available to the Commission.

Modeled probability distributions for hurricane intensity, eye diameter, forward speed, radii for maximum winds and radii of hurricane force winds are consistent with observed historical hurricanes in the Atlantic Basin. The basis for developing probability distributions for each of these parameters is the record of historical hurricanes. For certain parameters, such as central pressure and forward speed velocity, the smoothing technique discussed in Standard 5.2.6 is utilized. Probability distributions for parameters and the details of the methodology used to smooth the historical data and derive the probability distributions are available for on-site review by the Professional Team.

The modeled maximum wind speed and minimum central pressures are documented and well supported by RMS analysis and the scientific literature. The modeled maximum one-minute sustained 10-meter windspeed and the minimum central pressure for storms landfalling in Florida are 180 188 mph and 883 887 mb, respectively, both consistent with the Atlantic Basin historical extremes. In addition to a well-supported upper bound on hurricane intensity, the RMS hurricane model utilizes a sound and defensible approach to developing the probabilities of extreme events throughout Florida. Details of the sea-surface temperature based methodology used to develop the upper bound intensity of Florida hurricanes and the methodology used to develop extreme event probabilities will be available for review by the Professional Team.

5.2.8 Land Friction

Land friction shall be used in the model to reduce wind speeds over land, shall be based on scientific methods, and shall provide realistic wind speed transitions between adjacent zip codes, counties, and territories. The magnitude of friction coefficients shall be consistent with accepted scientific literature, consistent with geographic surface roughness, and shall be implemented with appropriate geographic information system data.

Land friction effects are modeled in the RMS hurricane model. The methodology is based on scientific methods and provides realistic windspeed transitions between adjacent ZIP Codes, counties, or territories areas of different surface roughness. The magnitude of the friction coefficients is consistent with accepted scientific literature. The following discussion provides an overview of the methodology used for incorporating the impact of land friction.

Impact of Roughness Conditions. The wind field model estimates the cumulative effect of roughness on the hurricane wind field at a location and upstream along the path of which the wind is coming from. Therefore, the direction from which the wind approaches, which may change over time, is considered when determining the impact of roughness on windspeeds. The further inland the airflow travels, and the more dense the localized roughness conditions, the greater the magnitude of the friction coefficient. Impact of Distance to Coast and Roughness Conditions. The magnitude of friction coefficients is a function of distance to coast (DTC) and localized terrain conditions (Roughness). In the RMS hurricane model these two effects are accounted for in the form of Degradation Curves. These curves provide the reduction in windspeed that is applied depending on the distance a given location is from the coast, and localized Roughness conditions. The further inland the airflow travels, and the more dense the localized Roughness conditions, the greater the magnitude of the friction coefficient.

Determination of Roughness. The starting point for the determination of land friction effects is the creation of a database that describes the surface roughness in terms of the roughness length. The definition of the roughness length arises from standard wind engineering techniques. The use of a roughness length to describe the underlying surface roughness also allows a physically based model to be used to calculate both local and upstream surface roughness effects on the wind speed.

The database itself is created using the National Land Cover Data (NLCD) dataset produced by the USGS. This dataset is derived from early to mid-1990’s Landsat Thematic Mapper satellite data and provides coverage of the entire continental United States at a horizontal resolution of 30-metres, using a 21-class land cover classification scheme. Further processing of areas classified as urban or suburban in this database is then undertaken by RMS to differentiate areas of differing building heights. This is done primarily using data on the construction square footage by ZIP Code. At the same time, those land cover classes whose effects on the surface wind speed are similar are merged into a single land use class. The end result is a 10-class land cover database with land cover classes ranging from water to high-rise buildings. Finally, a representative roughness length is assigned to each of the 10 land cover classes, using published mapping schemes from the scientific literature. In the RMS hurricane model, the DTC is determined by calculating the distance of the location to a digitized high resolution coastline. In Florida, the distance to coast also depends on the direction in which the storm approaches. A Roughness index is assigned to each zip code based on the following data:

• U.S. Census housing and population density data • Databases representing construction square footage by ZIP Code • Land use land cover (LULC) from the United States Geological Survey describing various forms of natural and man made terrain.

Quantification of Roughness Impacts. Coefficients describing the impact of land friction are then calculated by using the roughness database in conjunction with GIS software to sample both the local and upstream roughness conditions by direction at each point of interest. Both local and upstream roughness conditions are sampled because the wind speed at a particular location is determined by the local surface roughness and also by any changes in the surface roughness conditions upwind of the location being considered. As the upstream roughness will generally vary with direction about a particular location, sampling of the upstream roughness must also be undertaken by direction. Information on the sampled roughness length values and their distance from the location are then used in conjunction with a physically based model to determine an appropriate set of coefficients describing the impact of land friction effects at the location by direction. Quantification of DTC and Roughness Impacts. In the RMS hurricane model, the magnitude of the friction coefficients is quantified from RMS’ analysis of empirical hurricane data and review of the existing scientific literature on this topic. Windfield envelope contours for a number of hurricanes have been digitized, normalized for pressure filling, and analyzed to determine the impact of land friction on windspeed as a function of distance to coast. Local variability in the friction impacts due to localized terrain and Roughness is quantified through a detailed analysis of actual event claims data. These analyses result in a reliable representation between roughness conditions, windspeed reductions, and resulting damage.

The variation in windspeed reduction in the RMS hurricane model in Florida is illustrated in Figure 5.4. Each point on the chart represents a ZIP Code in Florida and the friction coefficient (0.85 = a 15% reduction in wind speed) windspeed reduction factor used by the model for that specific ZIP Code. As can be seen, there is a clear trend that the reduction in wind speed due to friction effects increases as the distance to coast (DTC) increases. However, at any given DTC, reductions will vary due to the impacts of local roughness conditions, and that in some cases ZIP Codes further inland experience less friction effects than those closer to the coast. However, the range of variability is realistic and can be supported.

Note Each point represents a zip code in Florida

0.95

0.9

0.85

0.8 Windspeed Reduction Factor

0.75

0.7 Distance to Coast

Figure 5.4 Variation in Friction Coefficients for Florida ZIP Codes

5.2.9 Hurricane Overland Weakening Rate

The hurricane overland weakening rate methodology used by the model shall be provided to the Commission and shall be shown to be (1) reasonable as observed in comparison to historical records, and (2) documented in accepted scientific literature or in modeler information accepted by the Commission.

The RMS hurricane model provides realistic weakening rates no less and no greater than observed extremes in historical records for Florida. The windspeed decay for each storm follows the functional form of the Kaplan & DeMaria model. The rate of windspeed decay is fixed once landfall occurs but varies from one storm to another, allowing the stochastic (simulated) storms to properly reflect the significant variation in the filling behavior of the historical storms. The model produces a mean windspeed reduction within 20% of the Kaplan-DeMaria decay value and consistent with historical hurricanes affecting Florida. There are three key components to the pressure filling methodology used in the RMS hurricane model. First, the rate of pressure filling is related to the historical behavior of storms on a regional basis, and thus for Florida principally reflects the experience of Florida storms. the rate of pressure filling is not modeled

as a fixed relationship, but as a variable, allowing the stochastic (simulated) storms to properly reflect the significant variation in the filling behavior of the historical storms. The model allows can accommodate the potential for the re-intensification of a storm’s pressure if it exits over water, allowing for the possibility for subsequent landfalls.

A major benefit of utilizing the above described methodology is that the behavior of the stochastic storms reflects the full range of behavior of historical storms in the region, storms that decay slowly and quickly, and storms that exit land, re-intensify, and make a subsequent landfall, while at the same time remaining consistent with the historical averages and within historic observed extremes. During the Professional Team visit, the methodologies used to model pressure filling and related life cycle storm behavior will be presented.

5.2.10 Temporal and Spatial Wind Field Characteristics

The time variant wind field, including the radial distribution of wind speeds, shall be demonstrated to be consistent with accepted scientific principles, such as: 1. The radius of maximum winds shall reflect specified hurricane characteristics. 2. The magnitude of the asymmetry shall increase as translational speed increases, all other factors held constant. 3. The wind speed shall decrease with increasing surface roughness (friction), all other factors held constant.

The RMS wind field model is a time stepping wind field model, which allows full advantage to be taken of the directionally dependent roughness impacts on surface winds. In this model, the radius of maximum winds is a function of the hurricane characteristics, the magnitude of the asymmetry increases with increasing translational speeds and the wind speeds decrease with increasing surface roughness, all other factors being held constant.

5.3 Vulnerability Standards

5.3.1 Derivation of Vulnerability Functions

Development of the vulnerability functions is to be based on one or more of the following: (1) historical data; (2) tests; (3) structural calculations; (4) expert opinion. Any development of the vulnerability functions based on structural calculations and/or expert opinion shall be supported by tests and historical data to the extent such data are available.

The derivation of the vulnerability functions shall be described and demonstrated to be theoretically sound.

Any modification factors/functions to the vulnerability functions or structural characteristics and their corresponding effects shall be disclosed and shall be clearly defined and their theoretical soundness demonstrated.

The development of vulnerability functions for residential classes of construction in Florida (including mobile homes) for each of structure, contents, and ALE coverages, is principally based upon structural and wind engineering principles and detailed analyses of historical claims data, supplemented by expert input as well as an extensive review of published literature on damage assessment opinion and the results of engineering methods of damage assessment. Damage curves for all classes of construction, including mobile homes, are developed separately. As outlined in the Module 2 documentation, the individuals within RMS involved in the development of vulnerability functions have extensive experience in the field of structural and wind engineering and data analysis.

A major component of this process is the collection and review of hurricane loss data in recent hurricanes. A sample of claims data for wood frame structures from four recent hurricanes is shown in Figure 5.5.

Overall, approximately $7 billion in claims data from various insurance companies in seven recent, significant hurricanes has been utilized in the development of the RMS hurricane vulnerability curves. The general procedure used to process such data included:

a) Collect and audit claims data b) Associate claims data with exposure data in-force at the time of the hurricane c) Develop loss ratios by company by ZIP Code, by construction class, by coverage type d) Associate loss ratios with best estimate of the windfield from each historical hurricane e) Perform regression analysis to develop vulnerability curves from claims data f) Evaluate vulnerability curves against expert opinion and engineering methods g) Validate curves against loss experience from various insurance portfolios h) Validate curve against industry loss across a large set of events.

100.0

10.0

1.0 MDR (%) MDR Andrew Hugo 0.1 Erin Georges Bob Vulnerability Function 0.0 Peakgust (mph) Figure 5.5 Sample Event Claims Data - Wood Frame Construction

Modifications may be made to the base vulnerability functions to take into consideration specific building characteristics or mitigating measures. These modifications are based on an analysis and study of claims data, damage statistics, building codes, engineering studies and wind tunnel experiments. See the responses to Standards 5.3.4 and 5.3.5 for further details.

5.3.2 Required Vulnerability Functions

Vulnerability functions shall separately compute damages for building structures, mobile homes, appurtenant structures, contents, and additional living expense.

Vulnerability functions compute, separately, damage for building structures, mobile homes, appurtenant structures, contents, and additional living expense.

5.3.3 Wind Speeds Causing Damage

Damage associated with a declared hurricane event shall include damage incurred for wind speeds above and below the hurricane threshold of 74 mph. The minimum wind speed that generates damage shall be specified.

Damage associated with a declared hurricane includes damage incurred for wind speeds above and below the hurricane threshold of 74 mph. The minimum wind speed that generates damage is 50 mph (peak gust).

5.3.4 Construction and Codes

In the derivation and application of vulnerability functions assumptions concerning construction type, construction characteristics, new building codes, and revisions to existing building codes shall be demonstrated to be reasonable and appropriate.

As described in Standard 5.3.1, historical event loss data has been used extensively in the development of vulnerability functions. For the loss information used in development of the vulnerability functions, construction characteristics and insured value information is supplied directly to us by our clients. This information is assumed to be correct, but is also subjected to checks by RMS. Summaries of exposure and loss data sets and their use in the development of vulnerability functions will be available for on-site review by the Professional Team.

The model can consider the impact on vulnerability due to changes in building codes, both for buildings and mobile homes. The changes in building codes and building construction practices are modeled through “Year Modifiers” that scale the base vulnerability functions based on the year of construction of the building. The development of the key year bands and year modification factors are a result of RMS research into the changes of building codes and construction practices in Florida as well as across the U.S. A region’s experience with natural catastrophes is also considered in developing the year bands. The development of modification factors is based on comparing analytical models of buildings, simulating building characteristics conforming to various years of construction.

The model can also consider the impact of specific building characteristics (e.g. roof geometry) through the use of Secondary Modifiers that scale the base vulnerability functions as explained in section 5.3.5.

5.3.5 Mitigation Measures

Modeling of mitigation measures to improve a building’s wind resistance and the corresponding effects on vulnerability shall be disclosed and demonstrated to be theoretically sound. These measures shall include, but not be limited to, fixtures or construction techniques that enhance: Roof strength Roof covering performance Roof-to-wall strength Wall-to-floor-to-foundation strength Opening protection Window, door, and skylight strength.

The percentage changes in the statewide, zero deductible personal residential non-mitigated loss costs that would be produced in the output ranges due to each mitigation measure shall be individually and specifically provided to the Commission, including ranges of possible impacts on damage for each mitigation measure listed.

Methods for estimating the effects of mitigation measures shall be shown to be reasonable both individually and in combination.

The RMS hurricane model can consider building specific characteristics or mitigation measures through the application of secondary modifiers The mitigation measures could be building- characteristic specific (e.g. improved roof sheathing or anchors) or external (e.g. storm shutters), and The secondary modifiers modify the base case vulnerability functions according to specific building characteristics or mitigation measures. These characteristics must be specifically selected by the user. The default case is to not include any modifications. If modifiers are selected they are clearly identified in the input files and output reports. Some of the main secondary modifiers that are available in the model are: a) Roof shape b) Roof slope c) Roof covering d) Roof sheathing e) Roof anchors and f) Storm shutters.

The following secondary modifiers are available in the model:

a) Roof covering b) Roof sheathing strength c) Roof anchor d) Roof geometry e) Wind resistance of window openings f) Wind resistance of doors openings and g) Foundation system

RMS uses a component-based vulnerability methodology to model the damage to each component of a building. The vulnerability of each component is based on its failure load, obtained from building codes, results of wind tunnel experiments, field observations and many technical reports. These include building codes and standards such as ASCE (1998), IBC (2000), SBC (1997) and SFBC (1994). The reports and studies include those by FEMA (FEMA 1992), the U.S. Department of Housing and Urban Development (HUD 1993), Natural Hazards Research and Applications Information Center, University of Colorado (Ayscue 1996) and Florida International University (Mitrani et al. 1995 and Peacock et al. 1998). A detailed list of references is given in Module 1. Component vulnerabilities are combined to obtain the overall vulnerability function of buildings with unique building characteristics simulating a range of mitigated, un-mitigated and average buildings. The “mitigated” building is compare with “average” building, using engineering checks, to develop modification factors. The modifiers are finally used to compute changes to industry loss costs resulting in loss cost relativities due to various mitigation measures.

The database of secondary modifiers has been developed by RMS engineering staff, in cooperation with Professors Dale Perry and Norris Stubbs of Texas A&M University, and Professor Peter Sparks of Clemson University. The modifiers are based on an extensive review of research material, building codes, engineering analyses, damage surveys and damage data Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 17 2002 Standards: 5.4 Actuarial Standards

contained in many technical reports. These include building codes and standards such as ASCE (1998), IBC (2000), SBC (1997) and SFBC (1994). The reports and studies include those by FEMA (FEMA 1992), the U.S. Department of Housing and Urban Development (HUD 1993), Natural Hazards Research and Applications Information Center, University of Colorado (Ayscue 1996) and Florida International University (Mitrani et al. 1995 and Peacock et al. 1998). A detailed list of references is given in Module 1. The details of the development of the secondary modifiers will be available for on site review by the Professional Team.

The RMS hurricane model can consider mitigation measures to improve a building’s wind resistance. The mitigation measures could be building characteristic specific (e.g. improved roof sheathing or anchors) or external (e.g. storm shutters). These mitigation measures can be addressed in the model by secondary modifiers that modify the base vulnerability functions according to specific building characteristics or mitigation measures. The secondary modifiers are discussed in standard 5.3.5. The details of the development of the mitigation measures will be available for on site review by the Professional Team

5.3.6 Additional Living Expenses (ALE)

In the estimation of Additional Living Expenses (ALE), the model shall consider hurricane damage including storm surge damage to the infrastructure.

The ALE vulnerability function shall consider the time it will take to repair/reconstruct the home.

In the RMS hurricane model, Additional Living Expense losses include only factors that are hurricane related, are theoretically sound, and consider the time to repair the structure. For personal lines loss cost analyses, ALE losses are determined based upon the estimated damage to the structure. Additionally, ALE loss functions have been calibrated/validated with actual hurricane event ALE coverage losses. In performing this vulnerability function development and calibration for ALE, storm surge damage to infrastructure was not excluded, and so was included where applicable.

5.4 Actuarial Standards

5.4.1 Underwriting Assumptions

When used in the modeling process or for verification purposes, adjustments, edits, inclusions, or deletions to insurance company input data used by the modeler shall be based on accepted actuarial, underwriting, and statistical procedures. The methods used shall be documented in writing.

For damage estimates derived from or validated with historical insured hurricane losses, the assumptions in the derivations concerning (1) construction characteristics, (2) policy provisions, (3) claim payment practices, and (4) relevant underwriting practices underlying those losses shall be identified and demonstrated to be reasonable and appropriate.

As noted in section 5.3.1, historical loss information is used in the development of the RMS vulnerability functions. This information, including construction type, line of business, policy structure, and insured value, is supplied directly to us by our clients. The information is subjected to a review by RMS and peculiarities, if any, are clarified directly with the client. Underwriting practices are assumed to be representative of residential insurance underwriting in general, that is, the vulnerability of property observed in historical events is assumed to be indicative of vulnerability of such property types in future events where the property is subjected to similar wind loads. Detailed summaries of the exposure and loss data sets, a description of the data review process, and the use of the data in the development of vulnerability functions are available for review by the Professional Team.

Any adjustments, edits, inclusions, or deletions to insurance company input are based upon accepted actuarial, underwriting, and statistical procedures and are documented in writing.

5.4.2 Actuarial Modifications

All actuarial modifications made to the model shall be disclosed to the Commission and based on accepted engineering and actuarial criteria.

There are no modifications to actuarial functions beyond the building characteristic modifiers discussed in standard 5.3.5.

5.4.3 Loss Cost Projections

Loss cost projections produced by hurricane loss projection models shall not include expenses, risk load, investment income, premium reserves, taxes, assessments, or profit margin. Hurricane loss projection models shall not make a prospective provision for economic inflation.

RMS loss cost calculations do not include expenses, risk load, investment income, premium reserves, taxes, assessments or profit margin. RMS loss projections do not make any prospective provision for economic inflation. Vulnerability functions project losses as a percentage of coverage values. Coverage values are input by the user and no modifications are made within the program to account for economic inflation.

5.4.4 Insurer Inputs

The modeler shall disclose any assumptions, fixed and/or variable, that relate to insurer input. Such assumptions shall be demonstrated to be actuarially sound. Assumptions that can vary by specific insurer shall be disclosed in a model output report. Fixed assumptions, that do not vary, need to be disclosed to the Commission.

Input data to the RMS hurricane model is explicitly provided by the user for each particular analysis. The model assumes that inputs provided by the user are reflective of actual exposures. Specifically:

Insurance to Value. The model does not make any assumptions regarding insurance to value. The location value and insurance limits are provided as separate input. No adjustments are made to these values within the model.

Demographic Assumptions. The model itself does not make adjustments for demographic differences in exposure unless the user is unable to specify either the type of construction or the specific location of the policy. If the construction class is unknown, the model defaults to a county statewide level average mix of construction in Florida. If the user is unable to specify the ZIP Code, but is able to specify the county, the model allocates the exposure to ZIP Codes within the county in proportion to the appropriate exposure for the line of business under consideration. However, RMS does not perform loss costs analyses for ratemaking purposes on data at the county level, and strongly advises its clients not to do so as well.

Appurtenant Structures. Values of appurtenant structures for each location are a user input. The model does not make assumptions regarding the value of appurtenant structures.

Contents. Contents limits and values are part of the user input. No assumptions are made within the model.

Additional Living Expenses. ALE limits and values are part of the user input. The model assumes that the value represents one year of ALE.

Insurer Exposures by ZIP Code. As part of the analysis process, each location analyzed is “geocoded” (i.e. geographically positioned). If the location does not geo-code (such as if a ZIP Code is invalid), the location is excluded from the analysis. All locations that are not included in the analysis are easily identified. If the analysis is run at ZIP Code level, the exposure is assumed to be distributed across the ZIP. Given that all exposure information is provided as part of the user analysis input, this information can be summarized and clearly identified as part of any rate filing submission. See the Analysis Summary Report in Module 3, Section IV, #11.

5.4.5 Demand Surge

Loss cost projections shall not explicitly include demand surge. Any adjustment to the model or historical data to remove implicit demand surge, shall be disclosed and demonstrated to be reasonable.

Loss cost projections produced by the RMS hurricane model do not explicitly include demand surge. RMS has a separate model for factoring demand surge into loss estimates, but this methodology is not utilized for loss cost and rate making applications. Estimated implicit demand surge impacts in historical claims data used to calibrate the model have been taken into account, specifically loss data for Hurricane Andrew.

5.4.6 Logical Relation to Risk

Loss costs shall not exhibit an illogical relation to risk, nor shall loss costs exhibit a significant change when the underlying risk does not change significantly.

Loss costs generated by RMS do not show an illogical relation to risk nor do they exhibit a significant change when the underlying risk does not change significantly. The general trend is for loss costs to be greatest in areas of past historical hurricane activity and greater on the coast than inland.

1. Loss costs produced by the model shall be positive and non-zero for all Zip Codes.

Loss costs produced by the model are positive and non-zero for all ZIP Codes for which exposure has been matched.

2. Modelers shall produce color-coded maps for the purpose of comparing loss costs by five-digit zip code within each county and on a statewide basis.

As an illustration of the maps requested, see Figure 5.6, which shows the variation in loss costs by ZIP Code along with historical hurricane landfalls by region. See also maps produced for Module 3, Section V

Northeast

Northwest

15 01000

Cat 1 Cat 2 Cat 3 Cat 4 Cat 5

5 3 00

Cat 1 Cat 2 Cat 3 Cat 4 Cat 5 Southeast

Southwest 8 6 5

8 2 1

4 Cat 1 Cat 2 Cat 3 Cat 4 Cat 5 2 11

6 to 16.4 Cat 1 Cat 2 Cat 3 Cat 4 Cat 5 4 to 6 2.01 to 4 1 to 2.01 0.5 to 1 0.16 to 0.5 No Exposure

Figure 5.6 Relative Loss Costs by ZIP Code and Historical Hurricane Landfalls (category assigned based on 1-minute windspeeds)

3. Loss costs cannot increase as friction or roughness increase, all other factors held constant.

Loss costs do not increase as friction or roughness increase, all other factors held constant. In the model, the reduction in windspeed always increases with increased roughness and the calculated losses for all vulnerability functions always increase with increasing windspeed. Therefore, all other factors held constant, loss costs cannot increase as roughness or friction increases (see Section 5.2.8).

4. Loss costs cannot increase as the quality of construction type, materials and workmanship increases, all other factors held constant.

Loss costs do not increase as the quality of construction type, material and workmanship increases, all other factors held constant.

5. Loss costs cannot increase with the presence of fixtures or construction techniques designed for hazard mitigation, all other factors held constant.

The model does not consider the presence of fixtures or construction techniques designed for hazard mitigation explicitly, but includes them as secondary modifiers as explained in section 5.3.7. Loss costs do not increase above those in the absence of such measures, all other factors held constant.

6. Loss costs shall decrease as deductibles increase, all other factors held constant.

Loss costs decrease as deductibles increase, all other factors held constant.

7. Loss costs cannot increase as the quality of building codes and enforcement increases, all other factors held constant.

The model does not explicitly address building code quality and enforcement, but these factors are included as secondary and year modifiers as explained in section 5.3.4 and 5.3.5. Loss costs do not increase as quality increases.

8. The relationship of loss costs for individual coverages (A, B, C, D) shall be consistent with the coverages provided.

The above tests are intended to apply in general. There may be certain anomalies that are insignificant or are explainable by special circumstances. This standard applies separately to each coverage.

The relationship between the loss costs for each separate coverage (A, B, C and D) and the coverages provided is consistent and reasonable.

5.4.7 Deductibles and Policy Limits

The model shall provide a mathematical representation of the distribution of losses to reflect the effects of deductibles and policy limits, and the modeler shall demonstrate its actuarial soundness.

The relationship among the modeled deductible loss costs shall be shown to be reasonable. Differences in these relationships from those previously found acceptable, if applicable, shall be explained and shown to be reasonable. If applicable, changes in the methods used to reflect the effects of policy limits shall be disclosed.

The vulnerability functions in the model produce two outputs at a given level of wind speed: the mean damage ratio (MDR), and a measure of the variation in the MDR, the coefficient of Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 23 2002 Standards: 5.4 Actuarial Standards

variation (CV). From a beta distribution, the model uses the MDR and CV to provide a measure of the variability in damage. Figure 5.7 provides an example of the variability modeled around the expected losses from a hypothetical MDR and CV. Given the loss distribution as illustrated below, and the deductible amount, losses to the insured and insurer are calculated by integrating the loss curve above and below the deductible value, respectively. Using such a distribution provides a mathematical representation of the distribution of losses in a suitable degree of detail to properly reflect the effects of deductibles, coinsurance, and various forms of sub-limits. Probability Density

Loss ratio Figure 5.7 Example Loss Distribution

The quantification of the CVs, and the relationship between the CVs and the MDRs is made from an analysis of the actual claims data discussed in Standard 5.3.1. The relationship among the modeled deductible loss costs is reasonable.

5.4.8 Contents

The model shall provide a separate mathematical representation of contents loss costs, and the modeler shall demonstrate its actuarial soundness.

The relationship between the modeled building and contents loss costs shall be shown to be reasonable. If applicable, differences and the reasons for those differences from prior submissions in the relativities between loss costs for the building and the corresponding loss costs for contents shall be explained and shown to be reasonable.

Losses to contents are dependent on the damage to the structure or building. From an engineering standpoint, losses to contents will be relatively small in comparison to structure losses until the envelope of the structure is breached. At that point, both structure and contents damage will quickly escalate with increasing wind speeds with the contents damage curve approaching that of the structure as windspeeds increase.

In the RMS hurricane model, separate contents damage curves have been developed from actual coverage specific loss data. Figure 5.8 shows a representative sampling of claims data used by RMS to develop the contents damage curves. For each ZIP Code in the sample, the contents and building damage ratios are plotted; the solid line is a best fit to the data. The dotted 45º line marks where contents and building damage ratios would be equal. This data clearly validates the

general engineering principals outlined in the paragraph above; at low wind speeds, the average levels of contents damage ratios are below the average levels of building/structure damage. At higher wind speeds, the ratios begin to converge.

The RMS hurricane model’s treatment of contents damage is derived from and reflects the relationships apparent in the data.

The independent development of contents damage curves as described above provides a separate mathematical representation of damage to contents in order to provide a reasonable representation of contents only policies and policies without contents coverage. The specifics of the development of the contents damage curves, including the review of actual hurricane contents specific claims data, can be made available for review by the Professional Team Contents Damage Ratio

Structure Damage Ratio Figure 5.8 Relative Structure and Contents Damage Ratios: Actual Claims Data

5.4.9 Additional Living Expenses (ALE)

The model shall provide a separate mathematical representation of Additional Living Expense (ALE) loss costs, and the modeler shall demonstrate its actuarial soundness.

The relationship between the modeled building and ALE loss costs shall be shown to be reasonable. If applicable, differences and the reasons for those differences from prior submissions in the relativities between loss costs for the building and the corresponding loss costs for ALE shall be explained and shown to be reasonable.

The modeler shall disclose the methods used in the model to incorporate ALE losses from damage to the infrastructure and the methods shall be shown to be reasonable.

In a manner similar to contents, losses to time element coverages are dependent on the damage to the structure. Time element loss ratios will be relatively small compared to structure loss ratios up to the point where the structure is severely damaged resulting in the building being uninhabitable.

In the RMS hurricane model, separate ALE damage curves have been developed from actual coverage specific loss data. Figure 5.9 shows a representative sampling of claims data used by RMS to develop the ALE damage curves. For each ZIP Code in the sample, the ALE and building damage ratios are plotted; the solid line is a best fit to the data. The dotted 45º line marks where contents and ALE damage ratios would be equal. ALE Damage Ratio Damage ALE

Structure Damage Ratio Figure 5.9 Relative Structure and ALE Damage Ratios: Actual Claims Data

The independent development of ALE loss functions as described above provides a separate mathematical representation of loss due to ALE coverage in a suitable degree of detail to allow for a reasonably accurate loss cost for polices not having ALE.

5.4.10 Replication of Known Hurricane Losses

The model shall be shown to reasonably replicate incurred losses on a sufficient body of past hurricane events, including the most current data available to the modeler. This standard applies separately to personal residential and mobile homes to the extent data are available. Personal residential experience may be used to replicate building-only and contents-only losses. The modeler shall demonstrate that the replications were produced on an objective body of loss data by county or an appropriate level of geographic detail.

The RMS model is able to reliably and without bias reproduce incurred losses on a large body of past hurricanes, both for personal residential and mobile homes. Validations of known storm losses have been performed in several ways, including:

For recent events, on an industry basis. The RMS model is able to reasonably reproduce aggregate incurred industry losses in recent events.

For recent events, on a company-specific basis. The RMS model is able to reasonably reproduce aggregate incurred losses for a diverse set of insurers.

For recent events, on a geographic and demographic basis. The RMS model is able to reasonably reproduce the geographic spread of company specific losses, and the spread of losses between various lines of business and between various types of coverages.

For less recent events, on an industry basis. The RMS model is able to reasonably reproduce industry losses for less recent hurricanes, both in aggregate and on a broad geographic basis, for which some level of industry loss data is available1.

Figure 5.10 and 5.11 show the results of representative samples of the comparative analyses that have been performed.

28,000 Actual Loss* 24,000 RMS Estimate**

20,000

16,000

12,000 Loss ($ millions) ($ Loss 8,000

4,000

0 Georges Fran Opal Erin Andrew Bob Hugo

Figure 5.10 Industry Loss Estimates (Residential and Commercial) for Recent Storms *PCS estimate, adjusted to 2000 dollars** Loss does not include demand surge or loss adjustment expense

1 From 1950 onwards, Property Claims Services (PCS) has tracked the aggregate industry losses from hurricanes. While these estimates, particularly the older ones, are potentially unreliable and must be adjusted to reflect current demographic and economic conditions, these older events do provide a means for checking potential bias in the model. Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 27 2002 Standards: 5.4 Actuarial Standards

$100.0

Losses indexed such that Observed Loss 100 = maximum company loss RMS Estimate*

$10.0

$1.0

$0.1 Andrew-Res Andrew-Res Hugo-Res Hugo-Res Opal-Res Fran-Res Erin-Res Bob-Res Georges-Res Hugo-MH Fran-MH

Figure 5.11 Company Specific Loss Comparisons *Loss does not include demand surge or loss adjustment expense

5.4.11 Comparison of Estimated Hurricane Loss Costs

The model shall provide the annual average zero deductible statewide loss costs produced using the list of hurricanes in 5.2.3 historical hurricanes in Florida based on the 1998 Florida Hurricane Catastrophe Fund’s (FHCF) aggregate personal residential exposure data, as of November 1, 1999. These will be compared to the statewide loss costs produced by the model on an average industry basis. The difference, due to uncertainty, between historical and modeled annual average statewide loss costs shall be demonstrated to be statistically reasonable.

The RMS hurricane model calculated historical annual average zero deductible loss for the Florida Hurricane Catastrophe Fund’s (FHCF) aggregate exposure database is $1.21 $1.12 billion per year. This value considers the historical storms in Standard 5.2.3 over the past 1020 years. The RMS hurricane model simulated annual average zero deductible loss for the same exposure database is $1.51 $1.27 billion per year. The difference between historical and modeled annual average statewide loss costs provided in this year’s submission is statistically reasonable. The historical and modeled loss costs provided in last year’s submission represented gross and not zero deductible loss costs, and were consequently lower.

5.4.12 Output Ranges

Any model previously found acceptable by the Commission shall provide an explanation suitable to the Commission concerning the differences in the updated output ranges. Differences between the prior year submission and the current submission shall be explained in the submission including, but not limited to:

1. Differences and the reasons for those differences from the prior submission of greater than ten percent in the weighted average loss costs for any county shall be specifically listed and explained in the modeler’s submission to the Commission. The submission shall include a specific listing of each county affected county. 2. Differences and the reasons for those differences from the prior submission of ten percent or less in the weighted average loss costs for any county shall be explained in the aggregate in the modeler’s submission to the Commission.

Output ranges have been provided in Module 3, Section V.4. Differences from last year’s output ranges are a result of the improvements RMS has made to its U.S. Hurricane model. Each component of the model has been rebuilt and/or recalibrated since last year’s submission (see Module 3, Section A.9). Changes in the output ranges are the result of one or more of the following:

The methodology to generate stochastic events remains the same but the smoothing technique to set the landfall probabilities has changed. At a state level, total landfall rates have not changed very much but there are slightly more Cat 3-5 storms and slightly fewer Cat 1-2 storms. For individual coastal gates there are some significant changes in probabilities.

The windfield model has been formulated in a time stepping mode and now utilizes directional roughness modification factors. The roughness factors themselves are computed from higher resolution roughness data than before. For the same storm parameters the new model gives higher windspeeds very close to the coast but similar winds inland. This increase in windspeeds is particularly important where the coastline is complex and/or exposure is concentrated near the coastline.

The damage functions have been revised to reflect a better understanding of the damage to building components; new claims data; new research in wind engineering; and new information obtained from post-event reconnaissance. Vulnerability changes vary by coverage and by windspeed. For the same windspeed the vulnerability of buildings is now a little higher while the vulnerability of contents is a little lower. For buildings, the increase in vulnerability is relatively higher at low windspeeds than at high windspeeds.

Differences in county-level total exposure loss costs are shown in absolute terms in Figure 12 below. The principal drivers of changes by county are given in Table 5.1.

Table 5.1: Differences in County Level Loss Costs and Principal Drivers of Change Percent COUNTY change in loss Explanation cost ALACHUA 81 Vulnerability changes in combination with rate and windfield changes BAKER 37 Primarily vulnerability changes BAY 47 Vulnerability changes in combination with rate and windfield changes BRADFORD 75 Vulnerability changes in combination with rate and windfield changes BREVARD 27 Vulnerability changes in combination with rate and windfield changes BROWARD 12 Primarily vulnerability changes CALHOUN 160 Primarily rate and windfield changes CHARLOTTE 11 Primarily vulnerability changes CITRUS 21 Primarily vulnerability changes CLAY 61 Vulnerability changes in combination with rate and windfield changes COLLIER -7 Changes in vulnerability offset by decreases in hazard. Little overall change COLUMBIA 103 Primarily rate and windfield changes DADE 0 Changes in vulnerability offset by decreases in hazard. Little overall change DE SOTO 16 Primarily vulnerability changes, slight decrease in hazard DIXIE 39 Vulnerability changes in combination with rate and windfield changes DUVAL 63 Vulnerability changes in combination with rate and windfield changes ESCAMBIA 53 Vulnerability changes in combination with rate and windfield changes FLAGLER 16 Vulnerability changes in combination with rate and windfield changes FRANKLIN 182 Primarily rate and windfield changes GADSDEN 110 Primarily rate and windfield changes GILCHRIST 82 Primarily rate and windfield changes GLADES 10 Primarily vulnerability changes, slight decrease in hazard GULF 31 Vulnerability changes in combination with rate and windfield changes HAMILTON 114 Primarily rate and windfield changes HARDEE 24 Primarily vulnerability changes, slight decrease in hazard HENDRY 25 Primarily vulnerability changes HERNANDO 28 Vulnerability changes in combination with rate and windfield changes HIGHLANDS 17 Primarily vulnerability changes, slight decrease in hazard

Percent COUNTY change in loss Explanation cost HILLSBOROUGH 51 Vulnerability changes in combination with rate and windfield changes HOLMES 62 Vulnerability changes in combination with rate and windfield changes INDIAN RIVER 41 Vulnerability changes in combination with rate and windfield changes JACKSON 115 Primarily rate and windfield changes JEFFERSON 83 Vulnerability changes in combination with rate and windfield changes LAFAYETTE 124 Primarily rate and windfield changes LAKE 76 Vulnerability changes in combination with rate and windfield changes LEE 16 Vulnerability changes in combination with rate and windfield changes LEON 49 Vulnerability changes in combination with rate and windfield changes LEVY 86 Primarily rate and windfield changes LIBERTY 159 Primarily rate and windfield changes MADISON 147 Primarily rate and windfield changes MANATEE 39 Vulnerability changes in combination with rate and windfield changes MARION 105 Vulnerability changes in combination with rate and windfield changes MARTIN 27 Vulnerability changes in combination with rate and windfield changes MONROE 8 Small increase in loss, primarily due to vulnerability changes NASSAU 44 Vulnerability changes in combination with rate and windfield changes OKALOOSA 39 Vulnerability changes in combination with rate and windfield changes OKEECHOBEE 14 Primarily vulnerability changes, slight decrease in hazard ORANGE 35 Primarily vulnerability changes, slight decrease in hazard OSCEOLA 27 Primarily vulnerability changes, slight decrease in hazard PALM BEACH 29 Vulnerability changes in combination with rate and windfield changes PASCO 30 Vulnerability changes in combination with rate and windfield changes PINELLAS 87 Primarily rate and windfield changes POLK 34 Primarily vulnerability changes, slight decrease in hazard PUTNAM 38 Vulnerability changes in combination with rate and windfield changes SANTA ROSA 69 Primarily rate and windfield changes SARASOTA 26 Vulnerability changes in combination with rate and windfield changes SEMINOLE 29 Primarily vulnerability changes, slight decrease in hazard

Percent COUNTY change in loss Explanation cost ST. JOHNS 43 Vulnerability changes in combination with rate and windfield changes ST. LUCIE 10 Primarily vulnerability changes, slight decrease in hazard SUMTER 50 Vulnerability changes in combination with rate and windfield changes SUWANNEE 124 Primarily rate and windfield changes TAYLOR 46 Vulnerability changes in combination with rate and windfield changes UNION 81 Vulnerability changes in combination with rate and windfield changes VOLUSIA 26 Primarily vulnerability changes WAKULLA 49 Vulnerability changes in combination with rate and windfield changes WALTON 36 Vulnerability changes in combination with rate and windfield changes WASHINGTON 123 Primarily rate and windfield changes

Differences in county loss costs per mille 1 to 5 0.5 to 1 0.25 to 0.5 0 to 0.25 0 to 0 -0.25 to 0

Figure 5.12 Absolute Changes in Loss Costs by County

5.5 Computer Standards

5.5.1 Primary Document Binder

A primary document binder, in either electronic or physical form, shall be created, and shall contain fully documented sections for each subsequent Computer Standard. Development of each section shall be indicative of accepted software engineering practices. All computer software (i.e., user interface, scientific, engineering, actuarial) relevant to the modeler’s submission must be consistently documented.

A Computer Standards primary document binder has been prepared by RMS and is available for on-site review by the Professional Team. The primary document binder contains an index which links each subsequent Computer Standard to one or more sections within the binder and, where appropriate, to other more detailed documents such as the RiskLink System Administration manual. The material contained within the primary documents binder is indicative of the accepted software engineering practices that are followed by the RiskLink development team. Through as the use of various techniques such as documentation templates and standards, the RiskLink software is documented in a consistent manner. RMS maintains technical, testing, user, maintenance and security documentation that describes how the various software routines are implemented in the model. RMS uses a configuration management process to track not only the files that go into each release of the software, but also for each interim “build” during the development cycle. This process, which is managed by the Quality Assurance department, includes CD archiving of all files and documentation including the date, time stamp, and originator of each file. In this way, there is an audit trail for all files produced and used in the model. The internal documentation is available for on site review by the Professional Team. This includes flowcharts of the overall hurricane model, data flow diagrams of the model components, diagrams of class inheritance and other class properties, and engineering specifications.

5.5.2 Requirements

The modeler shall document all requirements specifications of the software, such as interface, human factors, functionality, documentation, data, human and material resources, security, and quality assurance.

RMS maintains documentation of user interface / human factors requirements, functional specifications, documentation requirements, data specifications, human resource requirements, security measures, quality assurance requirements, and coding standards.

This internal documentation is available for on-site review by the Professional Team.

5.5.3 Model Architecture and Component Design

The modeler shall document detailed control and data flow diagrams, interface specifications, and a schema for all data files along with field type definitions. Each network diagram shall contain components (including referenced sub-

component diagrams), arcs, and labels. A model component custodian shall be identified and documented.

RMS maintains documentation of detailed control and data flow, interface specifications, and the schema definitions for all data files and database tables. Data flow diagrams are used to illustrate the relationship between software components and data using a network representation consisting of labeled component processes connected by data arcs, with components expanded into more detailed sub-component diagrams where appropriate. user interface / human factors requirements, functional specifications, documentation requirements, data specifications, human resource requirements, security measures, quality assurance requirements, and coding standards, including the appointment of model custodians. This internal documentation is available for on- site review by the Professional Team.

5.5.4 Implementation

The software shall be traceable from the flow diagrams and their components down to the code level. All documentation, including document binder identification, shall be indicated in the relevant component. The highest design level components shall incrementally be translated into a larger number of components until the code level is reached.

Available for review by the Professional Team during their on-site visit will be detailed control flowcharts of the overall hurricane model, detailed data flow diagrams of the model components, component input/output diagrams, diagrams of class inheritance and other class properties, and engineering specifications. Data flow diagrams are organized hierarchically, with highest design level components incrementally translated into a larger number of components. A data dictionary provides the linkage of the data flow components to the source code.

5.5.5 Verification

1. General

The modeler shall employ and document procedures employed, such as code inspections, reviews, calculation crosschecks, and walkthroughs, sufficient to demonstrate code correctness. The code shall contain sufficient logical assertions, exception-handling mechanisms, and flag-triggered output statements to test the correct values for key variables that might be subject to modification.

RMS Engineering and Quality Assurance departments rigorously check output generated from the model. Calculations are performed outside the model and compared to the software- generated results to ensure that they are correct. A series of test cases are run to ensure that the computer program generates consistent and reasonable results on a wide variety of client data. Data sets include end-condition test cases using very large and very small values, large-data- volume test datasets of many locations spread across multiple ZIP Codes, and data sets focused on testing specific areas of the model.

Code inspections, reviews, and walkthroughs are performed on a regular basis to verify code correctness. Both Software Management and Risk Modeling engineers participate in this process. Reviewers check code both during and after initial development. Code changes are often isolated and inspected using the features of our source code management system. Reviewers also use source-code debugging tools to verify run time behavior.

The software source code contains numerous logical assertions, exception-handling mechanisms, and flag-triggered output statements which are used to test the values of key variables for correctness.

With respect to the accuracy of input data, the software program tests that data values pass validation edits (for example, financial values are positive numbers, location addresses contain valid ZIP Codes, building classes are one of a predefined set of classes, etc.). Data that fails one of these edits is invalidated and excluded from the analysis.

2. Testing

Tests shall be documented for each software component, independent of all other components, to ensure that each component provides the correct response to inputs. The test specifications, procedures, and results shall also be documented to establish that the integration of all components produces model behavior that functions correctly.

Test plans are developed for each release of the RMS Hurricane model. These plans identify the scope and procedures for testing the model. For each discrepancy, a problem report is created and filed in a dedicated database to ensure that all errors are tracked and their resolution documented. Sample test plans are available for the Professional Team to review during their RMS on-site visit.

With respect to the accuracy of input data, the software program tests that data values pass validation edits (for example, financial values are positive numbers, location addresses contain valid zip codes, building classes are one of a predefined set of classes, etc.). Data that fails one of these edits is invalidated and excluded from the analysis

5.5.6 Model Maintenance and Revision

The modeler shall specify all policies and procedures used to maintain the code, data, and documentation. For each component in the system decomposition, the modeler shall list the installation date under configuration control, the current version number, and the date of the most recent change(s). The modeler shall use tracking software to identify all errors, as well as modifications to the code, data, and documentation.

As noted above (see Section 5.5.6),A problem report (“incident”) record is created using the Visual Intercept incident tracking system for each requested model change. This is done whether the change is viewed as a new feature or a bug fix. Each incident record is maintained throughout the life cycle of that incident, including resolution and re-testing a problem report is Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 36 2002 Standards: 5.6 Statistical Standards

initiated when an error is found and is maintained through the life cycle of that error including resolution and re testing. Documentation is developed for proposed engineering enhancements to the model and this documentation, including the software specification, is used in the development of the test plan for that release. This documentation, including the problem reporting system, is available for review by the Professional Team during their RMS on-site visit.

Microsoft Visual Source Safe is used to track modifications of all source code. In addition, documentation in our problem report database summarizes changes made to the source code, and provides a list of the source files affected by the change. For each software component, RMS will provide documentation of the installation date under configuration control, current version number and the date of the most recent change(s).

5.5.7 User Documentation

The modeler shall have complete user documentation including all recent updates.

User documentation for the RMS Hurricane model includes on-line help, help within the RiskLink product, RiskLink System Administration guide, RiskLink DLM User Guide, and RiskLink DLM Reference Guide. an Import Manual, a System Administrator’s Guide, DLM Reference Guide, and a User Manual. In addition, release notes are created for each revision of the model. This information is available for review by the Professional Team during their RMS on-site visit.

5.6 Statistical Standards

5.6.1 Use of Historical Data

The use of historical data in developing the model shall be demonstrated to be reasonable using rigorous methods published in the scientific literature.

RMS uses empirical methods in model development and implementation to match stochastic storm generation to historical data. For the modeling of the windspeed decay pressure filling component used in the RMS hurricane model, the rate of windspeed decay pressure filling is modeled as a variable. A graphical depiction of the modeled decay rates over time compared to the Kaplan-DeMaria decay rate and the +/- 20% range is presented in Module 3, Section V.7.

5.6.2 Comparison of Historical and Modeled Results

The modeler shall demonstrate the agreement between historical and modeled results using accepted scientific and statistical methods

RMS has employed a random-walk model to simulate the track and pressure distribution along the track of hurricanes in the entire North Atlantic basin by considering the characteristics of hurricanes. The parameters of the random-walk model are obtained from analyses of historical hurricanes for the time period 1900 to 2001. The results of the model are checked at all stages of Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 37 2002 Standards: 5.6 Statistical Standards

the development of the model to insure that physically realistic hurricanes, that preserve the statistical characteristics of the historical hurricanes, are simulated in the stochastic storm set. Statistical comparisons between historical and modeled data have been performed for the following:

a) Central Pressure b) Forward Speed Velocity c) Radius-to-maximum Winds (RMax) d) Landfall Rate e) Track crossing rates over a grid covering Florida f) Radius to hurricane force winds

In addition, vulnerability curves have been developed based largely on actual event insured loss data. Statistical comparisons of model vulnerability functions and the actual loss data have been made.

5.6.3 Uncertainty Characterization

The modeler shall provide an assessment of uncertainty using confidence intervals or other accepted scientific characterizations of uncertainty.

RMS has developed confidence intervals for the following, which will be presented to the Professional Team during their on-site visit: a) Central Pressure b) Forward Speed c) RMax d) Vulnerability relationships e) Uncertainty in historical AAL

5.6.4 Sensitivity Analysis for Model Output

The modeler shall demonstrate that the model has been assessed with respect to sensitivity of temporal and spatial outputs to the simultaneous variation of input variables using accepted scientific and statistical methods. Statistical techniques used to perform sensitivity analysis shall be explicitly stated and the results of the analysis shall be presented in graphical format.

In addition to the required Form F analysis, RMS has conducted independent sensitivity analyses to evaluate the impact on losses of varying key storm parameters such as central pressure, RMax, and forward speed, among other variables directly related to loss. During the Professional Team’s on-site visit, the results of the analysis will be available for review.

5.6.5 Uncertainty Analysis for Model Output

The modeler shall demonstrate that the temporal and spatial outputs of the model have been subjected to an uncertainty analysis using accepted scientific and Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 38 2002 Standards: 5.6 Statistical Standards

statistical methods. The analysis shall identify and quantify the extent that input variables impact the uncertainty in model output as the input variables are simultaneously varied. Statistical techniques used to perform uncertainty analysis shall be explicitly stated and results of the analysis shall be presented in graphical format.

In addition to the required Form F analysis, RMS has conducted independent uncertainty analyses to evaluate the impact on variance of loss through varying key storm parameters such as central pressure, RMax and forward speed, among other variables directly related to loss. During the Professional Team’s on-site visit, the results of the analysis will be available for review.

5.6.6 County Level Aggregation

At the county level of aggregation, the contribution to the error in loss costs estimates induced by the sampling process shall be demonstrated to be negligible using accepted scientific and statistical methods.

The RMS hurricane model stochastic storm set has been developed using a random-walk technique. This technique produces a set of tracks with a uniform annual rate of occurrence. The more tracks, the longer the simulated time. The submitted results were calculated using a stochastic track set representing approximately 100,000 years of simulated time. A stratified sampling of this full stochastic event set was then conducted. Loss convergence testing has been performed and verified that the error in loss costs estimates induced by the sampling process is negligible.

MODULES

MODULE 1

I. General Description of the Model

A. In General

1. Specify the model and program version number reflecting the release date.

The current model is RiskLink version 4.3a4.2 SP1a

2. Provide a complete and concise description of the model, with a one-page introductory summary. Include a description of the methodology, particularly the wind components, the damage components, and the insured loss components used in the model. Indicate where probability distributions have been fit to historical data and demonstrate their agreement. Describe sensitivity and uncertainty analyses used in the development of the model. Describe the computer language/code in which the computer program is written and what type of computer hardware is required. Specify the details of translation from model structure to program structure.

2.1 Introductory Summary

From 1900 to 2001, approximately 160 hurricanes of Saffir-Simpson Category 1 strength (NHC classification) or greater have made landfall on the continental United States, averaging approximately 1.6 hurricanes per year. Of these, 56 have made landfall within the state of Florida. Many of these events have been quite severe, and given the current population demographics of the state, would result in catastrophic insurance losses if they were to re-occur today.

Despite this significant record of hurricane activity, the direct loss experience from these events is not sufficient to form the basis for projecting future hurricane risk within Florida. First, the frequency and severity of historical activity has not been constant over time. The evidence clearly indicates that despite the recent occurrences of hurricane Andrew and Opal, Florida experienced a far higher incidence of significant hurricanes prior to the early 1960s. The tremendous population growth and related demographic changes in Florida since 1960, coupled with significant changes in construction practices, materials and costs, makes any direct loss experience from these earlier events of little relevance to today’s insurance industry. Second, the geographic pattern and physical characteristics of the historical record does not reflect the full range of outcomes that we are likely to see in the future. For example, while Florida has experienced many intense hurricanes, it is possible that hurricanes of even greater intensity will make landfall within the state. In a similar vein, the specific landfall locations of the historical hurricanes, if interpreted literally, may provide a misleading picture of future risk given the relatively short duration of our historical observations.

To address these challenges and provide the insurance industry with a reliable long term view of hurricane risk in Florida and throughout the United States, Risk Management Solutions, Inc. (RMS) has developed its hurricane model. This hurricane model provides users with a “virtual future” of hurricane risk represented by thousands of stochastically defined hurricanes, each with the physical characteristics of real hurricanes. These characteristics are central pressures, wind profiles (radii to maximum and hurricane force winds), tracks, forward velocities and windfields over water or land as well as a probability of occurrence. based upon the likelihood of the combination of the characteristics and landfall locations. The more severe or unusual the combination of characteristics and the more unlikely the landfall location, the lower the probability of that particular stochastic storm in RMS’ model. Overall, the RMS The underlying Monte Carlo hurricane model for the Atlantic Basin contains approximately 400,000 modeled hurricanes, equivalent to approximately 100,000 years of simulated activity. For loss cost determinations the Monte Carlo event set is importance sampled to improve computational efficiency. The event set used for loss costs determination contains 20,394 storms causing loss in Florida. This smaller event set provides the same representation of hazard as the much larger, Monte Carlo set it is derived from. , of which approximately landfall along Florida, equivalent to tens of thousands of simulated years of possible hurricanes.

For each of these stochastic (simulated) storms, the model is capable of overlaying the physical characteristics of the storm against a description of the insured exposures throughout a geographic region. Such exposures, described by their geographic locations, values, policy forms (coverages, limits and deductibles), and construction characteristics are related to computed levels of wind speed to model associated damages and financial losses. To compute the average annual loss, or loss costs for specific exposures, the losses from each simulated storm are then weighted by the probability of the storm’s occurrence.

Each component of the RMS hurricane model - the storm forecasting module, the windfield, or wind hazard module, the vulnerability, or damage module, and the financial loss module - is developed through a well-balanced combination of engineering and physical sciences “first principals”, and thorough analyses of historical data. Decisions made by RMS scientists and engineers in creating this model are well supported by scientific literature and research performed and documented by such staff. All model components are independently calibrated and validated to ensure that they describe a set of outcomes that is physically plausible, covers the range of physical variability, and satisfies a fitness measure to historical observations, without bias. Once integrated, the overall model has been extensively validated against actual observations of loss in a broad range of historical hurricanes, across a diverse set of insurance portfolios, and at all levels of geographic and demographic specificity.

2.2 Description of The Model

RMS’s proprietary RiskLink hurricane model for the United States is built around four major model components, or Modules:

a) Forecasting Module b) Windfield, or Wind Hazard, Module c) Vulnerability, or Damage Assessment, Module d) Financial Loss Module Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 42 MODULE 1: I. General Description of the Model

Each of these will be discussed below. Also described are the computer language and requirements as well as the model development methodology.

2.2.1 Forecasting Module

The historical storms database by itself cannot adequately serve as a basis for a model because it contains too few events for a reliable statistical analysis. For average annual loss calculations, the hurricane model contains 20,394 stochastic storms affecting Florida. approximately 50,000 stochastic storms making landfall

The RMS hurricane model uses a random-walk technique to generate a set of stochastic storm events covering the entire Atlantic Basin. Each event consists of a track (location, forward speed and direction, central pressure and radius of maximum wind) defined throughout the life of the storm from its genesis to its dissipation. Importance sampling of the simulated tracks is performed to create the computationally efficient event set used for loss cost determinations. Each event in the storm set has the same annual rate of occurrence. The total rate of occurrence of stochastic events is the observed mean annual rate of occurrence of historical storms.

Historical database. The development of stochastic (simulated) storms is derived from an analysis and parameterization of historical storm data. The historical storm database was developed with the participation of Charles J. Neumann, a meteorologist and one of the original researchers from the NHC, who compiled the HURDAT North Atlantic Basin Storm database (Jarvinen, Neumann, & Davis, 1984). The HURDAT database contains four pieces of information for each tropical cyclone recorded: time and date, latitude and longitude position, maximum sustained wind speed, and central pressure (when available). Working with Mr. Neumann, RMS engineers researched the background data on historical storms as well as specific information on several hurricanes. The key background references include Neumann (1990), NOAA (1993), and Simpson (1981). The RMS historical database was developed by incorporating the most reliable available information from this research. The investigation resulted in corrections of the time history of pressure and wind speed data along the track, as well as a more accurate definition of storm characteristics at landfall. Only storms which reached Category 1 or above were used in the development of the model. RMS consulted with other experts including Dr. Alan Davenport and Dr. Dale Perry to collect more data and to seek their opinion on specific storms. The final RMS-developed database was again reviewed by Charles Neumann. Results of the National Hurricane Center reanalysis project were also reviewed The hurricane model uses the same set of hurricanes specified by the Commission in the November 2, 2002 Report of Activities.

Modeling of storm tracks. Storm tracks are simulated using a random-walk technique. This method creates realistic synthetic events, which preserve the statistical behavior of the historical events (mean and variance of translational velocity). Tracks are simulated in two steps. Firstly, the tracks are created and secondly pressure histories are added to the tracks. The storms are then importance sampled to obtain the event set used for loss-cost determinations. Each simulated event has a uniform annual rate of occurrence.

The track model is calibrated across the Atlantic basin by comparing the rates of storms crossing a grid of cells covering the basin. More detailed calibration is performed at coastlines by Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 43 MODULE 1: I. General Description of the Model

calculating the rate of crossing and probability density functions (pdf) of forward speed and direction on linear gates along the coastlines.

Landfall (“strike”) probabilities. Calibration of landfall probabilities is performed on a series of segments, approximately 50 nautical miles (nmi) in length that bound the entire US coastline Florida’s geography. The target historical probabilities are computed from the historical database by application of a smoothing algorithm that corrects for inappropriate variability in the historical record due to the limited length of record. The stochastic model is then calibrated to match the historical rates of landfall.

Storm track characteristics (parameter) probabilities. Probability density functions (pdf) are measured for each of Florida’s 50 nmi segments for forward velocity, and angle of landfall (azimuth). Due to the limited length of the historical record, the target, historical pdf s are computed from the historical database by application of a smoothing algorithm. Extrapolations, primarily by including larger regional behavior, are also made to reflect the probability of occurrences beyond the range of the observed record at a gate. Calibration of forward speeds is performed by computing probability density functions (pdf) of forward speed from the historical record. Due to the limited length of the historical record, the calibration is performed at a regional level by grouping neighboring gates together.

Pressure filling. Pressure histories are added to the synthetic tracks using a second random-walk process. The rates of change of pressure along the synthetic tracks are defined through the mean and variance of pressure changes quantified from historical events. Storms tend to intensify faster over warm water than over cold water. Storms fill as they cross areas of land and may reintensify if they move back out over the water. The filling rates for storms landfalling on Florida is modeled using the same functional form as the model of Kaplan & DeMaria are based on observed historical behavior of storms in the Florida region. Minimum pressures are constrained by theoretical arguments relating central pressure to Sea Surface Temperature (SST). The pressure history of each storm thus depends on the track of the storm as it crosses areas of different SST and encounters topography.

The pressure history model is calibrated by specifying the pressure pdf on linear segments cells across the basin and on linear segments around the coastline. The pressure history of each event is individually scaled so that the pressure pdf for each segment is obtained. In this way the random-walk model defines realistic pressure histories and the calibration ensures the correct intensities of simulated storms. The methodology to generate stochastic storms at a location is described by the following steps:

Step 1: Quantify the translational velocity behavior of the historical storm set. The model uses a random-walk technique frequently applied to turbulent dispersion problems by considering each hurricane to be advected by a 2D “turbulent” translational velocity field superimposed on a “mean” translational velocity field. Both mean and turbulent velocity fields are inhomogeneous in two directions so the translation equations have been formulated to incorporate the interaction of these inhomogeneities. Model inputs are computed from the tracks of historical events in the HURDAT catalogue on a regular array of grid cells covering the whole Atlantic basin as shown in Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 44 MODULE 1: I. General Description of the Model

Figure 1.I.1. Historical tracks are classified into five types, depending on their point of formation and path. Each type is simulated separately. Type 1 storms (such as Floyd 1999) form in the Atlantic ocean and recurve up the East Coast of the US, Type 2 storms (e.g. Georges 1998) form in the Atlantic ocean and travel through to the Gulf of Mexico, Type 3 storms form off the East Coast of the US, Type 4 storms (e.g. Mitch 1998) form in the Caribbean Sea and Type 5 storms (e.g. Opal 1995) form in the Gulf of Mexico.

Step 2: Simulate the storm tracks and calibrate against historical rates of occurrence The storm tracks are modeled first and the intensity histories are added as a separate step using a random-walk technique for the pressure. The model has been extensively calibrated. The rates of storms crossing each model cell have been computed and compared against the historical record. Figure 1.I.2 shows 150 simulated ‘Type 2’ tracks.

Figure 1.I.1 Mean Translational Velocities for ‘Type 2’ Hurricanes on a 2º x 2º Grid

Figure 1.I.2 150 Simulated ‘Type 2’ Tracks

Step 3: Calculate target historical landfall rates and track parameter PDFs along the Florida coastline The US coastline is first divided into segments of about 50 nautical miles. This translates into 22 coastal segments (segments 17 to 38) for the State of Florida as shown Fig. 1.I.3. There are also 4 coastal segments capturing the coastline of the neighboring states of Georgia, Alabama and Mississippi. Historical crossing are computed for each coastal segment by smoothing across extensions to the segments Probability density functions for central pressure are developed for each segment from landfalling data supplemented by nearby, offshore track information. Pressure CDFs are then smoothed by normalizing landfall rates by Cat to match the historical record at a regional level.

Probability density functions (pdf) of forward speed are developed for groups of coastal segments for the parameters that define the windfield (i.e. central pressure difference, forward velocity, and angle of storm track). These distributions are based on a smoothing with nearby segments. Lower and upper bounds are developed for all parameters based on regional hurricane characteristics to keep the parameters within a realistic range.

NW NE

SE SW

Figure 1.I.3 Segments Used for Parameter and Rate-Smoothing

Step 4: Calibrate the storm tracks against landfall rates and forward speed track parameter pdfs at the coastline.

Step 5: Add the pressure histories to each stochastic event taking into account changes in SST and encounters with land along the way

Step 6: Calibrate the pressure histories against the pressure pdfs for each coastal gate.

Step 7: Importance sampling of the Monte Carlo basin-wide storm set to produce the event set used for loss-cost determination Each simulated storm has a uniform annual rate of occurrence, which depends on the number of simulated storms and the mean annual rate of occurrence of storms in the basin.

2.2.2 Windfield or Wind Hazard Module

The Windfield or Wind Hazard Module calculations determine the maximum localized wind speed associated with a storm event (historical or stochastic) over its life cycle. The wind speeds are calculated at a site identified by its latitude and longitude, taken either from a street address specific geocode or derived from the weighted centroid of a ZIP Code. The key storm parameters used in wind speed calculations include: central pressure, radius to maximum wind and wind profile, forward speed velocity, direction, landfall location, and track.

The theoretical and analytical formulations of the Windfield Module are taken from a methodology originally developed at the Boundary Layer Wind Tunnel, University of Western Ontario, Canada (Georgiou 1983, 1985). The wind speed is calculated from the formula relating the site location relative to the storm track, the landfall location, and the physical parameters of the storm. A sketch showing the relationship between the parameters used to compute the windfield values at a given time during the life cycle of the storm is presented in Figure 1.I.4.

(Azimuth, ∆P, VT,Rmax)

D min Site

Coastline Distance to Coast Hurricane Track

Figure 1.I.4 Sketch Showing Hurricane Model Parameters

The windfield calculations include the following steps:

Step 1: Estimation of over water gradient balance wind speed Vg

The mean gradient wind speed, Vg, is calculated from the formula:

g = T α fRSinVV ))((5.0 +− 1 B B  Rmax  2  −   (1)  2  ∆P  Rmax   R   T α ))((25.0 +−  BfRSinV   e   ρ  R     where R = radial distance from the storm to the site; α = angle from storm track to site (clockwise is positive); ∆P = central pressure difference; VT = storm translational speed; ρ = air density; f = Coriolis parameter (function of latitude); B = pressure profile coefficient and Rmax = radius to maximum winds.

Step 2: Estimation of over water windfield at 10m height Vs The 10-minute over-water wind speed, Vs, is a function of the gradient wind speed and relative position of site to the storm track and is obtained from:

 R  R   b −− c max  V  R 2R  s += ea  max   (2) Vg where a, b, and c are constants. These parameters are region-dependent.

Step 3: Estimation of over water peak gust windfield at 10m height The over water surface peak gust wind speed, Vgust, is calculated as:

gust = *VaV s (3) where a is a constant.

Step 3: Estimation of over land peak gust The model calculates overland gust wind speeds at a location by modeling both the effects of the local surface roughness and any change in the surface roughness conditions upwind of the location being considered. As the upstream roughness generally varies with direction about a particular location, the model considers the effects of upstream roughness by direction.

The calculation of peak gust includes consideration of the distance from the site to the coastline as well as local site conditions. Wind speeds experienced at ground level and on buildings are influenced by local roughness conditions. Friction from terrain roughness causes turbulence which reduces windspeed (i.e. the wind speed decreases with increased roughness).

The starting point for the determination of land friction effects is the creation of a database that describes the surface roughness in terms of the roughness length. The definition of the roughness length arises from the use of a logarithmic velocity, or log- law, profile to describe the variation of the wind speed with height in the region immediately adjacent to the surface. Use of the log-law requires a measure of the

underlying surface roughness, which is achieved through the use of the roughness length to parameterize the effect of surface roughness on the wind speed. The use of a roughness length to describe the underlying surface roughness also allows a physically based model to be used to calculate both local and upstream surface roughness effects on the wind speed

Coefficients describing the impact of land friction are then calculated by using the roughness database in conjunction with GIS software to sample both the local and upstream roughness conditions by direction at each point of interest. Both local and upstream roughness conditions are sampled because the wind speed at a particular location is determined not only by the local surface roughness, but by any change in the surface roughness conditions upwind of the location being considered. As the upstream roughness will generally vary with direction about a particular location, sampling of the upstream roughness must also be undertaken by direction. Information on the sampled roughness length values and their distance from the location are then used in conjunction with a physically based model to determine an appropriate set of coefficients describing the impact of land friction effects at the location by direction. The roughness classification is based on land use, land cover data published by the U.S. Geological Survey (USGS 1978, 1990), and the density of building construction. The roughness classification is developed from a matrix that combines the natural roughness with man made roughness into exposure classifications for each ZIP Code. The final roughness classification assigned to the ZIP Code is based on smoothing the roughness classification for the ZIP Code itself with those of adjacent ZIP Codes to account for the transition zone required for wind to adjust to new terrain conditions. This roughness category is then used to adjust the wind speeds (peak gust) to account for the local conditions based on calibration of roughness conditions with historical storm loss data.

2.2.3 Vulnerability or Damage Assessment Module

Given an event, the model estimates the wind and surge (optional) hazards present at a user- specified site. Local wind and surge hazards are measured in terms of peak gust wind speed and flood depth, respectively. These parameters are then used to drive the estimate of damage to a specific location. Estimated damage is measured in terms of a Mean Damage Ratio (MDR) and a deviation around the mean represented by the Coefficient of Variation (CV). The MDR is defined Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 49 MODULE 1: I. General Description of the Model as the ratio of the repair cost divided by replacement cost of the structure. The curve that relates the MDR to the peak gust is called a vulnerability function. RMS has developed vulnerability functions for 2624 building classifications. Each classification has a vulnerability function for damage to buildings due to wind and a vulnerability function for damage to contents due to wind, as well as similar vulnerability functions for surge damage. Time element (ALE) vulnerability functions are based upon the building damage function and the occupancy of the structure.

The development of building classifications is based on a combination of the construction material, building usage, and number of stories as explained in Section I.A.7. The vulnerability functions consist of a matrix of wind speed levels (measured as peak gust in mph) and corresponding MDRs. To calculate a MDR for a given location, RiskLink first determines an expected wind speed, then looks up the corresponding MDRs for building and contents based on the building classification. RMS has also developed CVs associated with each MDR. The CV is used to develop a probability distribution for the damage at each wind speed and for each classification. A beta distribution is used for this purpose.

The vulnerability functions are based, both on engineering principles using the new Component Vulnerability Model (CVM) as well as analyses of loss data. The CVM allows an objective modeling of the vulnerability functions, especially at higher windspeed ranges where little historical loss data is available, as well as the modeling of vulnerability of various classes of buildings. The CVM is used, not only to develop vulnerability functions, but also to obtain the vulnerability relativities by building class as well as for gaining better insights into hurricane mitigation.

The engineering model based on the CVM is calibrated using historical claims data at ZIP Code resolution for each of building, contents, and Additional Living Expenses (ALE) coverages. The calibration process involves a comparison of modeled Mean Damage Ratio (MDR) with that obtained from observed losses. Since the vulnerability model is a function of the windspeed, the calibration involves varying both windspeed and vulnerability within the bounds established by a) the science and historical observations governing the hazard at a given location and b) the engineering and historical observations governing the damageability of property at that location. Thus, one primary goal of calibration is to ensure that the vulnerability function is confined within the high and low vulnerability bounds as established by the CVM.

Development of vulnerability curves is based primarily on detailed analyses of insurance claims data. However, RMS also makes use of published documents, expert opinion, and conventional structural engineering analysis. RMS has reviewed research and data contained in more than 50 technical reports, special publications, and books related to wind engineering and damage to structures due to wind. These publications include studies performed for the National Science Foundation, for the Veterans Administration, and for the Federal Emergency Management Agency; as well as studies performed by the Army Corps of Engineers, and by Don Friedman at the Travelers. Published loss studies, damage surveys, and evaluation studies were also used in the development of the vulnerability curves. Section I.A.8 provides the source of documents and the research done to develop the vulnerability functions.

RMS engineers have conducted several post-event reconnaissance in the aftermath of hurricanes e.g. Hurricanes Opal (1995) and Erin (1995) in Florida, Hurricane Marilyn (1995) in the Virgin Islands and Puerto Rico, Hurricanes Fran (1996) and Bonnie (1998) in North Carolina, Hurricane Georges (1998) in the U.S. Gulf Coast and Puerto Rico and Hurricane Floyd (1999) in North Carolina, Typhoon Paka (1997) in Guam, Typhoon Angela (1995) (Rosing) in the Philippines and Typhoon Ryan (1995) in Japan. The knowledge and data gathered during these site visits have helped calibrate and validate the vulnerability functions.

The vulnerability of buildings modeled by each of the 26 building classes vulnerability functions represents the “average” vulnerability of a portfolio of buildings in that class. The vulnerability will vary depending upon specific characteristics of buildings in that portfolio. This variation can be addressed in the model through the use of secondary modifiers which can consider secondary building characteristics or mitigation measures to improve a building’s wind resistance. The secondary modifiers could be building-characteristic specific (e.g. improved roof sheathing or anchors) or external (e.g. storm shutters). These secondary modifiers modify the base, “average” vulnerability functions according to specific building characteristics or mitigation measures. The secondary modifiers are discussed in section 5.3.5.

Wind dependent functions, known as the secondary modifiers that modify the vulnerability curves according to specific building characteristics, have been developed as outlined in Standard 5.3.5. RMS also obtained input from researchers at the Boundary Layer Wind Tunnel Laboratory at the University of Ontario on performance of structures based on wind tunnel tests. This laboratory operates the largest wind tunnel laboratory in the world.

2.2.4 Financial Loss Module

To calculate losses, the damage ratio derived in the Vulnerability Module is translated into dollar loss by multiplying the damage ratio by the value at risk. This is done for each coverage at each location. Using the mean and CV, a beta distribution is utilized to represent the loss distribution. From the loss distribution one can find the expected loss and the loss corresponding to a selected confidence level by taking the integral under the loss curve up to the confidence point.

RiskLink uses the loss distribution curve to estimate the portion of loss carried by each participant within a financial structure (insured, insurer, reinsurer). The insurer-to-insured distribution is determined by the structure of the insurance policy. The insurer-to-reinsurer(s) distribution is determined by various contracts that an insurer may have with reinsurers. The calculation of the financial liabilities is done at each location and then integrated by policy and/or by portfolio while making the appropriate calculation of means and coefficients of variation. Hence, a loss curve can be developed for each single asset at a location, an account (i.e. a collection of several sites), or a portfolio (i.e. a collection of several accounts). For loss calculation purposes, a location associated with the centroid of the ZIP Code is typically considered.

2.2.5 Computer Languages and Requirements

The primary language for the development of RiskLink is C++. RiskLink runs on Intel-based processors under Microsoft Windows 98 and later releases. Minimum hardware requirements Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 51 MODULE 1: I. General Description of the Model include a CD-ROM drive, Pentium 400 MHz processor, 3.6 GB hard disk space, 512 MB RAM, and an 800 x 600 display. A more detailed description of requirements is listed in the RiskLink 4.3 System Administration manual. The primary languages for the development of RiskLink are C and C++. RiskLink currently runs on an Intel based processor under Microsoft Windows operating systems. Hardware requirements include a CD ROM Drive, 100 MB hard disk capacity (minimum), 4Mb RAM (minimum), and VGA/SVGA Video.

2.2.6 Model Development Methodology

As described above, the architecture for the hurricane model involves breaking the basic components into smaller modules and sub-modules, such as the wind hazard module and the vulnerability module. This structure is carried over into the software architecture. Modifications or additions to the model are typically designed and prototyped by wind engineers. Prototypes are coded, for example, in spreadsheets or in programs written in C, C++, or FORTRAN. Once the concept has been proven in the prototype, a written specification is prepared to describe the purpose of the change and to provide a detailed description of the algorithm to be introduced to the production software. This description typically takes the form of narrative, “pseudo-code” (similar to computer code but stripped of computer language details for the sake of readability), control flowcharts, or data flow diagrams. This description is sometimes augmented by actual computer code from the prototype. The specification is peer-reviewed by other wind engineers and by senior software developers. Once the specification is approved, the changes are then made to the production software. These software updates are then reviewed by senior software developers and by wind engineers. Test cases are written and run by software developers, wind engineers, and by quality assurance engineers, to provide multiple levels of functional testing.

3. Describe the theoretical basis of the model. Provide precise citations to or, preferably, copies of, the representative or any primary technical papers that help describe the underlying theory that was relied on for any particular component of the model.

The overall architecture of the model represents the current state-of-the-art in probabilistic catastrophe modeling methodologies. A summary of the theoretical basis for the various model components is as follows:

3.1 Storm Forecasting

The random-walk technique is widely used in the areas of environmental fluid mechanics, particularly to simulate the dispersion of pollutants (e.g. Luhar, A. K. and Britter, R. E. (1989). “A Random Walk Model for Dispersion in Inhomogeneous Turbulence in a Convective Boundary Layer”, Atmospheric Environment, 23(9), 1911-1924). RMS is the first modeling company to introduce the methodology to hurricane modeling (Drayton, M. J. (2000). “A Stochastic ‘Basin-wide’ Model of Atlantic Hurricanes”, Proceedings of The 24th Conference on Hurricanes and Tropical Meteorology, 29 May - 2 June 2000, Fort Lauderdale, Florida).

Detailed calibration of the model by comparing storm parameters distributions at the Florida coastline follows the more traditional, general approach set forth in the National Weather Service Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 52 MODULE 1: I. General Description of the Model publication NWS-38, “Hurricane Climatology for the Atlantic and Gulf Coasts of the United States”.

3.2 Windfield and Wind Hazard

Many windfield models were reviewed in the development of the RMS hurricane model. These included models developed by Graham & Nunn (1959), Tryggvason (1976), Schwerdt et al. (1979), Batts et al. (1980), Holland (1980), Georgiou (1983, 1985), Neumann (1987), and Sanchez-Sesma et al. (1988). These models were developed by scientists in the field of meteorology and by engineers. Georgiou’s model was initially selected to be used in the windfield module. There were several advantages in using this model, including the fact that is has been used by the University of Western Ontario in engineering studies related to performance and design of structures.: a) it is one of the most recent models published in peer reviewed literature, b) has, and c) the wind hazard results for the US coastal areas using this model served as a basis for the development of wind hazard maps in the ASCE/ANSI wind load standards. Dr. Alan Davenport, who supervised Georgiou’s work at the Boundary Layer Wind Tunnel has been a consultant to RMS since 1991. He and his colleagues at the BLWT have been full participants in the development of RMS Hurricane model.

The treatment of both surface roughness effects on mean and gust wind speed changes are modeled based on peer-reviewed wind engineering literature (Cook, 1985; Wieranga,1993 and 2001).

3.3 Vulnerability Functions

The vulnerability relationships are developed using structural and wind engineering principles coupled with analyses of on the basis of historical storm loss data, building codes, published studies and RMS internal engineering developments in cooperation with wind engineering experts such as Dr. Dale Perry, Dr. Norris Stubbs of Texas A&M University, Dr. Peter Sparks of Clemson University, and Dr. Alan Davenport of the University of Western Ontario. A list of selected references for each subject is provided at the end of this document.

4. Provide classes, objects, and procedures that define how the model is represented and how the domain associated with hurricane catastrophe (including all hurricane-related entities) is mapped to elements in the computer program. Explain all interfaces and coupling assumptions.

The high-level flow chart is shown in Figure 1.I.5. Detailed, proprietary flow-charts of each of the model subsystems including class diagrams can be shown to the professional team.

Model Flowchart

Historical Storm Database: Forecast Module User input Historical storms Type of analysis: that have hit Determines the storm properties Determinnistic Florida to be used in the analysis. For a deterministic analysis, Stochastic properties are retrieved for the Stochastic Storm chosen storm. For a Loss Over Database. Time analysis, each stochastic storm is analyzed. Stochastic storms Affecting Florida

Central Pressure Radius to Maximum Wind Forward Velocity Track

Windfield Module Information retrieved Peak gust and flood height are from databases based determined at the site. on geocoded location: Intermediate variables calculated: User Input Elevation Distance to coast Site(s) Location Exposure classification Vg, over water mean gradient wind speed

Peak gust Flood height

(continued overleaf)

Model Flowchart (continued)

Peak gust Flood height

Vulnerability Module User Input Vulnerability Databases: Building Information Vulnerability curves Damage ratios are calculated for Primary: relating peak gust and Construction class both buildings and contents, for flood depth to building Number of Stories both wind damage and flood and contents damage. Building Usage damage. Secondary: Cladding type Roof framing type etc.

Financial Module User Input Financial loss is calculated by Values:: multiplying the damage ratios Building, Contents, by values. Loss is represented Business Interruption, by the mean damage ratio and CV. ALE For portfolio analyses, losses are Insurance Structure: statistically aggregated. Limits Deductibles etc.

Financial loss to policy participants

Figure 1.I.5 Model Flow Chart

5. Provide a list and a description of the model variables and the outputs from the model. In describing the variables, state which are qualitative and which are quantitative. Describe the possible range associated with each variable. Identify differences, if any, in how the model produces loss costs for specific historical events versus loss costs for events in the stochastic hurricane set. Indicate which model variables are critical as determined from a sensitivity analysis or suitable equivalent. The objective is to provide an assessment of the attendant uncertainty in the loss costs produced by meteorological variables (including both occurrence and wind field aspects), vulnerability variables and actuarial variables.

5.1 Model variables

Table 1.I.1 lists model variables along with an indication of the relative importance (Y = relatively critical, N = less critical).

Table 1.I.1 Model Variables

Variable Importance Exposure Data (Input Variables) Locations of risks Y Monetary values per risk for each coverage Y Insurance structure for each coverage/policy Y Reinsurance structures Y Hazard Related (Model Variables) Central pressure Y Forward velocity Y Radius to maximum winds Y Landfall location Y Geography (Model Databases) Coastline position Y Elevation Y Roughness Y Building Primary Characteristics (Input Variables) Construction class Y Year built Y Number of stories Y Occupancy Y Building Secondary Characteristics (Input N Variables)

Table 1.I.2 provides the list of variables shown in Table 1.I.1 along with the range of possible values the variable can take.

Table 1.I.2 Model Variables With Range of Possible Values

Variable Range Exposure Data (Input Variables) Location of Risk Street Address; ZIP Aggregates or County Aggregates Values (Building; Contents, and BI) >=0 Insurance Structure All standard policies can be modeled Reinsurance All standard contracts can be modeled Hazard Related Central Pressure Continuous function with regional bounds Forward Velocity Continuous function with regional bounds Radius to Maximum Winds Continuous function Landfall location Longitude/latitude Geography Coastline Location Database Elevation Database Roughness Database per VRG cell Building Primary Characteristics Construction class See question I.A.7. Year built See question I.A.7. Number of stories See question I.A.7. Occupancy See question I.A.7. Building Secondary Characteristics Each Building Secondary Characteristic consists of a set of questions regarding specific conditions of the building that could be known from a visual inspection and/or a review of construction documents. The user enters a single answer for each secondary characteristic. A set of modifiers is assigned to each secondary characteristic. The modifiers vary with the wind speed. If a secondary characteristic is entered by the user, the system will adjust the MDR by the amount of the modifier, either upward or downward. For example, the secondary characteristic “Construction Quality/Maintenance” includes the following conditions: 1) Unknown; 2) Certified design and construction; 3) Certificate of occupancy; 4) No design review/inspection; and 5) Obvious signs of duress or distress. The modifier for Unknown is zero. The CV is also changed when the MDR is adjusted. The more the user knows about the building, the less uncertainty there is. The uncertainty is never nil. The default for any secondary characteristic is “unknown”. The default results in no change to the vulnerability function or the CV.

5.2 Outputs

The hurricane model produces a wide range of outputs, both in terms of reports, and in terms of a variety of detailed electronic output files. At the finest level of granularity, the model can produce output which provides the amount of loss per coverage per location analyzed per simulated event. Related files provide the expected losses from all simulated events, or complete loss distributions on both an occurrence or an aggregate basis. Output can also be provided by location, by coverage, and by financial perspective (e.g. ground-up vs. gross and/or net losses).

5.3 Loss Costs From Historical Storms versus Stochastic Storms

Stochastic storms are modeled in the same way as historic storms for which the basic storm parameters (central pressure, forward speed, landfall, track angle, radius to maximum wind) are available. In cases where more detailed windfield information is available for a historic storm, that information is used in representing wind speeds when modeling that storm. Vulnerability and financial modeling functions are identical for both stochastic and historic storms.

6. Are there methods used in the model to incorporate modification factors to the actuarial functions or characteristics? If so, describe. In particular, to what extent are mitigation factors incorporated in the model.

The model includes the capability to specify building characteristics as described in section I.A.5.1.

7. Describe the number of categories of the different vulnerability functions (damage ratios) used within the model. Specifically, include descriptions of the structure types, lines of business, and coverages in which a unique vulnerability function is used. What is the basis for differentiation (e.g., engineering analysis, empirical data, etc.)?

There are a total of 2426 building vulnerability classes. Each class has both building and contents damage functions (a total of 5248 functions). The vulnerability classes depend on a combination of: a) Construction class b) Building height, and c) Occupancy

The possible classifications for each of the three primary characteristics are listed in Table 1.I.3.

Table 1.I.3 Building Classification

Construction Class # of Stories Occupancy Woodframe Unknown Unknown Masonry 1 Single family Reinforced Concrete 2-3 Multiple family Steel Frame 4-7 Commercial Light Metal Frame 8-14 Industrial Mobile Home without Tie- 15+ Downs Mobile Home with Tie-Downs

The various vulnerability classes were defined to allow for the grouping together of structures with similar performance under wind loads.

8. What are the primary or representative documents used or the research results utilized in the development of the model’s vulnerability functions (damage ratios)?

Vulnerability functions in the current hurricane model are primarily derived through internal analysis of actual insurer hurricane loss experience. Over The vulnerability functions are developed on the basis of structural and wind engineering principles coupled with analyses of historical storm loss data, building codes and published studies The vulnerability functions have also been calibrated using over $7 billion of loss data, with corresponding exposure information. was analyzed. In addition, RMS has reviewed research material, engineering analysis, damage surveys, and damage data contained in more than 50 technical reports including studies performed for the National Science Foundation (J.H. Wiggins Company, 1980; NBS, 1981), for the Veterans Administration (Texas Tech. University, 1978); studies done by the Army Corps of Engineers, FEMA and NOAA (USACE, 1990), the National Research Council (NRC, 1993), the Building Research Establishment in England (Cook, 1985), and Don Friedman at the Travelers (Friedman, 1987). Other pertinent references include Davenport et al. (1989), Hart (1976), Liu et. al. (1989), McDonald (1986, 1990), Mehta (1983, 1992), Minor (1979), Sparks (1988, 1990, 1993), Stubbs (1993), and Zollo (1993). RMS engineers reviewed damage reports and special studies, and participated in conferences on major hurricanes in the U.S. and around the world.

The RMS engineering staff includes several engineers with Ph.D.s in Civil and Structural Engineering. These engineers have significant experience and expertise in the understanding of building performance and structural vulnerability, and are dedicated to the development of vulnerability relationships for risk models worldwide. Dr. Boissonnade, and Dr. Gunturi from RMS have published several papers related to wind hazards and vulnerability. In conjunction with an agreement with Texas A&M University, RMS engineers participated in reconnaissance missions for Hurricanes Opal and Erin in Florida, Hurricanes Fran and Bonnie in North Carolina, and Hurricane Marilyn in the Virgin Islands and Puerto Rico. RMS engineers have also conducted several post-event reconnaissance in the aftermath of hurricanes e.g. Hurricanes Opal (1995) and Erin (1995) in Florida, Hurricane Marilyn (1995) in the Virgin Islands and Puerto

Rico, Hurricanes Fran (1996) and Bonnie (1998) in North Carolina, Hurricane Georges (1998) in the U.S. Gulf Coast and Puerto Rico and Hurricane Floyd (1999) in North Carolina, Typhoon Paka (1997) in Guam, Typhoon Angela (1995) (Rosing) in the Philippines and Typhoon Ryan (1995) in Japan. RMS engineers also performed reconnaissance missions for Hurricane Georges in the U.S. Gulf Coast and Puerto Rico, Hurricane Floyd in North Carolina, Typhoon Angela (Rosing) in the Philippines, Typhoon Ryan in Japan and Typhoon Paka in Guam.

RMS has also worked with several outside authorities on the development of the vulnerability functions including Professors Dale Perry and Norris Stubbs of Texas A&M University and Professor Peter Sparks from Clemson University. Other wind experts around the world who have contributed to the development of the hurricane model include Dr. Georgiou in Australia, Professor Mitsuta at University of Kyoto, Japan, and Dr. Greg Holland at the Bureau of Meteorology Research Center in Melbourne, Australia

9. What efforts have been made to update or revise the model or specific parts of the model? How many times have revisions been made? Discuss which changes are considered substantive and which are considered technical. When did the revisions occur? What specific revisions were made?

The only change in the model since last year's submission is a change in hurricane rates, reflecting one additional year of historical data. This results in a very slight change to modeled losses.

The RMS hurricane model was first released in 1993. A major revision occurred in 1997 as a result of a very comprehensive review of all aspects of the model. The major impetus of that update was a significant increase in the amount of hurricane data (most importantly wind speed data and client loss data) that RMS accumulated. That data allowed RMS to extensively calibrate hurricane parameter and building vulnerability relationships. In 1998, a major reconfiguration of the software structure was made to improve functionality and usability. These changes had no impact on engineering and actuarial models and did not have an impact on loss cost calculations.

In 2000, the methodology for modeling storm tracks was revised. This was the most significant update since the 1997 model revision. A random-walk approach was employed to generate the stochastic tracks. The advantages of this approach are that: • Storms can be simulated over the entire Atlantic basin. • Tracks have physically realistic behavior but are entirely synthetic. • Each synthetic track is different from every other track. • The tracks are not straight lines. • The pressure history of each track is realistic and respects the SST and topography crossed by the storm as well as the historically observed filling rates. • The calibration of the tracks and validation of the technique follows the more traditional methodology of NWS 38.

On going review of the 2000 model resulted in a revision of the RMS hurricane model, released in February 2003. The revised wind field model includes a time stepping model and a new treatment of surface effects. In addition, the wind hazard is calculated at grid cell points, using the Variable Resolution Grid (VRG). The use of VRG is a new process for the 2002 model, eliminating the use of centroid longitude/latitude coordinates for storing hazard values. The impact of this change is an increase or decrease of losses depending on the joint distributions of exposure and wind hazard within the ZIP Code.

The model also includes the impact of extratropical transition on hurricane windfields. The impact of transition on loss costs in Florida is very small.

The revised storm frequencies and filling rates of storms overland resulted in changes to modeled losses.

The damage curves were updated because of the new loss data obtained since the last revision. This resulted generally in a slight change of modeled losses.

The general policy of RMS has been to upgrade the hurricane model whenever new data or research becomes available which results in a non-trivial improvement in the loss modeling methodology. In the past, updates to accommodate new ZIP Codes were made at least every 24 months where possible. When ZIP Code files were updated, associated ZIP Code-related databases, such as those containing distance-to-coast and surface roughness, were also updated.

10. Describe methods and procedures available to the model user so that the user may incorporate modifications into the model See section I.A.5.1.

B. Loss Costs

1. Does the model produce the same loss costs if it runs the same information more than once (i.e., not changing the seed of the random number generator)?

The model produces the same results every time it analyzes the same information.

2. What is the highest level of resolution for which loss costs can be provided? What resolution is used for the reported output ranges? Describe how the model handles beach/coastal areas as distinct from inland areas.

Loss costs are generally provided at a ZIP Code level. RiskLink can analyze data at street address level (highest resolution) using VRG or at ZIP Code level. Since the resolution of the VRG cells is highest in coastal areas, particularly urban areas, where the gradient of hazard is greatest, beach/coastal exposures are explicitly evaluated at a higher level of resolution than inland areas. Output ranges submitted with Module 3 were calculated at ZIP Code resolution.

Loss costs are generally provided at a ZIP Code level. RiskLink can analyze data at street address level (highest resolution) or ZIP Code level. Street level resolution provides more accurate loss results because parameters such as distance to track and distance to coast, are directly related to the site and not to a fictitious exposure location such as ZIP Code centroid. Each site, whether geocoded to the street address or the ZIP Code centroid, is evaluated based on distance to coast, distance to storm track, and local topographical characteristics. As such, beach/coastal exposure is implicitly evaluated for each site. Output ranges submitted with Module 3 were calculated at ZIP Code resolution.

3. How does the model handle deductibles (both flat and percentage), policy limits, replacement costs, and insurance-to-value when estimating loss costs?

The Financial Module in RiskLink accurately models deductibles, limits, and reinsurance structures. RiskLink can analyze complex insurance and reinsurance policies and contracts at a site level, account (i.e., several sites under the same policy), or portfolio level for three coverages (i.e. structure, contents, and ALE). RiskLink computes the mean and CV for each loss and fits a beta distribution, then uses standard probabilistic computations to assign losses to each participant (insured, insurer, and reinsurer). RiskLink computes the loss as a percentage of the replacement costs which are input parameters. If the insured value is lower than the replacement cost, that value is treated as a limit to the insurer’s liability. Otherwise, the insured value is assumed to be the same as the replacement cost.

4. Are annual aggregate loss distributions available? What review or tests have been done on these?

Aggregate loss distributions are not used for ratemaking purposes, and are therefore not included in this version of RiskLink, although they are available in internal analysis and in previous versions of RiskLink. The financial module in RiskLink generates the Aggregate Exceedance Probability (AEP) curves that relate to aggregate loss distributions. These curves have been reviewed for reasonableness and consistency for all financial perspectives.

5. How are loss adjustment expenses considered within the loss cost estimates?

RiskLink loss results exclude loss adjustment expenses.

6. Can the model distinguish among policy form types, for example, home- owners, dwelling property, mobile home, renters, condominium owners etc., and if so, what are the assumptions? Does the model produce loss costs for different types of policies, for example, structure and contents, loss of use, mobile home, commercial residential, or contents only? Discuss in detail.

Policy forms vary in their terms and conditions, and RiskLink can model these variable terms. The modeling capabilities include variability in construction (several types of construction classes including mobile homes), coverages for building, contents and ALE, or A, B, C and D for Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 62 MODULE 1: I. General Description of the Model personal lines coverages. Given these variables as input, any combination or policy form can be modeled for either commercial or personal lines. 7. Provide a list of all engineering and actuarial modifications made to the model including the range of possible impacts on the loss costs produced by the modification.

As explained in section 5.3.5, the model can consider the impact of specific building characteristics or mitigation measures through the use of secondary modifiers. Each modifier listed in section 5.3.5 has several options that can be selected by the user. For instance, the modifier roof geometry has options like, gable roof low pitch, gable roof high pitch, hip roof, flat roof etc. The range of possible impacts on loss costs due to engineering modifications (or secondary modifiers) is given in Table 1.I.4. The specific impact will vary within these ranges depending on the modifier option used and the location of the exposure. There are no actuarial modifications made to the model.

Table 1.I.4 Impacts on Loss Costs Using Engineering Modifications Roof Strength State Avg. Impact Roof Shape State Avg. Impact Concrete Fill -17% Hip Roof -23% Concrete/Clay Tiles -11% Single Ply Membrane -8% Roof Covering Performance State Avg. Impact 6d Nails High Wind Nailing Schedule -3% Roof-to-wall strength State Avg. Impact 8d Nails Unknown Nailing Schedule -10% Metal or Bolt Anchors (High Strength) -5% 8d Nails High Wind Nailing Schedule -11% 10d Nails -11% Opening protection State Avg. Impact Simple Plywood Shutter -9% Wall-to-floor-to-foundation strength State Avg. Impact Well-designed Plywood Shutter -10% Engineered -4% Engineered Shutter -14% Window, door and skylight strength State Avg. Impact Good design for wind protection -23% Average design for wind protection -7%

C. Other Considerations

1. Describe how the model takes into consideration the following:

a. Socio-economic effects resulting from a large catastrophe, both upside as in FEMA mitigation and downside as in labor and material shortages.

RiskLink does not explicitly take into account socio-economic effects resulting from a large- scale catastrophe. There is no consideration of FEMA mitigation in the model.

b. Building code and enforcement differentiation.

The model does not explicitly address building code enforcement differentiation, but these factors are taken into consideration through the use of secondary and year modifiers as explained in section 5.3.4 and 5.3.5.This can be taken into consideration by designating a site specific characteristic secondary modifier as described in 1.A.5.1.

c. Specific construction characteristics (e.g., use of hurricane shutters)

This can be taken into consideration by designating a site-specific characteristic secondary modifier, see 1.A.5.

d. Storm surge and flood damage to the infrastructure.

While the RMS hurricane loss model is able to estimate hurricane losses from flood and storm surge, this is an optional analytical feature that must be specifically “turned on” to include such losses. If this option is not activated, flood and storm surge losses are excluded from all loss cost projections. Direct flood damage to infrastructure is not calculated in the model however, the impact on ALE losses due to storm surge damage to infrastructure was not excluded in calibrating ALE loss functions.

The storm surge model assesses the impact of storm surge associated with landfalling hurricanes in the United States. The model initially calculates the surge along the coast, and then attenuates the impact of the surge flooding further inland. To analyze the impact of storm surge at a particular location, the model first calculates the surge at the coastal point closest to the site of interest. In finding the surge at a point on the coast, the model considers the central pressure, distance from the landfall point, radius to the maximum winds, angle of coastal crossing, the forward velocity of the storm and the shoaling factor of the landfalling basin (dependent on the basin’s bathymetry).

The peak surge at any point along the coast is initially based on a linear relationship between surge and pressure. For a given location, surge is calculated using the pressure at the coastal point nearest to the site. This basic surge height is then adjusted for the forward velocity of the storm, the angle of landfall, and the shoaling factor for the landfall basin.

Once the surge level at the coast is calculated, it can be applied to locations further inland. For an inland location, the flood level at the coastal point nearest to the site is attenuated to the site based on distance to coast and elevation. The extent of storm surge flooding is limited to a few kilometers inland. After calculating the surge level at the inland site, the surge level is reduced by the elevation of the site to obtain the flood water depth at the site. Flood depth values at the site are then used to estimate the damage due to storm surge. The details of the storm surge modeling are available for review by the Professional Team.

2. List the input variables for all of the categories in 1 above. a. N/A. b. See I.A.5. c. See I.A.5. d. See I.C.1. Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 64 MODULE 1: I. General Description of the Model

Selected References

American Society of Civil Engineers-ASCE, (1994), Minimum Design Loads for Buildings and Other Structures, ANSI/ASCE 7-93. Approved May 12, 1994, ANSI Revision of ANSI/ASCE 7-88.

Batts, Martin et al., (1980), Hurricane Winds in the United States. Journal of the Structural Division, ASCE, Vol. 106, ST10.

Cook, N.J., (1985), The Designer’s Guide to Wind Loading of Building Structures. Building Research Establishment Report, Butterworths, London, England.

Davenport, Alan G., Gill Harris, Don Johnson, (1989),The performance of Low Industrial Buildings in Hurricane Gilbert with Special Reference to Pre-Engineered Metal Buildings. Presented at the Metal Building Manufacturers Association Meeting at the University of Western Ontario, Aug. 14-15, 1989, London, Ontario.

Friedman, D.G., and Travelers Insurance Company, (1987), US Hurricanes & Windstorms - a technical briefing. Based on a presentation at a DYP Insurance & Reinsurance Research Group Workshop, May 1987. London Insurance & Reinsurance Research Group Ltd., UK.

Georgiou, P.N., (1985), Design Wind Speeds in Tropical Cyclone-Prone Regions. BLWT-2-1985. Submitted in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy. University of Western Ontario, London, Ontario, Canada.

Georgiou, P.N., A.G Davenport., and P.J. Vickery, (1983), Design Wind Speeds in Regions Dominated By Tropical Cyclones. Journal of Wind Engineering & Industrial Aerodynamics, 13, 139-159.

Graham, H.E., and D.E. Nunn, (1959), Meteorological Considerations Pertinent to Standard Project Hurricane, Atlantic and Gulf Coasts of the United States. National Hurricane Research Project Report No. 33, U.S. Department of Commerce.

Hart, Gary C., (1976), Estimation of Structural Damage Due to Tornadoes. University of California Los Angeles, Symposium on Tornadoes: Assessment of Knowledge and Implications for Man, Texas Tech University, Texas.

Herbert, Paul J., Jerry D. Jarrell., and Max Mayfield, (1993), The Deadliest, Costliest, and Most Intense United States Hurricanes of this Century (And other Frequently Requested Hurricane Facts.) NOAA Technical Memorandum NWS NHC-31, National Hurricane Center, Coral Gables, Florida.

Ho, Francis P., James C. Su., Karen L. Hanevich., Rebecca J. Smith., and Frank P Richards., Silver Spring, (1987), Hurricane Climatology for the Atlantic and Gulf Coasts of the United States. Study completed under agreement EMW-84-E-1589 for Federal Emergency Management Agency. U.S. Department of Commerce. NOAA Technical Report NWS 38.

Holland, Greg, (1980), An Analytic Model of the Wind and Pressure Profiles in Hurricanes. Monthly Weather Review, Vol. 8, August 1980.

J. H. Wiggins Company (1980), Assessment of Damageability for Existing Buildings in a Natural Hazards Environment, Volume 1: Methodology, Prepared for The National Science Foundation, Washington, D.C. Technical Report No. 80-1332-1.

Jarvinen, B.R., C. J Neumann, and Davis (1984), A Tropical Cyclone Data Tape for the North Atlantic Basin 1886-1983:Contents Limitations and Uses. NOAA Technical Memorandum NWS NHC 22, U.S. Department of Commerce. Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 65 MODULE 1: I. General Description of the Model

Kaplan/DeMaria (1995), “A Simple Empirical Model for Predicting the Decay of Tropical Cyclone Winds After Landfall,” Journal of Applied Meteorology, 34(11).

Liu, Henry., Wind Engineering - A Handbook for Structural Engineers. Prentice Hall, Englewood Cliffs, New Jersey.

Liu, Henry., Herbert S. Saffir., and Peter R. Sparks, (1989), Wind Damage To Wood-Frame Houses: Problems, Solutions And Research Needs. Journal of Aerospace Engineering, Vol.2, No. 2, April 1989.

McDonald, James R., and John F. Mehnert, (1990), A Review of Standards Practice of Wind Resistant Manufactured Housing. Journal of Wind Engineering and Industrial Aerodynamics, 36 (1990) 949-956.

McDonald, James R., and W.Pennington Vann, (1986), Hurricane Damage to Manufactured Homes. Presented at American Society of Civil Engineers Structures Congress ‘86, New Orleans, Louisiana, September 14-19, 1986.

Mehta, K.C., J.E.Minor., and T.A. Reinhold, (1983), Hurricane Speed- Damage Correlation in Hurricane Frederic, Journal of Structural Engineering, 109(1).

Mehta, Kishor C., Ronald H. Cheshire., and James R. McDonald, (1992), Wind Resistance Categorization of Buildings for Insurance. Journal of Wind Engineering and Industrial Aerodynamics, 41-44 (1992) 2617-2628.

Minor, J.E., and K.C. Mehta, (1979), Wind Damage Observations and Implications. Journal of Structural Division, ASCE, 105(ST11), Proc. Paper 14980, November 1979, pp. 2279-2291.

National Bureau of Standards-NBS (1981), Hurricane-Induced Wind Loads. PB82-132267 Prepared for the National Science Foundation, Washington, D.C.

National Research Council, Committee on Natural Disasters, (1993), Wind and the Built Environment. U.S. needs in Wind Engineering and Hazard Mitigation.. Panel on the Assessment of Wind Engineering Issues in the United States, Commission on Engineering and Technical Systems, National Academy Press, Washington, D.C.

Neumann, C. J., (1987), The National Hurricane Center Risk Analysis Program (HURISK). NOAA Technical Memorandum NWS NHC 38.

Neumann, C.J. et al., (1998), Tropical Cyclones of the North Atlantic Ocean, 1871-1998, with updates to 2000. Historical Climatology Series 6-2, National Climatic Data Center, Asheville, North Carolina.

NOAA (1979), Meteorological Criteria for Standard Project Hurricane and Probable Maximum Hurricane Windfields, Gulf and East Coasts of the United States, NOAA Technical Report NWS 23, Washington, D.C., September, 1979

NOAA (1987), Hurricane Climatology for the Atlantic and Gulf Coasts of the United States, NOAA Technical Report NWS 38, Washington, D.C., April, 1987

North Atlantic Storm Data Base, HURDAT

Sanchez-Sesma, Jorge., et al., (1988), Simple Modeling Procedure for Estimation of Cyclonic Wind Speeds. Journal of Structural Engineering, ASCE, Vol. 114, No. 2.

Schwerdt, R. W., et al., (1979), Meteorological Criteria for Standard Project Hurricane and Probable Maximum Hurricane Wind Fields, Gulf and Atlantic Coasts of the United States. NOAA Technical Report NWS23, U.S. Department of Commerce.

Simiu, Emil., and Robert H. Scanlan, Wind Effects on Structures - An Introduction To Wind Engineering. Second Edition. A Wiley-Interscience Publication. John Wiley & Sons (1986).

Simpson, Robert H., and Herbert Riehl, (1981), The Hurricane and Its Impact. Louisiana State University Press, Baton Rouge and London.

Sparks, P. R., and Herbert S. Saffir, (1990), Mitigation of Wind Damage to Non-Engineered and Marginally Engineered Buildings. Journal of Wind Engineering and Industrial Aerodynamics, 36 (1990) 957-966.

Sparks, P.R., M.L.Hessig, J.A.Murden., B.L.Sill, (1988), On the Failure of Single-story Wood-Framed Houses in Severe Storms. Journal of Wind Engineering and Industrial Aerodynamics, 29 (1988) 245-252.

Sparks, Peter R., and Shiraj A. Bhinderwala, (1993), Relationship Between Residential Insurance Losses and Wind Conditions in Hurricane Andrew. Conference - December 1993, Miami, Florida.

Sparks, Peter R., Damages and Lessons Learned from Hurricane Hugo. Dept. of Civil Engineering, Clemson University, Clemson, SC.

Stubbs, Norris., and A. Boissonnade, 1993, Damage Simulation Model for Building Contents in a Hurricane Environment. Proceeding of the 7th U.S. National Conference on Wind Engineering June 27, 1993, University of California Los Angeles.

Texas Tech Univ., Lubbock Inst. for Disaster Research, (1978), A Study of Building Damage Caused by Wind Forces. Prepared for Veterans Administration, Washington D.C., Office of Construction, U.S. Dept. Of Commerce National Technical Information Service, PB-286-604.

Tropical Prediction Center/National Hurricane Center (TPC/NHC), Tropical Cyclones of the North Atlantic Ocean, 1871-1998, with updates

Tryggvason, B. V., et al., (1976), Predicting Wind-Induced Response in Hurricane Zones. Journal of the Structural Division, ASCE, Vol. 102, No. 102, No. ST12.

U.S. Army Corps of Engineers-USACE (1990), Tri-State Hurricane Loss and Contingency Planning Study Phase II: Alabama, Florida, Mississippi. Executive Summary and Technical Data Report, US Army Corps Engineers, Mobile District.

U.S. Geological Survey, (1978), A Land Use and Land Cover Classification System for Use with Remote Sensor Data. Geological Survey Professional Paper 964, U.S. Dept. of Interior, Reston, Virginia.

U.S. Geological Survey, (1990), Digital Elevation Models -- Data Users Guide 5. U.S. Dept. of the Interior, Reston, Virginia. Wieringa, J., (1993), ‘Representative roughness parameters for homogeneous terrain’, Boundary- Layer Meteorol 63, 323-393.

Wieringa, J., (2001), ‘New revision of Davenport roughness classification’, in Proc. 3EACWE, Eindhoven, The Netherlands, July 2-6, 2001, 285-292.

Zollo, R.F., (1993), Hurricane Andrew August 24, 1992, Structural Performance of Buildings in Dade County, Florida, Technical Report No. CEN 93-1 Univ. of Miami, Coral Gables, Florida. II. Specific Description of the Model

A. Model Variables

1. Using the list of model variables provided in response to I.A.5, describe the source documents and any additional research that was performed to develop the model’s variable functions or databases. Particularly describe all such information, including a description of the historical database(s), for the model’s hurricane wind speeds and hurricane frequencies. Were there any assumptions used in creating any of these databases? Describe any deviation from the Commission’s hurricane set. Describe intensities used for these hurricanes

1.1 Hazard Variables and Databases

1.1.1 Historical Storms Database RMS engaged the services of Mr. Neumann to help develop its own, updated and improved, catalog of historical events. The updated version of HURDAT database was used as a starting point. RMS then performed extensive research on all historical storms and cross-checked the data from several references for accuracy, reliability and completeness. For a discussion of the development of the historical database, see Module 1, Section I.A., subsection 2.2.1.

1.1.2 Stochastic Storms Database The RMS stochastic database is generated using a random-walk simulation approach. For more information see the general description of the model in Module 1, Part 1, Section I.A.2.

1.2 Exposure Variables and Databases

1.2.1 Geocode Data Base A proprietary data base is being used for assigning a geographical coordinate to a location for which information is input at ZIP Code level. If the building location is input at a street level, then the proprietary geocoding software is used to determine the latitude and longitude based on the street address. If the building location is entered as a ZIP Code, then the model uses wind speeds for that ZIP Code, calculated from the underlying higher resolution VRG cells.

1.2.2 Building Inventory Data Base RiskLink has a database of building inventory for each county in each state, as well as for a number of ZIP Codes in certain urban commercial areas. The database indicates the distribution of construction class for the residential and commercial occupancies. The inventory database is used in the case where the building construction class is unknown. In that case RiskLink will develop a composite curve to calculate damageability based on occupancy, number of stories and year of construction. RMS conducted a survey of building officials and structural engineers to confirm construction class mixes based on occupancy and building height.

1.3 Building Characteristics Variables and Databases

1.3.1 Damage Database Model variables covered in the damage curves database include vulnerability class as a function of construction material, number of stories, and occupancy. For each class the damage ratios at different wind speed levels are stored. Damage functions were calibrated with historical loss data. These databases are developed for building damage and contents damage due to wind, surge (low velocity flood) and wave (high velocity flood). Damage functions for Additional Living Expenses (ALE) are based on reviews of insurance loss data and elicitation of opinions from claim adjusters.

1.3.2 Secondary Characteristics Database For a discussion of the development of secondary modifiers see Standard 5.3.5 and Module 1, Section I.A.5.

Some of the references used in the development of the secondary modifiers database are listed below:

ASCE (1998), “ASCE7-98 - Minimum Design Loads for Buildings and Other Structures”, American Society of Civil Engineers, Reston, VA.

Ayscue, J. K. (1996), “Hurricane Damage to Residential Structures: Risk and Mitigation”, Natural Hazards Research and Applications Information Center Institute of Behavioral Science, University of Colorado.

Baskaran, A. and Dutt, O. (1997). “Performance of Roof Fasteners Under Simulated Loading Conditions”, Journal of Wind Engineering and Industrial Aerodynamics, 72, 389-400.

Chiu, G. L.F., Perry, D. C. and Chiu, A. N. L. (1994). "Structural Performance in Hurricane Iniki." In Proceedings of Seventh United States National Wind Engineering Conference, Volume I, Gary C. Hart, Editor. Washington, D.C.: National Science Foundation.

Cook, R. L., Jr. (1991). "Lessons Learned by a Roof Consultant." In Hurricane Hugo One Year Later, Benjamin A. Sill and Peter R. Sparks, Editors. New York: American Society of Civil Engineers.

Copple, J.H. (1985), “A review and analysis of building codes and construction standards to mitigate coastal storm hazards,” Hazard Mitigation Research Program, The Center for Urban and Regional Studies, The University of North Carolina, Chapel Hill, North Carolina.

Crandell, J. H., Gibson, M. T., Laatsch, E. M. and Overeem, A. V. (1994). “Statistically-Based Evaluation of Homes Damaged by Hurricanes Andrew and Iniki”, Hurricanes of 1992, ASCE.

Cunningham, T. P. (1994). “Evaluation of Roof Sheathing Fastening Schedules for High Wind Uplift Pressures”, Hurricanes of 1992, ASCE.

FEMA (1992). “Building Performance: Hurricane Andrew in Florida; Observations, Recommendations, And Technical Guidance”, Federal Emergency Management Agency

FWUA (2000). Florida Windstorm Underwriting Association: Manual of Rates, Rules and Procedures, FWUA, Jacksonville, Florida. Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 69 MODULE 1: II. Specific Description of the Model

HUD (1993). “Assessment of Damage to Single-Family Homes Caused by Hurricanes Andrew and Iniki”, U.S. Department of Housing and Urban Development, Office of Policy Development and Research.

IBC (2000), “2000 International Building Code”, International Code Council, Falls Church, Virginia.

Kumamaru, M. Tsuru, N. Maeda, J. and Miyake, A. (1999), “Some effects of roof shapes on housing wind loads,” Proceedings of 10th ICWE, Copenhagen, Denmark.

Manning, B. R. and Nichols, G. R. (1991). "Hugo Lessons Learned." In Hurricane Hugo One Year Later, Benjamin A. Sill and Peter R. Sparks, Editors. New York: American Society of Civil Engineers.

Minor, J. E. and Behr, R. A. (1994). “Improving the Performance of Architectural Glazing in Hurricanes”, Hurricanes of 1992, ASCE.

Mitrani, J. D.; Wilson C. B. and Jarrell, J. (1995). “The effectiveness of Hurricane shutters in Mitigating Storm Damage. Technical Publication No.116. Miami, Florida: Department of Construction Management, Florida International University, Miami, Florida.

Oliver, C. and Hanson, C. (1994). “Failure of Residential Envelopes as a result of Hurricane Andrew in Dade county, Florida”, Hurricanes of 1992, ASCE.

Peacock, W. G., Morrow, B. K. and Gladwin, H. (1998). “South Florida Mitigation Baseline Survey Report”, International Hurricane Center and Institute for Public Opinion Research, Florida International University, Miami, Florida.

SBC (1997), “1997 Standard Building Code”, Southern Building Code Congress International, Birmingham, Alabama.

SBCCI (1999a), “SBCCI Test Standard for Determining Wind Resistance of Concrete or Clay Roof Tiles - SSTD 11-99”, Southern Building Code Congress International, Birmingham, Alabama.

SBCCI (1999b), “SBCCI Test Standard for Determining Impact Resistance From Windborne Debris - SSTD 12-99”, Southern Building Code Congress International, Birmingham, Alabama.

Smith, T. L. (1994). "Causes of Roof Covering Damage and Failure Modes: Insights Provided by Hurricane Andrew." In Hurricanes of 1992, Ronald A. Cook and Mehrdad Soltani, Editors. New York: American Society of Civil Engineers.

Wolfe, R. W.; Ramon M. Riba; and Mike Triche. 1994. "Wind Resistance of Conventional Light-Frame Buildings." In Hurricanes of 1992, Ronald A. Cook and Mehrdad Soltani, Editors. New York: American Society of Civil Engineers.

2. List the current primary databases used by the model and the aspects of the model to which they relate. Indicate which databases are “public” and which are “proprietary.”

Table 1.II.1 lists the primary databases used by the model.

Table 1.II.1 Primary Databases Used by the Model Primary Databases Status How Used RMS Historical Proprietary Used in calculating the wind speeds due to historical storms. Storms Database This information is used in generating losses for historical storms.

Stochastic Storms Proprietary Used for calculating the wind speeds due to stochastic storms at Database each location. This information is used in calculating losses for stochastic storms.

Damage Database Proprietary Used for estimating damage at a given wind speed

Secondary Proprietary Used for modifying damage at a given wind speed based on Characteristics specific building characteristic. Database Year Modifier Proprietary Used for modifying damage based on the year of construction database of the building.

Building Inventory Proprietary Used for estimating damage at a given wind speed when Database building class is unknown

3. What assumptions are made in the following areas:

a. Meteorology

In modeling wind speeds, hurricanes are characterized by the central pressure, the radius of maximum wind, forward velocity, track and direction. The wind speeds are calculated based on the windfield model presented by Dr. Peter Georgiou (1983, 1985) and calibrated with historical storm data.

b. Damageability

The RMS building classification adequately characterizes the major differences between residential construction building stock.

c. Insurance Coverage

Policies are current and enforceable.

How does the model address the issue of demand surge?

Loss cost projections produced by the model do not include demand surge.

4. Are there other major or significant assumptions not listed above? If so, describe. There are none.

5. Describe the nature and extent of actual insurance claims data that have been used to develop the model’s vulnerability functions (damage ratios). Describe in detail what is included, such as, number of policies, number of insurers, and number of units of dollar exposure; separate into personal lines, commercial, and mobile home.

RMS has collected loss data from its clients for the purpose of developing and calibrating the model’s vulnerability functions. The datasets have different resolutions and are used for different validation purposes. Data containing detailed information on damage, loss by construction class and exposure by ZIP Code is used for calibration of vulnerability functions. Aggregated data is used primarily for sensitivity analyses. To adequately use loss data for development of vulnerability functions, the data must contain several types of information including: loss per coverage (A, B, C, and D), line of business, exposure value per coverage, description of structures (construction type, etc.), and actual location of structures. Overall, RMS has used over $7 billion of loss data and over $1 trillion in corresponding exposure data in the development and calibration of damage functions. A sample of the datasets is shown in Table 1.II.2 Table 1.II.2 Sample of Datasets Used for Development and Calibration of Vulnerability Functions LOB* Storm Company Data Resolution RES Andrew A ZIP/Coverage/Construction Class RES Andrew B ZIP/Coverage RES Andrew C ZIP/Construction Class RES Bob A ZIP/Coverage/Construction Class RES Erin A ZIP/Coverage/Construction Class RES Fran A ZIP/Coverage/Construction Class RES Fran B ZIP/Coverage RES Hugo A ZIP/Coverage/Construction Class RES Hugo B ZIP/Coverage RES Opal A ZIP/Coverage/Construction Class RES Georges D ZIP/Coverage/Construction Class MH Fran E ZIP/Coverage/Construction Class MH Hugo F ZIP/Coverage/Construction Class *RES – Residential; MH – Mobile Homes

B. Methodology

1. Specify the wind speed(s) (e.g., one-minute sustained, peak gusts, etc.) used for loss estimation.

Three-second peak gust wind speeds are used for loss estimation.

2. How is the asymmetric nature of hurricanes considered?

The asymmetric nature of hurricanes is considered in a number of ways. Firstly, by taking into account the forward velocity of the hurricane, the asymmetry in windspeeds on the left and right

Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 72 MODULE 1: II. Specific Description of the Model hand sides of the hurricane is modeled. The forward velocity term and the hurricane cyclonic windspeed terms are additive, which will cause the windspeed on the right hand side of the hurricane to be higher, and on the left hand side of the hurricane to be lower. Secondly, as hurricanes begin to undergo extratropical transition, an asymmetry in the size of the radius to maximum winds begins to occur, in which the radius to maximum winds on the right hand side gets larger relative to that on the left hand side. This asymmetry is captured in the sampling of the radius to maximum winds values for transitioning storms. The mean gradient wind speed, Vg, is calculated from the formula:

g = T α fRSinVV ))((5.0 +− 1 B B  Rmax  2  −   (4)  2  ∆P  Rmax   R   T α ))((25.0 +−  BfRSinV   e   ρ  R     where R = radial distance from the storm to the site; α = angle from storm track to site (clockwise is positive); ∆P = central pressure difference; VT = storm translational speed; ρ = air density; f = Coriolis parameter (function of latitude); B = pressure profile coefficient and Rmax = radius to maximum winds. The VT term (storm translational speed) accounts for the fact that the translational storm speed is additive on one side of the hurricane and subtractive on the opposite.

3. Describe the nature of the filling rate function used.

See “Pressure Filling” in Module 1, Section I.A, sub-section 2.2.1.

4. Other than the hurricane’s characteristics, what other variables affect the wind speed estimation (e.g., surface roughness, topography, etc.)? Describe the database used for land friction calculation and its compatibility with the friction model.

Variables that affect modeled windspeed are the surface roughness conditions, both at the site and upstream to the site by direction. The database that describes the surface roughness in terms of the roughness length is the National Land Cover Data (NLCD) dataset produced by the USGS. This dataset is derived from early to mid-1990’s Landsat Thematic Mapper satellite data and provides coverage of the entire continental United States at a horizontal resolution of 30-metres, using a 21-class land cover classification scheme. Further processing of areas classified as urban or suburban in this database is then undertaken by RMS to differentiate areas of differing building heights. This is done primarily using data on the construction square footage by ZIP Code. At the same time, those land cover classes whose effects on the surface wind speed are similar were merged into a single land use class. The end result is a 10-class land cover database with land cover classes ranging from water to high-rise buildings. Finally, a representative roughness length is assigned to each of the 10 land cover classes, using published mapping schemes from the scientific literature. Variables that affect the windspeed estimation are the distance of the site from the coastline and local roughness conditions. An initial index of local roughness conditions is developed for each ZIP Code from construction density and land use data. A smoothing algorithm then adjusts this index to reflect the impact of roughness in the immediate neighboring regions. Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 73 MODULE 1: II. Specific Description of the Model

5. Identify the characteristics (e.g., central pressure, radius of maximum winds, etc.) of a hurricane that are used in estimating wind speeds and how this information is applied for the entire state of Florida.

The parameters are: central pressure difference, forward velocity, Coriolis parameter, radius of maximum wind, angle of track, landfall location, and storm track (i.e., distance from site to the eye of storm). For a complete description, see response to Module 1, Section I.A, “General Description of the Model”.

6. Which variables in the wind speed component are dependent, and how is this dependence incorporated in the model?

See Module 1, Section II.B, #11 below and the general description of the model in Module 1, Section 1.A.

7. Describe how the coastline is segmented (or partitioned) in determining the parameters for hurricane frequency used in the model. Provide the hurricane frequency distribution by intensity for each segment.

See Module 1, Section I.A., Sub-section 2.2.1 for a description of coastline segmentation. The historic and stochastic frequencies of storms by intensity for each segment are shown in Figure 5.3 in the Standards as well as in Module 3, Table 3.I.3.

8. If stochastic simulation techniques are used, describe how the hurricanes are generated from the underlying probability distributions. How are landfall sites, hurricane paths, and decay rates determined?

See Module 1, Section I.A, Sub-section 2.2.1.

9. Does the model produce confidence intervals for:

a. Wind speed estimates given a set of hurricane parameters?

The model incorporates confidence intervals on the wind speed given the storm parameters. A lognormal distribution is used for representing the variability. Variability of wind speed has been estimated by comparing predicted versus observed (NHC) wind speeds at diverse locations for diverse storms.

b. Damage estimates given a wind speed estimate?

The model produces confidence intervals for damage estimates given a windspeed estimate. RiskLink uses the beta distribution to represent the uncertainty in the damage. See response to Module 1, Section I.A, Sub-section 2.2.4 for further information.

c. Annual loss costs?

The model does not produce confidence intervals for annual loss costs. Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 74 MODULE 1: II. Specific Description of the Model

Characterize the uncertainties in the model, for example, with an uncertainty analysis. Uncertainty refers both to possible model misspecifications and inherent random variation. (Standards 5.6.3 and 5.6.5)

During the Professional Team’s visit, the results of an independent uncertainty analysis will be presented.

10. Describe the method or methods used to estimate annual loss costs needed for ratemaking. Identify any source documents used and research performed.

Expected losses associated with each stochastic storm are multiplied by the annual probability of occurrence for the corresponding storm. These are summed over all storms to determine the average annual loss. If the analysis is done on a ZIP Code basis, the storm intensity (peak gust) and frequency (mean annual rate) of the stochastic storm database are already pre compiled into RiskLink for each ZIP Code to speed up calculations.

11. What functions or variables does the model consider to be independent? For those functions that are not independent, describe the source of dependence such as latitude. Are there limitations on the functions or variables that are a function of latitude? If so, describe. What are the intermediate (endogenous) variables that are part of the calculations between the inputs and outputs described in I.A.5?

Hurricane tracks (location, forward velocity and track angle) are calibrated against the historical record in all parts of the basin. Tracks are independent of the central pressures but the central pressures are dependent on the tracks as they determine which SSTs and topography the storms cross.

Windfield Model Dependent Variables

R Radial distance from storm to the site (dependent on site location) α Angle from storm track to site (dependent on site location) ρ Air density (dependent on pressure) f Coriolis parameter (dependent on latitude) B Pressure profile coefficient (dependent on latitude) Rmax Radius to maximum wind (dependent on central pressure and latitude) Roughness coefficient (dependent on surface roughness conditions site location)

Intermediate variables are the wind speed variables (Vs, Vg) as discussed in the response to Module 1, Section I.A, Sub-section 2.2.2.

12. Identify the form of the probability distributions used for each function or variable, if applicable. What statistical techniques were used for distributions that are estimates? What tests were used for goodness-of-fit?

The parameter distributions are based on smoothed historical observations. Upper-bound storm intensities (minimum obtainable pressures) were estimated based upon more recent research into the relationships between sea-surface temperature and other physical parameters of the hurricanes.

Distributions of storm azimuth and forward speed were calculated from the stochastic tracks and compared against the smoothed historical observations at gates around the coast of Florida. Chi- square tests and other statistical tests have been performed for goodness of fit. Based upon these tests, the random-walk technique was calibrated as explained in Module 1, Section I.A., Sub- section 2.2.

13. What is the most sensitive aspect of the model? Is this sensitivity based on a) an assumption, b) an underlying datum unique to the model, or c) a technique that the model employs? Discuss fully and provide an example to illustrate how (to what degree) this sensitivity affects output results.

Testing during the model development process confirmed no inappropriate sensitivity of the RMS hurricane model to assumptions or variations in parameter values. The areas of greatest significance to model loss cost output are the following:

The rate of hurricanes. This is based on the past 100 years of historical observations, a relatively short period. Because total damage rises sharply with hurricane intensity, intense hurricanes [Saffir-Simpson Intensity scale (CAT) of 3 and above] contribute disproportionately to loss costs. Consequently, a small increase in the rate of occurrence of intense storms could lead to a relatively large increase in average annual losses. Refer to Section 5.2.7 of the Report of Compliance with Standards for a discussion of the technique used to smooth inconsistencies and inappropriate variability in the relatively brief historical record.

Local roughness assignments. As described in Section 5.2.9 of the Report of Compliance with Standards, the model recognizes the impact of surface friction and reduces modeled wind speeds accordingly. At some wind speeds, a change in wind speed can have a proportionately larger impact on the output results. Directional coefficients, which take into consideration the impact of roughness up to 50 miles upwind for 8 different directions, are stored at the VRG level of resolution.

Vulnerability functions. These functions translate wind speeds into damage ratios for the various types of construction and coverage (see Module 1, Section I.A, Sub-section 2.2.3). As noted above, damage rises sharply at some wind speeds as wind speed increases.

14. Are there other aspects of the model that may have a significant impact on the sensitivity or variation in output results?

Every parameter in the model has an impact on estimated loss costs, and accordingly has been studied in detail by RMS.

15. What sensitivity and uncertainty analyses have been performed on the model?

Numerous analyses have been performed to quantify, calibrate, and validate the impact of various portions of the model on loss estimates. Through the development process, several model iterations were generated and tested thoroughly in order to arrive at final values for such variables as the roughness assignments, parameter distributions, windfield profiles, and damage curves. In addition, sensitivity and uncertainty analyses have been done using central pressure, forward velocity, radius to maximum winds, and vulnerability curves as variables.

C. Validation Tests

1. What were the nature and results of the tests performed to validate the wind speeds generated?

Windspeeds predicted for many hurricanes such as Georges (1998), Fran (1996), Opal (1995), Erin (1995), Andrew (1992), Bob (1991), and Hugo (1989), have been compared to observed windspeed data obtained from Monthly Weather Review publications and NHC reports. Additionally, predicted windspeeds have been compared to event windspeed contour maps produced by the Hurricane Research Division of NOAA.

2. What were the nature and results of the tests performed to validate the expected loss estimates generated? If a set of simulated hurricanes or simulation trials was used to determine these loss estimates, specify the convergence tests that were used and the results. Specify the number of hurricanes or trials that were used.

The losses produced by the set of stochastic storms have been compared to losses produced by historical storms impacting Florida.

RMS has validated estimates by first comparing the modeled frequency of various storm characteristics with the historic record. The number of modeled storms of various intensities landfalling in each of the segments aligns with the historical record.

The losses produced by the set of stochastic storms have been compared to losses produced by historical storms impacting Florida. For example, industry loss levels at various return periods calculated using the stochastic model were compared to loss levels for return periods in the historical record. In addition, the geographic progression of loss costs by ZIP was reviewed for smoothness, consistency and logical relation to risk.

In order to ensure that the set of stochastic storms is sufficient and converges, multiple trial runs were performed using increasing numbers of storms. Ratios of average annual loss were compared for successive runs for each ZIP Code. The final number of storms used in the model was then chosen so that the change in average annual loss on a county and ZIP Code was within an acceptable range. The final model uses 20,394 approximately 50,000 landfalling storms for loss cost calculations.

3. What were the nature and results of the tests performed to validate the damage estimates generated?

Insurance companies have supplied RMS with datasets containing the locations and building types associated with coverage and loss amounts. These datasets have been run against historical storms and the computed losses have been compared to the actual losses. Table 1.II.3 shows a sampling of aggregated loss comparisons by company. Table 1.II.3 Sample Client Loss Data Comparison (losses normalized such that maximum actual loss = $1,000,000) Comparison Storm TIV Actual Loss Predicted Loss Ratio A Andrew $21,493,025 $670,765 $516,476 0.77 B Andrew $12,459,123 $1,000,000 $933,547 0.93 C Hugo $7,306,514 $57,255 $55,977 0.98 D Hugo $3,795,591 $92,775 $85,689 0.92 E Fran $7,014,865 $50,448 $34,315 0.68 F Georges $254,159 $6,298 $4,776 0.76

Additionally, RMS has calculated losses for all historical storms which have made landfall in the U.S. during the last century. Table 1.II.4 shows a comparison between industry losses as reported by the Property Claims Service (PCS) and RMS modeled estimates for significant recent storms. The PCS loss numbers have been adjusted to correspond to 2000 loss numbers to account for increases in exposure and inflation. Table 1.II.4 Comparison of Actual and Estimated Industry Loss ($ million) Storm Year Actual Loss* RMS Estimate** Georges 1998 $1,300 $1,400 Fran 1996 $2,000 $2,200 Opal 1995 $2,700 $2,600 Erin 1995 $490 $400 Andrew 1992 $25,300 $21,500 Bob 1991 $900 $950 Hugo 1989 $5,700 $5,200 *PCS estimate, adjusted to 2000 dollars ** Loss does not include demand surge or loss adjustment expense

4. Were insured losses from ancillary perils included within the annual loss cost estimate? If so, describe which perils, the basis for the loss estimation, and the validity testing or peer review that was performed on these calculations.

Insured losses from ancillary perils were not included.

5. What were the nature and results of any validation tests on any other aspects of the model?

Modeled vs. observed loss for various historical events at both the industry and individual company level. Comparison of modeled vs. historical probabilities of exceeding various loss levels. Comparison of modeled vs. historical average annual loss. Comparison of modeled vs. historical rates of hurricanes by intensity along the Florida coastline.

6. Provide documentation of all validation tests performed.

The documentation is provided along with the respective sections described above. Further documentation on validation will be available for review during the Professional Team visit.

MODULE 2

Background/Professionalism

1. Company Background

A. Describe the ownership structure of the modeling company. Is the company affiliated with any other company? If so, describe the nature of the relationship.

Risk Management Solutions, Inc. (RMS) is a wholly owned subsidiary of DMG Information, Inc., Harmsworth Publishing, Ltd., part of the Daily Mail and General Trust plc, a UK Corporation.

B. How long has the company been in existence?

RMS was founded in 1988.

C. In what year was the model developed?

The original RMS Hurricane Model, IRAS Hurricane, was developed between 1991 and 1992 and first released in 1993. The RMS Hurricane Model has been thoroughly examined in 1996 and 1997. The model was revised again as part of a major two-year effort between 2001 and 2002, and was released in February 2003. The current submitted model is RiskLink version 4.3a.4.2 SP1a

D. How long has the model been used for ratemaking purposes?

The model has been used for ratemaking purposes since 1994.

E. In which states has use of the model been attempted for ratemaking purposes? Has the model been accepted for use in any state? If so, what state or states? Provide the Commission with the name of a contact person in all the states where the model has previously been used for ratemaking purposes. (The Commission may contact these persons to discuss the work performed.)

Individual companies as well as ISO file loss costs in most hurricane-prone states using the RMS Hurricane Model. These loss costs have been accepted in numerous states. In the past, Mr. George Burger, Assistant Vice President, Insurance Services Office, Tel: (898-5813212) was the ISO representative for RMS loss cost filings. However, as of June 2002, ISO is no longer a licensing client of RiskLink and the RMS Hurricane Model.

Two states where the RMS Hurricane Model was recently formally reviewed and accepted for filing insurance rates are Louisiana and Hawaii. Contacts in these states are:

Mr. Richard Piazza Louisiana Dept. of Insurance P.O. Box 94214 Baton Rouge, LA 70804-9214 (225)342-0895

Ms. Shelley Santo Rate & Policy Analysis Manager Insurance Division Department of Commerce & Consumer Affairs P.O. Box 3614 Honolulu, Hawaii 96811-3614

ISO has filed loss costs in most hurricane prone states using the RMS Hurricane Model. These loss costs have been accepted in numerous states. For information on the status of these filings, the ISO representative is: Mr. George Burger, Assistant Vice President Insurance Services Office Tel: (898 5813212)

F. Describe generally the modeling company’s services and the percentage of the company’s annual income derived from each.

RMS receives revenues from: • licensing of software products (83 90%) • providing analytical reports (8%) • consulting services (8 1%) • miscellaneous others (1%)

G. How long has the model been used for analyzing insurance company exposures or other such uses? Describe these uses.

The RMS hurricane model has been used by RMS and its clients for analyzing insurance and related property risks since 1993. Specifically, applications have included:

• Insurance company modeling of expected portfolio losses for specific events, loss cost development for rate filings, reinsurance program optimization, portfolio strategy and individual policy underwriting • Reinsurance modeling of contract and treaty structure, costing/pricing, portfolio strategy and capital allocation • Insurance Services Organization (ISO) modeling of industry loss costs rates by geographic zone, construction type, and line of business

• Broker and intermediary modeling of risk transfer and risk finance alternatives on a probabilistic cost/benefit basis • Corporate modeling of risk transfer needs and alternatives, including risk mitigation due to hurricane and storm surge risk • Issuer, investment bank and investor modeling of financial risk, expected yield, and risk correlation for bond issues based on catastrophe risk 2. Professional Credentials (Standard 5.1.2 for all items in this section)

A. List the names of the company’s technical staff and consultants with highest degree obtained (discipline and University), years of experience with hurricane modeling for ratemaking, and credentials and years of relevant experience.

Brief biographies of RMS’ technical staff are given below:

Richard R. Anderson, FCAS, MAAA, Chief Actuary Mr. Anderson is the Chief Actuary at Risk Management Solutions. Mr. Anderson’s responsibilities at RMS include research and development of the financial module of RMS’s catastrophe models, the modeling of uncertainty in the catastrophe models, and research and development of enterprise-wide risk modeling for property/casualty insurance companies. Mr. Anderson also has done research and development work on the systematic optimization of capital allocation and the inclusion of catastrophe model output into DFA models. Mr. Anderson earned his BS in Mathematics from Illinois State University. He is a Fellow of the Casualty Actuarial Society and a member of the American Academy of Actuaries. Hurricane Project Responsibilities: (1) Collecting Insurance Industry Loss for all historical events and updating the loss to current dollar value based on population growth and inflation, which is then used for loss calibration; and (2) Assessing uncertainty on model generated losses and assigning confidence levels (3) Sensitivity and uncertainty analyses.

Fouad Bendimerad, Ph.D., P.E., Technical Director Dr. Bendimerad holds M.S. and Ph.D. degrees in Civil Engineering from Stanford University. He has over twenty years experience in the field of structural engineering and risk analysis. He is known world-wide as an expert in damage and loss estimation from natural hazards and has published extensively in this subject. He is the secretary of the Earthquakes and Megacities Initiative, an international endeavor sponsored by the United Nations. His project oversight includes: (1) Probabilistic hazard modeling of natural hazards phenomena; (2) Modeling of structural performance of buildings, lifelines, and commercial / industrial facilities; (3) Earthquake damage estimation; and (4) Decision analysis. He is a principal in the highly complex team project "NIBS," developing nationally applicable standardized methods for assessing earthquake risks (physical damage, functional losses and economic losses) to buildings and other structural systems. Prior to RMS, Dr. Bendimerad spent seven years at Stanford University where he was in charge of the seismic risk program and maintained a Consulting Professorship in the Civil Engineering Department. Dr. Bendimerad is a Registered Professional Engineer in the State of California, and a member of several professional organizations including the American Society of Civil Engineers. Hurricane Project Responsibilities: Advisor on science and technical issues.

Auguste Boissonnade, Ph.D., Chief Scientist, Vice President of Model Development for Weather Risk Dr. Boissonnade was the original architect of the RMS hurricane catastrophe models and has over 20 professional experience in structural analysis and design, natural hazard modeling, and risk assessment of natural hazards in the US, Europe, Africa, and Asia. His expertise includes developing risk assessment models for natural hazards (earthquakes, extreme winds, floods and other weather phenomena) for applications in risk assessment of critical facilities and insurance exposures. Dr. Boissonnade has a B.S. from Ecole Superieure des Travaux Publics (France) and a Ph.D. from Stanford University where he has been a Consulting Professor. While at Stanford, Dr. Boissonnade performed research on damage estimation with application to the insurance industry. Prior to joining RMS, Auguste was a project leader at Lawrence Livermore National Laboratory with responsibilities for developing probabilistic seismic hazard guidelines for the U.S. Nuclear Regulatory Commission and guidelines on natural phenomena hazards for the Department of Energy. He is a member of the American Meteorological Society and the American Society of Civil Engineers and a reviewer for the National Science Foundation. Dr. Boissonnade has authored more than 50 publications, including one book. Hurricane Project Responsibilities: (1) Review of overall data generated for use in stochastic simulation; (2) Windfield definition/ degradation curves/roughness/vulnerability curves; (3) Historical and stochastic loss calibration; and (4) Advisor on science and technical issues.

Anders Brix, Ph.D., Principal Modeler Anders is a Senior Modeler based in RMS’ London office, with responsibility for researching and implementing advanced modeling techniques. Prior to joining RMS, he developed pricing models and conducted dynamic financial modeling as a statistician in the Instrat actuarial services unit of reinsurance broker Guy Carpenter. Anders received a Ph.D. in Mathematical Statistics from the Royal Veterinary and Agricultural University in Denmark and has conducted post-doctoral research in statistics at several universities throughout Europe. He received a Cand. Scient. degree in statistics from the University of Copenhagen. Hurricane Project Responsibilities: Review of model output and sensitivity analyses from a statistical viewpoint.

David Carttar, Lead Engineer Mr. Carttar has B.S. degrees in Geography and Architectural Studies from the University of Kansas, and a Master of City Planning degree from the University of California at Berkeley. For RMS, Mr. Carttar coordinates geocoding and mapping applications for the company's core technology. Mr. Carttar's experience revolves around the application of geographic modeling at a variety of technical levels. Hurricane Project Responsibilities: Updating geocoding capabilities for all hurricane states.

Han Yao Chen, Ph.D., Senior Software Engineer Dr. Chen graduated with a M.S. in Civil Engineering in Xian and with Ph.D. in Seismology at the Institute of Geophysics in Beijing. He is a member of the Research and Development Division of Engineering at RMS. He leads projects on loss estimation caused by fire following earthquake and earthquake ground motion estimation. His areas of research have been concentrated primarily on seismic design of structures, seismic hazard analysis, and seismic loss estimation. A scholar and engineer, Dr. Chen has written over twenty technical papers in the

Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 83 MODULE 2: Background/Professionalism area of Engineering Seismology. He has more than ten year’s experience in catastrophe risk modeling and probabilistic modeling. Hurricane Project Responsibilities: Software implementation.

Richard Dixon, Ph.D., Senior Modeler Dr. Richard Dixon joined RMS in January 2001 to undertake studies on the role of the jet- stream, in affecting the formation of severe windstorms. Having raised the public profile of the jet in generating catastrophic windstorms in Europe, he has most recently looked across the Atlantic to lead the meteorological work to understand the structure and statistics of transitioning hurricanes; again controlled by the jet-stream. Dr. Dixon has a first-class Honors degree in Meteorology and a Ph.D. from the University of Reading, concerning the processes involved in the development of intense extra-tropical cyclone windstorms. Hurricane Project Responsibilities: Lead researcher in the area of transitioning storms and activity rates, and the impact of transition on hurricane structure and windfields.

Michael Drayton, Ph.D., Head of Climate Hazards Practice Dr. Drayton holds a Ph.D. in Applied Mathematics from the University of Cambridge and a 1st class honors degree in Civil Engineering from New Zealand. Dr. Drayton is primarily involved in the research and development of hazard models. Since joining the RMS London office in early 1996 he has worked on the European windstorm model, the Atlantic hurricane models and the UK flood project. He has extensive experience of insurance-related hazard modeling and has also worked as a researcher investigating river flooding and pollution dispersion in the environment. Hurricane Project Responsibilities: Development of the stochastic basin-wide event set model.

Weimin Dong, Ph.D., Chief Risk Officer Dr. Dong is a co-founder of RMS . He has over 30 years of industrial, teaching, and research experience specializing in seismic hazard evaluation and insurance and financial risk assessment. He is the chief architect of RMS’ catastrophe models, and has overseen the company’s R&D efforts since its inception. Weimin is currently focusing his efforts on further developing the P&C RAROC methodologies, including the RAROC ASP development and various optimization routines. Prior to founding RMS, Dr. Dong served as the Director of Earthquake Research for the General Research Institute, Ministry of Machine Building in China. Dr. Dong received his Ph.D. from Stanford University, and his Master of Engineering Mechanics from Shanghai Jiao Tong University. During his career, he has published books, technical reports, and over 100 papers. Hurricane Project Responsibilities: Advisor on science and technical issues.

Uday Eyunni, Software Engineer Mr. Eyunni graduated with M.S. in Computer Science from the University of Alabama at Birmingham. Mr. Eyunni joined RMS in 1994. Since then, Mr. Eyunni has worked on various software products. At RMS, Mr. Eyunni's primary role is to design and develop software for RiskLink and RiskOnline products. Mr. Eyunni has published research papers on parallel computing and compilers. Hurricane Project Responsibilities: Software design and implementation.

Surya Gunturi, Ph.D., Director Dr. Gunturi holds B.S. and M.S. degrees in Civil Engineering from the Indian Institute of Technology in Madras, India. He earned the Standing First Rank in his master’s program. He

Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 84 MODULE 2: Background/Professionalism holds a Ph.D. in Civil Engineering from Stanford University. He was honored with a fellowship to the University of Stuttgart where he worked on non-linear dynamic analysis of structures. Dr. Gunturi has over twenty years experience as a researcher and project manager. At RMS, he heads the Wind Hazard Modeling group, investigating world wide wind hazards and developing analytical methods to predict windfield patterns, surge flooding, and the impact of extreme wind conditions. He has extensive working knowledge of computer expert systems. Dr. Gunturi has published over 30 technical papers on the area of structural engineering analysis and design and is a member of the American Society of Civil Engineers. Hurricane Project Responsibilities: Hurricane model implementation.

He-Jung Kim, Lead Risk Quantification Researcher He-Jung is a Lead Risk Quantification Researcher with RMS. She is a member of the Actuarial & Financial Modeling team which overseas the research and development of the financial model component of RMS' catastrophe modeling software. She has over 10 years of experience in the P&C insurance industry in the areas of enterprise risk management, reinsurance optimization, catastrophe modeling, insurance pricing and reserving. Prior to joining RMS, she was a project manager with Milliman USA's P&C actuarial consulting practice where she performed loss reserve analyses and property catastrophe rate analyses. She held various other positions in commercial reserving and personal lines pricing at Fireman’s Fund and Farmers Insurance. He- Jung has a B.S. in Math/Applied Science from the University of California, Los Angeles. Hurricane Project Responsibilities: (1) Assessing uncertainty on model generated losses and assigning confidence levels; (2) sensitivity and uncertainty analyses.

Atul C. Khanduri, Ph.D., Program and U.S. Hurricane Model Project Manager Since joining RMS in 1997, Dr. Khanduri has played a key role in developing hurricane vulnerability models as well as for researching, consolidating and maintaining all vulnerability and inventory parameters related to wind risk models. Experienced in hurricane reconnaissance surveys, he is involved in developing mitigation models and strategies for dealing with natural hazards. More recently, Dr. Khanduri has been involved in managing and coordinating model development projects in the Global Risk Modeling Business Unit. Dr. Khanduri holds B.Eng. and M.Eng. degrees in Civil Engineering from the University of Roorkee (India) and a Ph.D. from the Center for Building Studies, Concordia University (Canada). While in Canada, on a Commonwealth Scholarship, Dr. Khanduri performed research on wind effects on buildings, using experimental and computerized modeling methods and on the application of Artificial Intelligence techniques to civil engineering. Dr. Khanduri has a broad-based experience of over 14 years in civil engineering design, research, teaching and risk assessment. He has numerous publications in technical journals and conferences and holds memberships of the American Society of Civil Engineers, Canadian Society for Civil Engineering and the American Association of Wind Engineering. Hurricane Project Responsibilities: Development of hurricane vulnerability models and improving existing models as well as researching, consolidating and maintaining all vulnerability and inventory parameters related to wind risk models. Overall U.S. Hurricane model project manager responsibility.

Craig Miller, Ph.D., Lead Engineer2

2 Consultant to RMS since November 2002 Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 85 MODULE 2: Background/Professionalism

Dr. Miller joined RMS in September 1997. During his time at RMS Dr. Miller was primarily responsible for the development of surface wind field models for the modeling of risk due to both tropical and extra-tropical cyclones. This included the characterization of the effects of changes in the surface roughness and wind speed averaging times, as well as the effects of topography on surface wind speeds, both modeled and observed. Dr. Miller was also involved in post storm damage surveys following Hurricane Georges in Puerto Rico in 1998, and wind storm Anatol in Denmark in 1999. Prior to joining RMS Dr. Miller worked as a Research Fellow at the Building Research Establishment in England on a project examining the exposure of UK Meteorological Office anemograph sites, and the resulting impact on design wind speeds for the United Kingdom. Dr. Miller holds B.E.(Hons) and M.E. degrees in Mechanical Engineering from the University of Auckland, New Zealand, and a Ph.D. in Engineering Science from the University of Western Ontario, Canada. He is a member of the Wind Engineering Society, the Royal Meteorological Society, and the American Meteorological Society. Since leaving RMS in November 2002 to take up a faculty position associated with the Alan G. Davenport Wind Engineering Group in the Department of Civil and Environmental Engineering at the University of Western Ontario, Canada, Dr. Miller has acted as a consultant for RMS. Hurricane Project Responsibilities: Development of windfield models for the assessment of risk due and development of modeled effects including the effects of ground roughness changes and topography on the windfield structure.

Gilbert Molas, Ph.D., Lead Engineer Dr. Molas’ primary technical duties are to develop earthquake and windstorm stochastic models. He is also actively involved in several technical aspects of RMS’s worldwide risk models including calibration, validation, and product implementation. He has been a major contributor to the development of earthquake and windstorm models for the United States and Japan, including securitization projects for these models. Dr. Molas graduated Cum Laude from the University of the Philippines, with a B.S. in Civil Engineering. He received his M.S. and Ph.D. in Civil Engineering from the University of Tokyo in 1995. While in Japan on a Monbusho Scholarship, Dr. Molas worked on Earthquake Engineering and Disaster Mitigation research and developed new earthquake ground motion attenuation relations, and damage estimation techniques using artificial intelligence (Neural Networks). Prior to joining RMS, Dr. Molas was a member of the faculty at the Department of Civil Engineering, University of the Philippines, teaching structural analysis and design, and probability and statistics. He has worked on catastrophe risk model development for more than ten years. Hurricane Project Responsibilities: (1) Advisor on science and technical issues; and (2) Convergence studies.

Guy Morrow, Head of Science and Engineering, Principal Engineer Mr. Morrow holds a B.S. in Civil Engineering from the University of Illinois and an M.S. in Structural Engineering from the University of California in Berkeley. He is a registered Civil and Structural Engineer in the State of California. Mr. Morrow has over ten years of experience in the field of seismic analysis, structural design and risk assessment. Prior to joining RMS, Mr. Morrow was an associate in the structural engineering firm Degenkolb Engineers in San Francisco. Since joining RMS in 1994, Mr. Morrow has performed risk assessments of major commercial and manufacturing facilities located throughout the world. Most recently, he has been involved in the development and support of RMS’ Risk Models. Hurricane Project Responsibilities: Advisor on science and technical issues. Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 86 MODULE 2: Background/Professionalism

Chris Mortgat, Ph.D., Vice President, Principal Scientist Dr. Mortgat received his Ph.D. in Civil Engineering, an Engineer’s Degree in Geotechnical Engineering, and an M.S. in Structural Engineering from Stanford University; and has a B.S. in Civil Engineering from Tennessee Technological University. Dr. Mortgat has a broad background in earthquake engineering that ranges from structural analysis for buildings and earth dams to the development of seismic hazard maps. Dr. Mortgat has developed a unique Bayesian risk analysis methodology and has studied earthquake response spectrum shapes and their attenuation. He has directed or participated in major seismic risk analysis projects for Costa Rica, Nicaragua, Alaska, and Algeria. Following the 1980 Algerian earthquake, he participated as a member of the Stanford University research team and the Earthquake Engineering Research Institute’s reconnaissance team in Algeria. He has published numerous articles and reports in these areas. Dr. Mortgat has been responsible for civil/structural design review at several nuclear power plants in areas such as procedure and criteria review, structural dynamics modeling, steel and concrete design and design of suspended commodities. Recently, Dr. Mortgat has been involved in the severe accident assessments of advanced light water reactor designs. He has more than 25 years experience in catastrophe risk modeling. Since joining RMS in 1995, Dr. Mortgat has been the head of the Risk Engineering department and has been involved in all significant engineering efforts undertaken by RMS. Hurricane Project Responsibilities: Advisor on science and technical issues.

Robert Muir-Wood, Ph.D., Managing Director, Global Risk Modeling Dr. Robert Muir-Wood is Managing Director of the Global Risk Modeling group within RMS. He has developed probabilistic catastrophe models for earthquake, tropical cyclone, volcano, river flood, and storm surge in Japan, Australia, the Caribbean, and the U.K. Most recently he has led the project to build a new scientific foundation for European windstorm loss modeling. He has published 40 scientific papers, written more than 100 articles and reviews, lectured to audiences from the Soviet Ministry of Atomic Energy to the Royal Geographical Society Christmas Lecture, run courses on catastrophe risk for Lloyds of London and is the founding editor of the European Journal of Geo-sciences: Terra Nova. He has also published six books, and has been active in his field for more than 20 years. Hurricane Project Responsibilities: Advisor on science and technical issues.

Brian Owens, Director , Technical Marketing Mr. Owens joined RMS in October of 2001 and is responsible for managing the technical interface between RMS clients/client relationship managers and the model development team, particularly with respect to U.S. hurricanes. Prior to joining RMS, Mr. Owens worked for four years at Andersen Consulting, after which he worked at Chase Securities Inc., focusing on corporate finance transactions for the insurance and reinsurance industries, including catastrophe financing structures. Mr. Owens holds a B.S. in Computer Science from the National University of Ireland, an MBA in Finance from the Wharton Graduate School of Business, and an M.S. in Atmospheric Science from the Rosenstiel School of Marine and Atmospheric Science at the University of Miami, where he researched hurricane climatology, seasonal forecasts and tropical cyclone tracks in the North Atlantic basin. Hurricane Project Responsibilities: Windfield modeling, U.S. Hurricane model technical marketing, project manager for Florida Commission on Hurricane Loss Projection Methodology submission.

John Reed, Managing Director, Catastrophe Applications Mr. Reed joined RMS in 1993 as IRAS Product Manager. Before taking his current position, he managed a number of projects in both the Product Development and Quality Assurance departments. Prior to joining RMS he was Director of Development / Operations Manager for Greenleaf Medical Systems, as well as a development manager and an international software marketing liaison for Hewlett Packard. Mr. Reed has a B.S. in Computer Science and an MBA, both from the University of Michigan. He also has an MS in Medical Informatics from Stanford University’s Medical School. A long-standing member of the Healthcare Information Management Systems Society and the American Medical Informatics Association, Mr. Reed has written and presented papers on healthcare technology management and is active in both organizations. Hurricane Project Responsibilities: Software implementation, testing and quality assurance, reliance management.

John Reiter, Vice President, RiskLink Product Development Mr. Reiter has over seventeen years of experience in developing commercial software tools for the analysis of insurance and other financial risk. Mr. Reiter has a B.S. in Mathematics and Computer Science from the University of Illinois at Urbana-Champaign and an M.S. in Computer Science from the same university. Prior to joining RMS in 1994, Mr. Reiter worked for over ten years as a software developer at Syntelligence, Inc., building systems that provide underwriting advice to the property and casualty insurance industry and loan risk analysis for the banking industry. At RMS, Mr. Reiter’s primary role is manager of all software development for the RiskLink and RiskOnline products. Mr. Reiter is a member of the Association for Computing Machinery and has authored several software-related publications. Hurricane Project Responsibilities: Software design and implementation.

Hemant Shah, President and CEO Hemant Shah is co-founder, President and CEO of Risk Management Solutions, Inc. Mr. Shah received his B.S. and M.S. in Civil Engineering from Stanford University. Prior to becoming president of the company, he held the posts of Senior Vice President of Marketing and Strategic Planning and Director of RMS Europe. In the past 11 years, he has played a central role in the change that has taken place within the insurance and reinsurance industries in their adoption of technologies and risk-based strategies to manage catastrophe exposure. He has advised hundreds of insurers, reinsurers, and financial institutions and is a regular participant in industry initiatives, conferences, and trade press. Hurricane Project Responsibilities: Advisor on science and technical issues.

Mohan P. Sharma, Ph.D., Lead Engineer Dr. Sharma has over 15 years professional experience in teaching, structural analysis and design, natural hazard modeling and catastrophe modeling. Dr. Sharma has a B. Tech. from the Indian Institute of Technology, New Delhi, India and a M.S. and Ph.D. from Stanford. He has taught undergraduate and graduate courses at the Institute of Engineering, Kathmandu, Nepal and Santa Clara University, Santa Clara, CA. At RMS Dr. Sharma has led teams in the development of hazard and vulnerability models for hurricanes, tornado and hail, and extratropical storms. Hurricane Project Responsibilities: Researched hurricane parameters of historical database, developed methodology for statistical comparisons of historical and stochastic storm sets, and

Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 88 MODULE 2: Background/Professionalism working on procedures to enhance the numerical performance of the model. Responsible for the development of the revised storm surge module of U.S. Hurricane model.

Chessy Q. Si, Software Engineer Ms. Si holds a B.S. Degree in Economic Geography and Urban Planning from Beijing University and a Post-Graduate Diploma in Geographic Information Systems (GIS) from the Institute for Housing Studies, the Netherlands. She received her MA in GIS and MRP in Regional Planning from State University of New York, Albany. Prior to joining RMS, she practiced urban planning for 5 years and worked as a GIS Specialist with various public and private agencies. Ms. Si has 10 years experience with GIS application, spatial data analysis and digital cartography. She is currently involved in several RMS’ projects and is responsible for RMS’ spatial data warehouse. Hurricane Project Responsibilities: GIS software implementation.

Claire Souch, Ph.D., Senior Catastrophe Analyst Dr. Souch holds a B.Sc. in Environmental Science and a Ph.D. in Water Resource Management from Cranfield University in the U.K. Since joining RMS in October 2000, she has had responsibility for real-time hurricane monitoring and event response, using the RMS Hurricane model to pick storm tracks and produce industry loss estimates. In addition, she has extensive GIS analysis experience. Hurricane model responsibilities: coordination of material for this year’s submission to the Florida Commission on Hurricane Loss Projection Methodology.

Pane Stojanovski, Ph.D., Vice President, Chief of Development Operations Dr. Stojanovski holds MS and Ph.D. degrees from the University of Skopje, Macedonia. He has over twenty years of, research, practicing, and teaching experience in the field of earthquake and structural engineering, catastrophe loss modeling, and development of natural catastrophe loss estimation models. Before joining RMS he was professor at the Skopje University, Macedonia. Dr. Stojanovski was also a visiting Fulbright scholar / professor at the Blume Earthquake Engineering Center at Stanford University. Dr. Stojanovski is in charge of the model development operations of the Global Risk Modeling business unit of RMS. He also oversees te implementation and productization of all catastrophe models developed in GRM. urricane Project Responsibilities: Operational oversight and resource utilization for the preparation of the submittal to the FCHLPM.

B. Describe the credentials of the individuals or groups involved in the development of the following aspects of the model: 1. Meteorology 2. Vulnerability 3. Actuarial 4. Computer Science 5. Statistics State whether these persons are full-time employees or outside consultants.

Individuals Involved in Hurricane Model Development for All Model Versions (See section 2.A. above for credentials of individuals and more complete description of hurricane model development activities)

Table 2.2.1 Individuals Involved in Meteorological Aspects of the Model

Staff (S)/ Name Credentials Consultant (C) Dr. Fouad Bendimerad, P.E. Ph.D., Civil Engineering S Stanford University Dr. Auguste Boissonnade Ph.D., Civil Engineering and Meteorology S Stanford University Dr. Richard Dixon Ph.D. , Meteorology S University of Reading Dr. Michael Drayton Ph.D., Applied Mathematics S Cambridge University Dr. Surya Gunturi Ph.D., Civil Engineering. S Stanford University Dr. Craig Miller Ph.D., Wind Engineering S,C University of Western Ontario Dr. Chris Mortgat Ph.D, Civil and Geotechnical Engineering S Stanford University Dr. Robert Muir-Wood Ph.D., Earth Sciences S Cambridge University Mr. Brian Owens M.S., Tropical Meteorology S University of Miami Dr. Claire Souch Ph.D., Water Resource Management S Cranfield University Mr. Hemant Shah M.S., Civil Engineering S Stanford University Dr. Mohan Sharma Ph.D., Civil Engineering S Stanford University Dr. Alan Davenport Director, BLWTL, U. of Western Ontario, C Canada Mr. Charles Neumann Former Director of Research, U.S. C National Hurricane Center Dr. Dave Surry BLWTL, University of Western Ontario, C Canada (previous version of model)

Table 2.2.2 Individuals Involved in Vulnerability Aspects of the Model

Staff (S)/ Name Credentials Consultant (C) Dr. Fouad Bendimerad, P.E. Ph.D., Civil Engineering S Stanford University Dr. Auguste Boissonnade Ph.D., Civil Engineering and Meteorology S Stanford University Dr. Surya Gunturi Ph.D., Civil Engineering. S Stanford University Dr. Atul Khanduri Ph.D., Civil Engineering S Concordia University Mr. Guy Morrow, S.E. M.S., Structural Engineering S University of California, Berkeley Dr. Chris Mortgat Ph.D., Civil and Geotechnical Engineering S Stanford University Dr. Dale Perry Professor C Texas A & M University Dr. Timothy Reinhold Professor C Clemson University Dr. Peter Sparks Professor C Clemson University Dr. Pane Stojanovski Ph. D, Structural Engineering S University of Skopje, Macedonia Dr. Norris Stubbs Professor C Texas A & M University

Table 2.2.3 Individuals Involved in Actuarial Aspects of the Model

Table 2.2.4 Individuals Involved in Computer Science Aspects of the Model

Staff (S)/ Name Credentials Consultant (C) Mr. John Reiter M.S., University of Illinois S Mr. Uday Eyunnai M.S., University of Alabama S

Table 2.2.5 Individuals Involved in Statistical Aspects of the Model

Staff (S)/ Name Credentials Consultant (C) Mr. Richard Anderson Fellow, Casualty Actuarial Society S Ms. He-Jung Kim B.S., Mathematics/ Applied Science S U.C.L.A. Dr. Auguste Boissonnade Ph.D., Stanford University S Dr. Anders Brix Ph.D. Royal Veterinary and Agricultural S University, Denmark Dr. Han Yao Chen Ph.D., Institute of Geophysics S Dr. Chris Mortgat Ph.D., Stanford University S Dr. Weimin Dong Ph.D., Stanford University S Dr. Surya Gunturi Ph.D., Stanford University S Dr. Mohan Sharma Ph.D., Stanford University S

3. Multi-discipline Team

A. Indicate the different academic disciplines used to provide input and to construct the model.

Meteorology Staff: Dr. Auguste Boissonnade Dr. Richard Dixon Dr. Michael Drayton Dr. Steven Jewson Dr. Craig Miller Dr. Robert Muir-Wood Mr. Brian Owens Dr. Mohan Sharma Dr. Claire Souch

Consultants: Dr. Alan Davenport Mr. Chip Guard Mr. Charles Neumann Dr. Robert Sheets

Engineering Staff: Dr. Fouad Bendimerad Dr. Auguste Boissonnade Dr. Surya Gunturi Dr. Atul Khanduri Dr. Chris Mortgat

Mr. Guy Morrow Dr. Pane Stojanowski

Consultants: Dr. Dale Perry Dr. Timothy Reinhold Dr. Norris Stubbs

Actuarial Staff: Mr. Richard Anderson Ms. He-Jung Kim Mr. James Gant

Statistics Staff: Mr. Richard Anderson Ms. He-Jung Kim Dr. Auguste Boissonnade Dr. Anders Brix Dr. Han Yao Chen Dr. Weimin Dong Mr. James Gant Dr. Surya Gunturi Dr. Chris Mortgat Dr. Mohan Sharma

Economics Staff: Mr. Richard Anderson Mr. Weimin Dong Mr. Peter Ulrich

B. Of the disciplines listed above, which are represented by current employees with the company? Are other disciplines represented through consulting arrangements?

See 3.A.

C. Provide visual business workflow documentation connecting all personnel related to model design, testing, execution, and maintenance.

Figure 2.3.1 shows a typical workflow diagram used at RMS. Specific documentation related to model design, testing, execution, and maintenance is available for on-site review by the Professional Team.

Model Development

Develops Model

Engineering & Software

Engineering & Software Specs

QA & Engineering Software Engineering

QA Writes Test Plan Software Writes Code Engineering Creates Data Files

Testing Maintenance Software, Engineering, QA User

Software is Tested Problem Reported Using QA Plan

QA QA &Engineering

QA Plans all Pass Tests Problem Verified Fail

Software & Engineering Software / Files Released Solution Identified

Figure 2.3.1 Business Workflow Diagram

4. List of Clients

A. Provide a sample list of the company’s clients in the following categories: for ratemaking, for reinsurance and capital markets, for government. Regarding the ratemaking clients, state the number of clients in this category and the total residential market share, in Florida and nationwide, represented by these clients. For ratemaking clients, how many clients have a U.S. aggregate annual property and casualty insurance premium of $100 million or more? Do any of the ratemaking clients have a U.S. aggregate annual property and casualty insurance premium of over $1 billion? (The Commission may contact these persons or firms to discuss the company’s work.)

Table 2.3.1 Sample List of Clients

Ratemaking Clients Reinsurance Clients Government/Organization AIG American Re California Dept. of Insurance Allstate AXA Corporate Solutions Florida Hurricane Cat. Fund Chubb Employers Re Hawaii Insurance Division CAN General Re Farmers Renaissance Re Fireman’s Fund SCOR Re ITTThe Hartford Sorema N.A. St. Paul Swiss Re America State Farm ACE Tempest Re Travelers Western Re USAA Zurich

According to the latest information available, RMS has fifty-eight current ratemaking clients, which comprise an estimated 64% of the U.S. market and 69% of the Florida market. Thirty- nine clients have an aggregate annual premium of $100 million and one client has an aggregate annual premium of over $5 billion.

B. Describe the present mix of company clients (ratemaking, reinsurance, capital markets, government, etc.) and whether (and, if so, how) that mix differs from the mix over the last 3 to 5 years.

Table 2.3.2 Mix of Company Clients Over the Last 5 Years

Market Segment 1997 1999 2000 2001 2002 Primary Insurers 48% 42% 46% 43% 44% Reinsurers 19% 29% 36% 27% 31% Reinsurance Intermediaries 13% 11% 11% 5% 14% Financial Institutions 5% 1% 2% 3% 1% Brokers/Risk Managers/Engineering 6% 1% 2% 7% 6% Government/Other 8% 16% 3% 16% 4%

C. How long have the ratemaking clients been clients of the company? Table 2.3.3 Time Since Ratemaking Clients Became Clients

Ratemaking Client Client Since AIG 1993 Allstate 1995 Chubb 1992 CNA 1995 Farmers 1995 Fireman’s Fund 1996 ISO 1993 St. Paul 1992 State Farm 1995 Travelers 1993 USAA 1994 Zurich 1994

D. Provide the loss date of the insurance company data available for validation and verification of the model.

The year of the loss is given below:

Georges - 1998 Fran - 1996 Opal - 1995 Erin - 1995 Andrew - 1992 Bob - 1991 Hugo - 1989

5. Independent Expert Review

A. What independent peer reviews have been performed on the following parts of the model:

1. Meteorology 2. Vulnerability 3. Actuarial 4. Computer Science 5. Statistics

In addition to the extensive testing that RMS has itself performed on its current version of the RMS Hurricane model, and in addition to the many contributions by the outside experts listed above whose names and reputations rest upon the quality of their work, an overall review of the 1997 released version of the Hurricane Model was conducted by Dr. Robert Sheets, former director of the National Hurricane Center. A letter outlining his findings is attached in Attachment A. Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 96 MODULE 2: Background/Professionalism

ISO, a national industry group, has also reviewed the 1997 released version of the hurricane model. ISO elected to utilize RMS technology with the intention of using this technology as the basis for their loss costs filings in hurricane-prone states.

The current version of RMS’ Hurricane Model builds upon the strengths of its former models; we therefore include the following discussion of the reviews conducted on the original RMS Hurricane Model to illustrate the consistent and comprehensive approach that RMS takes to validate and substantiate its models.

Dr. Robert Simpson and Mr. Glenn Meyers reviewed the original version of the RMS Hurricane Model and did not receive any compensation from RMS. Other experts, such as Dr. Alan Davenport and Dr. Dale Perry, have worked in close association with RMS for several years and made extensive technical contributions to RMS Hurricane Model.

In 1993, the RMS hurricane model was selected by ISO to be the methodology upon which it would file revised catastrophe procedures in the calculation of property loss costs. Prior to this decision, the model was carefully examined and a validation procedure was performed comparing the model output to ISO losses for specific storms by a team of 10 members of the ISO actuarial staff over a six month period. Highlights of the validation efforts of RMS engineers, ISO, and RMS clients include:

Convergence. The statistical "completeness" of the stochastic database was tested, and was found to represent the range of potential storm occurrences.

Rate of occurrence. The modeled frequency of storm occurrences was compared to the historical record, and was found to closely replicate the historical rate of occurrence.

State-of-the art. The hurricane wind-field model was compared to the state-of the-art methodologies developed and utilized by the engineering community for the estimation of wind speeds for the purpose of hazard analyses of critical facilities. The evaluation concluded that the RMS approach was as well-founded as such methodologies.

Meteorological review. ISO retained Dr. Robert Simpson, the co-developer of the Saffir/Simpson scale and former Director of the National Hurricane Center, to perform an independent review of the hurricane model. He concludes his assessment by stating: “IRAS is an interactive expert system which can provide a broad and probably unparalleled base of information for insurance decision analysis. From a physical viewpoint, the model as a follow- on to similar stochastic purposes should provide the most comprehensive assessment of damage potential available, with discrimination over smaller scale areas than heretofore available.”

A copy of Dr. Simpson's full assessment report is attached in Attachment B.

The following experts were hired by RMS to contribute during key stages of the current RMS Hurricane model design and development:

Mr. Charles J. Neumann, a meteorologist who compiled the North Atlantic Basin Storm database (known as HURDAT), did a private review and update of the HURDAT database for

RMS, using knowledge and information not available to him or not used by the team at the time he originally compiled the database at the NHC.

Dr. Robert Sheets, the recently retired Director of the National Hurricane Center, gave input on the model overall and on the stochastic storm development in particular.

Dr. Dale Perry, of Texas A & M University, is was a consultant to the U.S. Corps of Engineers, the Applied Technology Council, the U.S. Department of Justice, the Southern Building Code Congress, and many other engineering firms regarding programs related to wind design.

Dr. Tim Reinhold, of Clemson University gave substantial input to the windfield modeling and vulnerability portions of the model.

B. Provide documentation of independent peer reviews of both the standards and modules and clearly identify any unresolved or outstanding issues as a result of these reviews.

Copies of Dr. Robert Sheets’ and Dr. Robert Simpson's assessment reports are attached in Attachments A and B.

C. Describe the nature of any on-going or functional relationship the company has with any of the persons performing the independent peer reviews. State which of the peer reviews described above were paid for by the company and which were performed for no compensation. Describe any review by an independent organization, such as Standard and Poor’s, Moody’s, etc.

There currently is no on-going or functional relationship with the reviewers. In 1993, the RMS hurricane model was selected by ISO to be the methodology upon which it would file revised catastrophe procedures in the calculation of property loss costs. Prior to this decision, the model was carefully examined and a validation procedure was performed comparing the model output to ISO losses for specific storms by a team of 10 members of the ISO actuarial staff over a six- month period.

D. Discuss any adversarial situations (such as a ratemaking hearing) in which the model was subjected to review.

RMS has collaborated with several DOI’s (CT, FL, HI, LA, NY) in the context of hurricane rate making. These relationships have not been adversarial.

MODULE 3

The following pages contain questions and follow-up tests of the computer simulation model. Answer each question with as much detail as possible. Answers that do not address the question directly may not help the Commission make the appropriate decisions regarding the model.

The written response and output files must be submitted to the Commission as specified in the Process for Determining the Acceptability of a Computer Simulation Model.

NOTE: The Commission is charged with evaluating the model as a ratemaking tool. Thus, modelers should focus on this charge while explaining the model.

Module 3 - Section I

Meteorology - Hurricane Set

1. Define an “event” in the model. Does it include only hurricanes making landfall (i.e., the eye of the hurricane crosses land) or does it also include any hurricane where hurricane force winds cause damage (i.e., the eye need not necessarily cross land)?

An event is defined as a tropical cyclone modeled throughout hurricane that during its lifetime in the North Atlantic basin is at least a Category 1 as per the Saffir Simpson scale. An event need not necessarily make landfall in the U.S. to cause loss for inclusion in the historical database and therefore the stochastic database contains both landfalling and by-passing storms.

2. What is the upper limit of wind speeds (maximum one-minute average wind at 10 meters height) per hurricane category (defined by the Saffir-Simpson scale wind speed) that the model produces?

As explained in Section II.B of Module 1, maximum wind speeds are a function of several parameters. Some of these parameters, such as translational speed and radius to maximum wind, are not explicitly included in the definition of the Saffir-Simpson Scale. The modeled maximum wind speeds per hurricane category are consistent with the Saffir-Simpson Scale listed in the table below in Table 3.I.1. Table 3.I.1 Saffir-Simpson Hurricane Scale (for displayed parameters)

Wind Speed Central Pressure Category (mph) (mb) 1 74 - 95 > 980 2 96 - 110 965 – 979 3 111 - 130 945 – 964 4 131 - 155 920 – 944 5 > 155 < 920

3. How does the model handle events with multiple landfalls? Are these defined as a single event or multiple events? How does this affect frequency assumptions used in the model?

As discussed in Module 1, Section I.A, Sub-section 2.2.1, the RMS hurricane model uses a random-walk technique to generate a set of stochastic storm events covering the entire Atlantic Basin. Each event is defined throughout the life of the storm allowing for the possibility of multiple landfalls.

The track model is calibrated across the Atlantic basin by comparing the rates of storms crossing a grid of cells covering the basin. More detailed calibration is performed at the U.S. coastline by calculating the rate of crossing on linear gates along the coastlines.

In developing frequencies of historical storms, gates that are roughly 50 nautical miles long are established along the coast. Historical storm landfall frequencies are then established for each gate using a spatial smoothing procedure across gates that are aligned parallel to the gate of interest. The landfall frequencies studied are first landfall, multiple landfall and exiting. The track model is calibrated against those rates by comparing the historical values to their respective values for the stochastic storm set to ensure that all aspects of the historical track landfall characteristics are preserved by the stochastic storm set.

4. How does the model handle the definition of an event from an insurance policy perspective? That is, does the model recognize the 72-hour limitation for an occurrence as defined by some insurance policies? From this perspective, are events with multiple landfalls greater than 72 hours apart considered as two events?

The basin-wide stochastic track set enables the model to consider multiple landfalls resulting from the same event. Losses from all landfalls resulting from such an event are calculated. As the model does not explicitly present the losses attributable to each individual landfall, no consideration is given to the time period separating landfalls from an insurance perspective, although the losses from each landfall are readily calculable from an analysis of the affected regions.

5. Describe the hurricane tracks in the model. Discuss the appropriateness of the hurricane tracks used by the model. What historical data are used as the basis for the model’s hurricane tracks?

Each model hurricane track is defined by the positions (longitude and latitude) and intensities of the event at regular time intervals throughout its life. All model hurricane tracks are generated by the random-walk model. This model creates physically realistic tracks which preserve the statistical behavior (means and variances of translational speeds and directions) of the historical events in each cell across the basin. The advantage of the random-walk method is that it produces a greater variety of track behavior than is seen in the historical record but ensures that the behavior of the simulated tracks is consistent with the variance of the historical record. By checking various historical storm track characteristics with those of the stochastic storms it is ensured that the regional behaviors (direction, curvature) of the storms have been preserved. The reference source for the RMS database of historical tracks is the North Atlantic Basin Storm database known as HURDAT. RMS worked with Mr. Charles Neumann, a former meteorologist Information submitted to the Florida Commission on Hurricane Loss Projection Methodology, February 2003 © 2003 Risk Management Solutions, Inc 100 MODULE 3: Meteorology - Hurricane Set from the National Hurricane Center to audit and improve the quality of HURDAT. For more details refer to Module 1, Section I.A., Sub-section 2.2.1.

6. Describe in detail the decay rates or hurricane degradation assumptions used by the model after the hurricane makes landfall. How far inland are hurricane force winds estimated for different category events (as defined by wind speed in the Saffir-Simpson scale)? Does the decay rate vary by region or hurricane segment?

Pressure histories are added to the synthetic tracks using a second random-walk process. The rates of change of pressure along the synthetic tracks are defined through the mean and variance of pressure changes quantified from historical events. Storms tend to intensify faster over warm water than over cold water. Storms fill as they cross areas of land and may reintensify if they move back out over the water. The overland filling rates for storms landfalling on Florida are based on the research of Kaplan & DeMaria (1995) but have been adjusted to take into consideration the observed historical behavior of storms in the Florida region. Once a hurricane makes landfall, its filling rate is constant until either it dissipates or it emerges over water, where it may begin to reintensify. However, the filling rate may vary from one hurricane to another. There is very little variation in the model inputs across the Florida region.

Minimum pressures are constrained by theoretical arguments relating central pressure to Sea Surface Temperature (SST). The pressure history of each storm thus depends on the track of the storm as it crosses areas of different SST and encounters topography.

The pressure history model is calibrated by specifying the pressure pdf on cells across the basin and on linear segments around the coastline. The pressure history of each event is individually scaled so that the pressure pdf for the cell or segment is obtained. In this way the random-walk model defines realistic pressure histories and the calibration ensures the correct intensities of simulated storms.

A further check is performed on the filling rates of stochastic storms after landfall to compare the degradation rates against the model of Kaplan & DeMaria (1995). The model filling rates are found to agree very well with their model.

7. Provide a graphic representation of the modeled degradation rates for Florida storms over time compared to the Kaplan-DeMaria decay rate. Include curves for +/- 20% of the Kaplan-DeMaria values.

0.9 Stochastic 0.8 0.8 K&DM K&DM 0.7 1.2 K&DM

0.6

0.5 Decay 0.4

0.3

0.2

0.1

0 0 5 10 15 20 25 Time (Hours)

Figure 3.I.1 Modeled Degradation Rates Compared to the Kaplan-DeMaria Curve (includes the decay rate and the +/- 20% range)

8. Name the source of the historical data set used to develop frequency distributions for specific hurricane characteristics. How many years worth of data does the data set contain? Were any modifications made to the data set? If so, describe in detail the modifications and their appropriateness.

The historical data set used to develop frequency distributions for storm characteristics is the 1021 years of data from 1900 to 20010 taken from the tropical cyclone data tape for the North Atlantic Basin Storm data base, HURDAT, as mentioned above. The HURDAT database has been reviewed and modified based on storm parameter data provided by the Florida Commission on Hurricane Loss Projection Methodology, data bases used in the NOAA Technical Report NWS 38 (1900-1984), NWS 23 (1900-1978), NHC 31 (1900-1995) and a review by Charles Neumann of all intense storms affecting mainland United States by investigating documents available at the NHC/TCP center and contacting regional meteorological centers. The HURDAT database has been reviewed and modified as described in Module 1, Section I.A, Sub-section 2.2.1. The modified HURDAT database is the basis of the number of storms, storm tracks, and storm properties at 6 hour intervals. The information from all four sources listed above was reviewed in order to determine the best estimate of storm properties at landfall.

8. Provide ranges for radius of maximum winds, radius of hurricane force winds and far field pressure used by the model for the central pressures provided in Figure 1.

The ranges are shown below in Table 3.I.2.

Table 3.I.2 Hurricane Parameters

Central Radius of Maximum Radius of Hurricane Far Field Pressure (mb) Winds (mi) Force Winds (mi) Pressure (mb) 900 6-28 33-128 1013 910 7-35 31-143 1013 920 7-38 31-153 1013 930 7-43 27-136 1013 940 8-50 24-140 1013 950 8-54 23-141 1013 955 8-55 22-128 1013 960 8-62 21-128 1013 965 8-62 18-105 1013 970 9-70 16-113 1013 975 8-71 13-106 1013 980 9-72 10-88 1013 985 9-73 9-73 1013 990 9-72 9-72 1013

10. Provide maps showing the maximum winds at the zip code level for the modeled 102 year historical storm set and also for a 100 year return period from the stochastic storm set.

100 Year Return Period Windspeeds (mph) > 120 110 to 120 100 to 110 90 to 100 80 to 90 70 to 80 60 to 70

(a) Modeled Historical (b) Stochastic Note: Windspeeds include the impact of inland degradation and roughness effects. Figure 3.I.2 Modeled 102-year Historical and 100-year Stochastic One-Minute Windspeeds (mph) (thematic plots)

(a) Modeled Historical (b) Stochastic Figure 3.I.3 Modeled 102-year Historic and 100-year Stochastic One-Minute Windspeeds (mph) (contour plots)

11. Provide frequency and annual occurrence rates from both the historical data set given and the data set that the model generates by hurricane category (defined by wind speed in the Saffir-Simpson scale) for the entire state of Florida and selected regions as defined in Figure 2.

The frequency and annual occurrence rates from both the historical data set given and the data set that the model generates by hurricane category (defined by wind speed in the Saffir-Simpson scale) for the entire state of Florida and selected regions is shown in Table 3.I.3.

Historical 30 Entire State Modeled 25 20 15 10 5 0 12345

1.5 Northeast 1.0

0.5

16 0.0 Northwest 12 12345 8 10 4 Southeast 0 12345 5

0 10 12345 8 Southwest 6 3 By-Passing 4 2 2 0 1 12345 0 12345 Figure 3.I.4 Comparison of Historical and Modeled Rates (category assigned based on 1- minute windspeeds)

12. Complete the tables in Figure 3 with modeled information for Florida in total and by region as defined in Figure 2. For each region, the column labeled “Hurricanes” is the number of hurricanes that made their initial landfall in that region. Category refers to the Saffir-Simpson Hurricane Scale for the hurricane at that landfall. “Coastal X-ings” refers to the total number of crossings in that region, either entering land or exiting land, as long as the storm was a hurricane as it crossed. It is counted in the bin at the strength when it crossed the land/sea boundary. It includes the initial landfall and all subsequent landfalls as well as exits as long as it was still a hurricane on exit. List the number of events, the relative frequency (percent of the total) and annual occurrence rate (probability of an event in a given year) per hurricane category.

See Table 3.I.3

Table 3.I.3 Historical and Modeled Hurricane Characteristics for Florida

Entire State of Florida