INTRARISK Applied Research Associates, Inc.

February 25, 2002 Applied Research Associates, Inc. FloridaFlorida CommissionCommission onon HurricaneHurricane LossLoss ProjectionProjection MethodologyMethodology

ASCE 7-98 Wind Map Statewide Inspections and Mitigation Certifications

State of Florida Predicted 100 Year Return Period Peak Gust Hurricane Wind Speeds (mph) In Open Terrain

100 - 110

110 - 120

120 - 130

130 - 140

140 - 150

Hurricane Hazard Mitigation Grant Program for the Hawaii Hurricane Relief Fund Hurricane Erin, 1995

60 70

70 80

80 90

90 100

Eglin AFB B71 Pensacola LLWSAS (^ (^ ( 100 110 Hurlburt (^ Shading Represents Panama City Beach 10 mph Ranges of Peak Storm Track Gust Speeds in Open Terrain

Prepared for: Prepared by: FCHLPM Applied Research Associates, Inc. State Board of Administration 811 Spring Forest Road, Suite 100 1801 Hermitage Boulevard Raleigh, North Carolina 27609-9199 Tallahassee, Florida 32308

Copyright ©2002 Applied Research Associates, Inc. For Evaluation by the Florida Commission on Hurricane Loss Projection Methodology

TABLE OF CONTENTS

2001 Standards ...... 1 5.1 General Standards ...... 1 5.2. Meteorological Standards ...... 6 5.3 Vulnerability Standards...... 12 5.4 Actuarial Standards...... 13 5.5. Computer Standards...... 22 5.6 Statistical Standards ...... 25 Module 1...... 28 I General Description of the Model...... 28 A. In General...... 28 B. Loss Costs ...... 35 C. Other Considerations...... 36 II. Specific Description of the Model ...... 37 A. Model Variables...... 37 B. Methodology ...... 39 C. Validation Tests ...... 43 Module 2...... 49 1. Company Background...... 49 2. Professional Credentials...... 51 3. Multi-discipline Team...... 56 4. List of Clients...... 57 5. Independent Expert Review...... 57 Module 3...... 59 I. Meteorology – Hurricane Set...... 59 II. Hurricane Wind Field...... 69 III. Vulnerability Functions Damage Estimates ...... 70 IV. Insurance Functions Company Loss Estimates...... 72 V. Average Annual Loss Functions Loss Costs...... 78 VI. General...... 88 VII. Baseline Tests ...... 89 OutPut Ranges...... 102 References...... 103

Applied Research Associates, Inc. i February 2002 Table of Contents

LIST OF FIGURES

Figure 1. Comparison of Modeled and Observed Wind Speeds at Inland Locations ...... 3

Figure 2. Comparison of Modeled and Observed Wind Speeds at Coastal Locations ...... 4

Figure 3 Comparison of Modeled and Observed Landfalling Counts of Hurricanes by Category (defined by wind speed) and Region ...... 10

Figure 4. Comparison of Modeled and Observed Mean Content Damage vs. Mean Building Damage ...... 18

Figure 5. Comparison of Modeled and Observed Mean ALE Damage vs. Mean Building Damage ...... 18

Figure 6. Comparison of Modeled and Observed Losses for Homeowner Policies ...... 20

Figure 7. Variation in Loss Costs vs. Number of Simulated Years ...... 22

Figure 8. Comparison of Historical and Modeled Distributions of the Central Pressure Deficit of all Tropical Storms Passing Within 155 miles of Milepost 1450...... 26

Figure 9. Comparison of Historical and Modeled Distributions of Storm Heading of all Tropical Storms Passing Within 155 miles of Milepost 1450...... 26

Figure 10. Overview of Hurricane Damage and Loss Modeling ...... 31

Figure 11. High Level Flowchart of Portfolio (Multiple Buildings – Multiple Site) Computer Model...... 32

Figure 12. The Five Florida Water Management Districts ...... 40

Figure 13. Comparison of Simulated and Observed Key Hurricane Statistics along the Gulf and Atlantic Coasts of the United States ...... 44

Figure 14. Comparison of Modeled and Observed Landfall Rates of Hurricanes as a Function of Intensity in Florida ...... 45

Figure 15. Comparison of Modeled and Observed Annual Number of Landfalling Hurricanes in Florida...... 46

Figure 16. Comparison of Modeled and Actual Losses as a Function of Peak Gust Wind Speed in Open Terrain...... 47

Figure 17. Comparison of Total Losses for Different Company (A, B, C, and D) and Storms for Homeowner Policies ...... 47

Figure 18. ARA Office Locations and Technical Specialties ...... 50

Applied Research Associates, Inc. ii February 2002 List of Figures

Figure 19. Example Comparisons of Modeled Degradation Rates to Kaplan-Demaria Decay Rates...... 61

Figure 20. Maximum One Minute Sustained Wind Speeds for the Historical Storm Set...... 64

Figure 21. Example of Maximum One Minute Sustained Wind Speeds From a Random 101 Year Period Taken From the 100,000 Year Stochastic Storm Set ...... 64

Figure 22. Region Definitions...... 65

Figure 23. Model Disclosure Summary ...... 77

Figure 24. Ground-Up Loss Cost for Wood Frame Houses...... 79

Figure 25. Ground-Up Loss Cost for Masonry Wall Houses...... 79

Figure 26. Ground-Up Loss Cost for Mobile Homes...... 80

Figure 27. Grid for Calculating Hourly Wind Velocities, Category 5 (Coordinates with Respect to Initial Storm Center at (0,0) at time 0)...... 97

Figure 28. Grid for Calculating Hourly Wind Velocities, Category 3 (Coordinates with Respect to Initial Storm Center at (0,0) at time 0)...... 98

Figure 29. Grid for Calculating Hourly Wind Velocities, Category 1 (Coordinates with Respect to Initial Storm Center at (0,0) at time 0)...... 98

Figure 30. Florida County Codes ...... 100

Figure 31. Florida Counties...... 101

Applied Research Associates, Inc. iii February 2002 List of Figures

LIST OF TABLES

Table 1. Model Variables...... 33

Table 2. Primary Databases Used by Model...... 38

Table 3. Model Variables...... 42

Table 4. Inland Distance of Hurricane Force Winds ...... 61

Table 5. Simulated Wind Radius Statistics...... 63

Table 6. Historical vs. Modeled Hurricane Frequencies...... 66

Table 7. Probability of Hurricane by Year...... 67

Table 8. Distribution of Hurricanes by Size ...... 68

Table 9. Example of Insurer Loss Calculation...... 72

Table 10. Validation Comparisons of Actual and Modeled Losses...... 74

Table 11. Historical Per Storm Ground-up Losses ...... 82

Table 12. Percentage Contribution of Hurricane Andrew to Historical Annualized Loss by Zip Code ...... 83

Applied Research Associates, Inc. iv February 2002 List of Tables

2001 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. The ARA hurricane model provides estimates of loss costs for personal lines residential property from hurricane events, excluding flood and storm surge, except as it applies to ALE. Losses are developed for buildings, the contents of a building, appurtenant structures, and additional living expenses. Gross losses are developed and insured losses are computed using policy information. The model begins to estimate damage to buildings when the peak gust wind speed (in open terrain) produced by a hurricane equals or exceeds 50 mph. This 50 mph peak gust threshold corresponds to a sustained (one minute average) wind speed (in open terrain) of about 40 mph. The methodology used in the model to predict damage and loss given a wind speed will be presented to the professional team during their visit. 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 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. Reference: Module 2, Section I, #2-#3 Reference: Module 2, Section I, #5 ARA’s full time staff involved in the development of the hurricane model has extensive experience in wind engineering and hurricane loss projection. The wind engineering field includes meteorology, structural engineering, bluff body aerodynamics, probability, statistics, and risk analysis. A number of our staff are considered experts in the field of wind engineering and hurricane modeling, as demonstrated through publishing of peer reviewed papers, acting as reviewers for papers related to hurricane modeling submitted to respected professional journals, and through their active involvement in the development of wind related codes and standards in the United States. Independent experts have also provided inputs and review to the model development evaluation. These individuals all abide by standards of professional conduct adopted by their professions.

Applied Research Associates, Inc. 1 February 2002 2001 Standards 5.1.3 Modelers Policy of Model Revision 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. Reference: Module 1,Section I, A.1 Reference: Module 1, Section I, A.9 ARA’s hurricane loss model is denoted HurLoss Version 3.0. We have developed and implemented a written policy for model revision, including methodology and data. 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. Reference: Module 1, Section II, B.11 Reference: Module I, Section II, B.13-15 Reference: Standard 5.5.3 All of the components of ARA’s hurricane model, including the wind speed, climatology, damage, and loss models have been individually developed and validated. The hurricane wind field model used in ARA’s hurricane model has been validated through comparisons of modeled and observed wind speed data using information collected from over 15 landfalling storms. Details describing the ARA wind field model, and the efforts taken to validate the model are given in Vickery, et al. (2000a). Example hurricane wind field validation plots are given in Figures 1 and 2. The damage and loss models used to produce vulnerability functions are based on first-principle approaches: engineering load and resistance analysis for physical damage estimation, and repair and reconstruction cost models for building loss estimation. These components have been independently validated. The actuarial components are theoretically sound and have also been validated separately without potential bias or compensation to the meteorology and vulnerability components. The relationships within the model among the meteorological, vulnerability, and actuarial components of the model are reasonable.

Applied Research Associates, Inc. 2 February 2002 2001 Standards Hurricane Elena, Pensacola NAS Hurricane Hugo, Charlotte Airport

360 360

270 270 n n o o i i t t c c e e r r i 180 i 180 nd D nd D i i W 90 W 90

0 0 0:00 4:00 8:00 12:00 16:00 0:00 4:00 8:00 12:00 Time UTC, 2 September 1985 Time UTC, 22 September 1989

150 150

ph) 120 ph) 120 m m ( ( d d e e e e

p 90 p 90 nd S nd S i i 60 60 W W e e nut nut i i

M 30 M 30 n- n- Te Te 0 0 0:00 4:00 8:00 12:00 16:00 0:00 4:00 8:00 12:00 Time UTC, 2 September 1985 Time UTC, 22 September 1989

150 150

120 120 ph) ph) m m ( ( d d e 90 e

90 e e p p

nd S 60 i

nd S 60 i W t W t s s u 30 u

30 G G

0 0 0:00 4:00 8:00 12:00 0:00 4:00 8:00 12:00 16:00 Time UTC, 22 September 1989 Time UTC, 2 September 1985

Hurricane Hugo, Columbia Airport Hurricane Hugo, Shaw AFB

360 360

270

n 270 n o o i i t t c c e e r r i 180 i 180 nd D nd D i i W 90 W 90

0 0 0:00 4:00 8:00 12:00 0:00 4:00 8:00 12:00 Time UTC, 22 September 1989 Time UTC, 22 September 1989

150 150 ) h ph) 120 p m 120 m ( d d ( e e e e p 90 p 90 S d nd S n i i 60 W 60 W e e t nut nu i i M 30 M 30 n- n- Te Te 0 0 0:00 4:00 8:00 12:00 0:00 4:00 8:00 12:00 Time UTC, 22 September 1989 Time UTC, 22 September 1989

150 150

120 120 ph) ph) m m ( ( d d e 90 e 90 e e p p nd S 60 nd S 60 i i W W t t s s u 30 u 30 G G

0 0 0:00 4:00 8:00 12:00 0:00 4:00 8:00 12:00 Time UTC, 22 September 1989 Time UTC, 22 September 1989

Figure 1. Comparison of Modeled and Observed Wind Speeds at Inland Locations.

Applied Research Associates, Inc. 3 February 2002 2001 Standards Hurricane Fran, RDU Airport Hurricane Frederic, Mobile

360 360

270 270

180 180

Wind Direction Wind 90 Wind Direction 90 0 20:00 0:00 4:00 8:00 12:00 0 20:00 0:00 4:00 8:00 12:00 Time UTC, 12-13 September 1979 Time UTC, 5 6 September 1996 150

150 120

120 90 90 60 60 30 30 Ten-Minute Wind Speed (mph) Speed Wind Ten-Minute

Ten Minute Wind (mph) Speed 0 0 20:00 0:00 4:00 8:00 12:00 20:00 0:00 4:00 8:00 12:00 Time UTC, 12-13 September 1979 Time UTC, 5 6 September 1996 150 150

120 120

90 90

60 60

30

30 (mph) Speed Wind Gust Gust Wind Speed (mph)

0 0 20:00 0:00 4:00 8:00 12:00 20:00 0:00 4:00 8:00 12:00 Time UTC, 5 6 September 1996 Time UTC, 12-13 September 1979

Figure 1 (concluded). Comparison of Modeled and Observed Wind Speeds at Inland Locations.

Hurricane Fran, Kure Beach Hurricane Bertha, Kure Beach 360 360

270 270

180 180 Wind Direction Wind Direction Wind 90 90

0 0 12:00 16:00 20:00 0:00 4:00 8:00 12:00 0:00 4:00 8:00 12:00 16:00 20:00 0:00 4:00 8:00 Time UTC, 5 6 September 1996 Time UTC, 12 13 July 1996

150 150

120 120

90 90

60 60

30 30 Ten (mph) Wind Speed Minute Ten (mph) Speed Wind Minute 0 0 12:00 16:00 20:00 0:00 4:00 8:00 12:00 0:00 4:00 8:00 12:00 16:00 20:00 0:00 4:00 8:00 Time UTC, 5 6 September 1996 Time UTC, 12 13 July 1996

150 150

120 120

90 90

60 60

30 30 Gust Wind Speed (mph) Wind Speed Gust Gust WindSpeed (mph)

0 0 12:00 16:00 20:00 0:00 4:00 8:00 12:00 0:00 4:00 8:00 12:00 16:00 20:00 0:00 4:00 8:00 Time UTC, 5 6 September 1996 Time UTC, 12 13 July 1996 Figure 2. Comparison of Modeled and Observed Wind Speeds at Coastal Locations.

Applied Research Associates, Inc. 4 February 2002 2001 Standards

Hurricane Elena, Dauphin Island Sea Lab Hurricane Hugo, Myrtle Beach AFB

360 360

270 270

180 180 Wind Direction Wind 90 Wind Direction 90

0 0 0:00 4:00 8:00 12:00 16:00 20:00 0:00 4:00 8:00 12:00 Time UTC, 2 September 1985 Time UTC, 21 22 September 1989

150 150

120 120

90 90

60 60

30 30 Ten (mph) Wind Speed Minute Ten (mph) Wind Speed Minute 0 0 0:00 4:00 8:00 12:00 16:00 20:00 0:00 4:00 8:00 12:00 Time UTC, 2 September 1985 Time UTC, 21 22 September 1989

150 150

120 120

90 90

60 60

30

30 (mph) Wind Speed Gust Gust Wind Speed (mph) Wind Speed Gust

0 0 20:00 0:00 4:00 8:00 12:00 0:00 4:00 8:00 12:00 16:00 Time UTC, 21 22 September 1989 Time UTC, 2 September 1985 Figure 2 (concluded). Comparison of Modeled and Observed Wind Speeds at Coastal Locations 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 date of the updated information shall be disclosed. Zip code centroids, when used in the model, shall be based upon 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. Reference: Module 3, Section VI, #1 Reference: Module 3, Form A Zip code centroids used in the model are the population centroids, and are updated at least every 12 months. Maps showing the zip code boundaries and the associated centroids will be available to the professional team. 5.1.6 Identification of Units of Measure of the Model All units of measure for model inputs and outputs shall be clearly identified. Reference: Module 1, Section I, C.2 All units of measure for model inputs and outputs are clearly identified. 5.1.7 Visual Presentation of Data

Applied Research Associates, Inc. 5 February 2002 2001 Standards 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. Reference: Module 3, Section V, #3 Color coded (Blue and Red) continuous tone maps have been produced and are presented in Module 3. Additional maps, including color contour maps, will be shown to the professional team during the on-site review. 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 final outputs of length, wind speed and pressure, for the purpose of estimating damage and loss, are given in terms of statute miles, miles per hour, and millibars, respectively. 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. Reference: Module 3, Section II, #2 The basic output from the wind field model is the mean hourly wind speed at a height of 10m above the surface. This mean hourly value is converted to a peak gust value for computation of wind induced damage. The conversion of the mean hourly wind speed to the peak gust wind speed is accomplished with the terrain dependent ESDU gust factor models, shown to be suitable for describing the gust characteristics of hurricane winds in Vickery and Skerlj (2000). 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 2000 hurricane season and/or the HURDAT (HURricane DATa) data set, is found in the Report of Activities as of November 1, 2001 under Section VII, Compliance With Standards and Related

Applied Research Associates, Inc. 6 February 2002 2001 Standards Information, #4. All proposed alternatives to the characteristics of specific storms in the storm set shall be subject to the approval of the Commission. Reference: Module 1, Section II, B.7-8 Reference: Module 3, Section I The storm set used by ARA to define the hurricane climatology includes all storms given in the HURDAT database. The storm database encompasses the period 1886 through 2000. The storm set used by ARA includes all storms defined in the official “event set”. 5.2.4 Hurricane Characteristics Methods for depicting all modeled hurricane characteristics (e.g., wind speed, 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. Reference: Module 1, Section II, B.1-8 Reference: Module 3, Section I Reference: Standard 5.6.2 The wind speeds associated with a modeled hurricane are estimated using ARA’s peer reviewed hurricane wind field model as described in Vickery, et al. (2000a,b). The following paragraphs summarize key elements. Using ARA’s storm track model, the number of storms to be simulated in any one year is obtained by sampling from a negative binomial distribution having a mean value of 8.7 storms/year and a standard deviation of 3.7 storms/year. The starting position, date, time, heading, and translation speed of all tropical storms as given in the HURDAT database are sampled and used to initiate the simulation. Using the historical starting positions of the storms (i.e., date and location) ensures that the climatology associated with any seasonal preferences for the point of storm initiation is retained. Given the initial storm heading, speed and intensity, the simulation model estimates the new position and speed of the storm based on the changes in the translation speed and storm heading over the current six-hour period. The changes in the translation speed, c, and storm heading, θ, between times i and i+1 are obtained from, ∆ =+ψ +λ + +θ +ε lnca12a a3 a4lncia5 i (1a) ∆θ =+ψ +λ + +θ +θ +ε bb12 b3 b4ciib5 b6 i−1 (1b)

where a1, a2, etc., are constants, ψ and λ are the storm latitude and longitude,

respectively, ci is the storm translation speed at time step i, θ i is the storm heading at time step i, θi-1 is the heading of the storm at time step i-1, and ε is a random error term.

The coefficients a1, a2, etc., have been developed using 5-degree by 5-degree grids over the entire Atlantic basin. A different set of coefficients for easterly and westerly headed storms is used. As the simulated storm moves into a different 5-degree by 5-degree square, the coefficients used to define the changes in heading and speed change accordingly.

Applied Research Associates, Inc. 7 February 2002 2001 Standards The central pressure of a storm is modeled through the use of a relative intensity parameter which is coupled to the sea surface temperature. Modeling hurricanes using this relative intensity concept was first used in single point simulations by Darling (1991). Note that while the actual central pressure of a hurricane is a function of more than the sea surface temperature (i.e., wind shear aloft, storm age, depth of warm water, etc.), the modeling approach is an improvement over traditional simulation techniques in that the derived central pressures are bounded by physical constraints, thus eliminating the need to artificially truncate the central pressure distribution. The relative intensity approach is based on the efficiency of a tropical cyclone relative to a Carnot cycle heat engine and the details of the approach given in Darling (1991). To compute the relative intensity, I, of a hurricane, we use the mean monthly sea surface temperatures in the Atlantic Basin (given in one-degree squares) at the location of the storm, combined with the central pressure data given in the HURDAT data base (see description in Jarvinen, et al., 1984), an assumed relative humidity of 0.75, and a temperature at the top of the stratosphere taken to be equal to 203o K (Emanuel, 1988). Using the approach given in Darling (1991), every central pressure measurement given in HURDAT is converted to a relative intensity. During the hurricane simulation process, the values of I at each time step are obtained from,

ln(Ic+−) =+cln(I) +cln(I) +cln(I−) +cT+c(T−T) +ε (2) ii101 2i13i24s5si+1 si

The coefficients c0, c1, etc., vary with storm latitude, storm intensity, basin (i.e., Gulf of Mexico or Atlantic Ocean), and heading (i.e., Easterly or Westerly direction). Near the US coastline, where more continuous pressure data is available, finer, regionally specific values of these coefficients are developed. These regionally specific coefficients take into account changes in the relationships between sea surface temperature and storm intensity that may be influenced by subsurface water temperatures as described, for example, in Chouinard, et al. (1997). These regional coefficients preserve the variations in local hurricane climatology along the coastline, and through small adjustments in the coefficients, the model can be calibrated to match historical landfall rates of hurricanes. In the modeling process, once a simulated storm makes landfall, the reduction in central pressure with time is modeled using the filling models described in Vickery and Twisdale (1995a). If a storm moves back over water, Equation 2 is again used to model the variation in central pressure with time.

The radius to maximum winds (Rmax given in statute miles) is modeled using: = + ϕ − ∆ 2 + ε 2 ε ln Rmax 1.587 0.02623 0.000031208 p ; r =0.325, = 0.397 (3)

where the error term, ε, is normally distributed. Equation (3) is an update of the Rmax model given in Vickery, et. al. (2000b). The updates include the addition of recent intense storms (such as Floyd, 1999 and Mitch, 1998). The modeling of the Holland pressure profile parameter, B, is unchanged from that given in Vickery, et. al. (2000b). 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 coastal location and to the modeled winds in each hurricane which causes

Applied Research Associates, Inc. 8 February 2002 2001 Standards 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: A scale from 1 to 5 that measures hurricane intensity.

Category Winds (mph) Central Pressure (MB) Damage

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 Reference: Module 3, Section I, #1-3 Reference: Module 3, Form B Reference: Standards 5.6.2 and 5.6.3 The storm intensity at the time of landfall (or any other time), as defined by the Saffir-Simpson scale, can be computed using either the one minute sustained wind speed or the central pressure. All comparisons of storm intensity presented herein are based on the one minute sustained wind speed at a height of 10 meters. 5.2.6 Hurricane Probabilities Modeled hurricane probabilities shall reasonably match the historical record through 2000 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 (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. Reference: Module 1, Section I, B.2 Reference: Module 1,Section II, B.7 Reference: Module 3, Section I Reference: Standards 5.6.2 and 5.6.3 Modeled probabilities of landfalling hurricanes by region are consistent with the limited observed (or historical) data. Figure 3 presents comparisons of modeled and observed landfall counts of hurricanes as a function of Saffir-Simpson category, as defined by wind speed. Data are shown as a function of region and for Florida as a whole. The agreement between the modeled and historical rates of landfalling storms as a function of intensity and region is reasonable, with the differences between historical and modeled results falling well within the range of uncertainty associated with the limited

Applied Research Associates, Inc. 9 February 2002 2001 Standards number of historical observations of landfalling hurricanes. Landfalling storms depicted in Figure 3 include both entering and exiting storms, as defined in Module 3 and in the “Official Storm Set”. The intensities of historical storms and the landfall region have been obtained directly from the information given in the “Official Storm Set”.

NORTHWEST FLORIDA - All Storms WEST FLORIDA - All Storms 16.0 16.0 Model Model 14.0 14.0 Historical Historical 12.0 12.0

10.0 10.0 t t n

8.0 u oun 8.0 C Co 6.0 6.0

4.0 4.0

2.0 2.0

0.0 0.0 12345 12345 Storm Category Storm Category

SOUTH EAST FLORIDA - All Storms NORTH EAST FLORIDA - All Storms 16.0 16.0 Model 14.0 Model Historical 14.0 Historical 12.0 12.0 10.0

t 10.0 n

u 8.0

ount 8.0 Co 6.0 C 6.0 4.0 4.0 2.0 2.0 0.0 12345 0.0 Storm Category 12345 Storm Category

FLORIDA - All Storms

45.0 Model 40.0 Historical 35.0 30.0

t 25.0 oun

C 20.0 15.0 10.0 5.0 0.0 12345 Storm Category

Figure 3. Comparison of Modeled and Observed Landfalling Counts of Hurricanes by Category (defined by wind speed) and Region. 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. Reference: Module 1, Section II, B.1 Reference: Module 1, Section II, B.7-8 Reference: Module 3, Section 1, #2 Reference: Module 3, Section 1, #8

Applied Research Associates, Inc. 10 February 2002 2001 Standards Reference: Standards 5.6.2 and 5.6.3 The modeled probability distributions of hurricane strength, forward speed, radii to maximum winds are consistent with that derived from historical storms in the Atlantic basin. The data used to derive the statistical models include the publication NWS-38, the HURDAT data base, and annual updates to the HURDAT database available from the National Hurricane Center/Tropical Prediction Center web site. Additional information on recent storms has been obtained from the Hurricane Research Division web site as well as detailed analyses of storms as published in AMS Journals including, Monthly Weather Review, Weather and Forecasting, etc. The resulting statistical models for radii to maximum winds, etc., used in the ARA model are an update of the models published in accepted scientific literature in Vickery and Twisdale (1995b) and Vickery, et al. (2000b). 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. Reference: Module 1, Section II, B.4-5 Reference: Module 3, Section I The effect of land friction is treated in the ARA hurricane wind field model through the use of widely accepted wind engineering boundary layer models. The wind speeds near the ground are reduced using the approach described in Vickery, et al. (2000a). The largest, and most rapid, portion of the reduction in wind speeds associated with a hurricane making landfall is produced by the ground friction, which is modeled using the ESDU (1983) boundary layer models. The approach used to model the effects of ground friction in the ARA model has been validated through comparisons of measured and modeled wind speeds from hurricanes taken offshore, at the coast, and inland from the coast. The database used to derive the ground roughness is based on the Florida Water Management District’s Land Use Land Cover database. 5.2.9 Hurricane Overland Weakening Rate The hurricane overland weakening rate used by the model shall be bounded by the observed extremes in historical records for Florida. The mean wind speed shall be within twenty percent (20%) of the Kaplan/DeMaria decay value or an alternative acceptable to the Commission. Reference: Module 1, Section II, B.3 Reference: Module 3, Section I ARA’s hurricane model uses the filling rates developed by Vickery and Twisdale (1995a) to describe the rate of weakening of a storm (as defined by the increase in central pressure) upon making landfall. The use of this peer reviewed filling model, coupled with

Applied Research Associates, Inc. 11 February 2002 2001 Standards the ESDU model used to model the reduction in wind speed associated with friction, produce reductions in wind speed within 20% of the Kaplan/Demaria model. 5.3 Vulnerability Standards 5.3.1 Derivation of Vulnerability Functions The method of derivation of the vulnerability functions shall be described and demonstrated to be theoretically sound. 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. Reference: Module 1, Section I, A.8 Reference: Module 3, Section III Reference: Module 3, Section IV, #3-6 Reference: Standard 5.6.2 ARA's vulnerability functions have been developed for residential buildings (including mobile homes) using a theoretically sound load and resistance modeling approach, coupled with empirically-derived elements. For each structure, the model estimates physical damage to the building. This physical damage is used in turn to estimate the loss to the building and contents separately. The building damage state is used in conjunction with a restoration model to estimate the costs associated with additional living expenses. ARA’s vulnerability functions include elements of each of the following: (1) historical data, (2) tests, (3) structural calculations, and (4) expert judgement. The resulting vulnerability functions separately compute damage for buildings, mobile homes, appurtenant structures, contents and additional living expense. The ARA personnel responsible for the development of the damage and loss models have extensive experience in wind load modeling, structural analysis, post- hurricane damage surveys and meteorology. 5.3.2 Required Vulnerability Functions Vulnerability functions shall separately compute damages for building structures, mobile homes, appurtenant structures, contents, and additional living expense. Reference: Module 3, Section III Reference: Module 3, Section IV, #3 ARA’s vulnerability functions separately compute damages 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.

Applied Research Associates, Inc. 12 February 2002 2001 Standards Reference: Module 3, Section III ARA’s vulnerability functions produce damage for windspeeds above and below the hurricane threshold of 74 mph. The minimum peak gust windspeed that produces damage is about 50 mph. 5.3.4 Construction Characteristics In the derivation and application of vulnerability functions, assumptions concerning construction type and construction characteristics shall be demonstrated to be reasonable and appropriate. Reference: Module 1, Section I, A.7 Reference: Module 3, Section III The assumptions related to construction types and construction characteristics are based on multiple datasets and engineering judgement. These data and assumptions will be discussed with the professional team. 5.3.5 Modification Factors Modification factors to the vulnerability functions or structural characteristics and their corresponding effects shall be disclosed and shall be clearly defined and their theoretical soundness demonstrated. Reference: Module 3, Section III, #3 Reference: Module 3, Section III, #6 No modification factors are used in the vulnerability functions. 5.3.6 Additional Living Expenses In the estimation of Additional Living Expenses (ALE), the model shall consider hurricane damage including storm surge damage to the infrastructure. The Additional Living Expense vulnerability function shall consider the time it will take to repair/reconstruct the home. Reference: Module 3, Section IV, #5-6 ARA’s Additional Living Expense model includes factors that are hurricane related, including the time it takes to repair/reconstruct the house and storm surge/wave damage to infrastructure. The model has been calibrated using insurance loss data. 5.3.7 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. ARA’s vulnerability model was originally developed to treat the wind resistive characteristics of buildings, including mitigation measures. Details of the model will be disclosed to the professional team. 5.4 Actuarial Standards 5.4.1 Underwriting Assumptions

Applied Research Associates, Inc. 13 February 2002 2001 Standards 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 upon accepted actuarial, underwriting, and statistical procedures. The methods used shall be documented in writing. For damage estimates derived from historical insured hurricane losses, the assumptions in the derivations concerning (1) construction characteristics, (2) policy provisions, and (3) relevant underwriting practices underlying those losses shall be identified and demonstrated to be reasonable and appropriate. Reference: Module 1, Section I, B.4 Reference: Module 1, Section II, A.3-5 Reference: Module 3, Section IV The amount and quality of insurance information on historical losses varies significantly. The assumptions involved in the comparisons of historical insured hurricane losses to model estimated losses are documented as part of each study. Any adjustments are based upon accepted actuarial, underwriting, and statistical procedures. Sensitivity analyses are often used to assess how certain assumptions affect the estimated losses to ensure the reasonableness of comparisons. 5.4.2 Actuarial Modifications All modification factors to the actuarial functions or characteristics including but not limited to building code, quality, age, occupancy, stories, or condition of structure and their corresponding affects shall be disclosed and shall be clearly defined and their actuarial soundness demonstrated. The disclosure of modification shall include a description of the impact upon loss costs of the modification in accordance with the following: A: < -50% B: -50% to -25% C: -25% to 0 D: 0 to 25% E: 25% to 50% F: > 50% Reference: Module 1, Section I, A.6 Reference: Module 1, Section I, A.10 Reference: Module 1, Section I, C.1.c Reference: Module 3, Section III, #3 ARA’s modeling approach does not use modification factors to actuarial functions. 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,

Applied Research Associates, Inc. 14 February 2002 2001 Standards assessments, or profit margin. Hurricane loss projection models shall not make a prospective provision for economic inflation. Reference: Module 1, Section I, B.4 Reference: Module 1, Section I, C.1.a Reference: Module 3, Section III, #2 Reference: Module 3, Section V Reference: Module 3, Section VII Loss costs projections include only the direct costs associated with rebuilding/replacing the damaged structures, contents, appurtenant structures and the costs associated with additional living expenses. Additional costs such as expenses, risk load, investment income, premium reserves, taxes, assessments, or profit, are not included in the loss costs projections. Economic inflation is not included in the loss projections. 5.4.4 Insurer Inputs The modeler shall disclose any assumptions, fixed and 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. Reference: Module 1, Section I, A.10 Reference: Module 1, Section I, B.4 Reference: Module 1, Section II, A.3-4 Reference: Module 3, Section IV Assumptions that relate to insurer input include: 1. Insurance to Value. Unless specified prior to an analysis or part of a separate study, the replacement cost of a building is assumed to be equal to the insured limit of the building. Losses are calculated based on replacement value and capped at the appropriate policy limit. If information is provided regarding insurance to value, this information and how it was treated by the model would be documented in the analysis report. 2. Appurtenant Structures. Assumptions pertaining to appurtenant structure exposure and value are made outside of the model and would be documented in the analysis report. 3. Contents. Unless otherwise specified, the value of the contents is assumed to be equal to the insured value of the contents. Assumptions regarding content limit and value would be documented in the analysis report. 4. Additional Living Expenses. The model for additional living expenses is based upon a restoration model supplemented with insurance data. A change in the ALE policy limit has no effect on losses associated with ALE, unless the estimated ALE losses exceed the ALE limit. 5. Insurer Exposure by Zip Code. Each policy in the portfolio is checked to ensure it is located in a valid zip code. The treatment of invalid or out-of-date

Applied Research Associates, Inc. 15 February 2002 2001 Standards zip codes would be discussed with insurer and documented in the analysis report. Assumptions that can vary by specific insurer are disclosed in the output report. 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 shall be demonstrated to be reasonable. Reference: Module 1, Section I, C.1.a Reference: Module 3, Section III, #2 Reference: Module 3, Section VII Loss cost projections do not explicitly treat demand surge. A separate analysis would be required with increased labor and material costs in order to produce loss functions that might be representative of demand surge. Separate studies to treat demand surge would be documented in the analysis report. 5.4.6 Loss Costs - Meaning of “Damage” In calculating loss costs, damage shall be expressed as insurable losses. Reference: Module 1,Section II, A.5 ARA’s building performance model quantifies damage in terms of failure of building components, and not percent structural damage. A repair and replacement model estimates the cost to repair or replace the damaged components/building. With policy and deductible information, damage is quantified in terms of insured losses. 5.4.7 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. 1. Loss costs produced by the model shall be positive and non-zero for all zip codes. 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. 3. Loss costs cannot increase as friction or roughness increase, all other factors held constant. 4. Loss costs cannot increase as the quality of construction type, materials 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. 6. Loss costs shall 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 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.

Applied Research Associates, Inc. 16 February 2002 2001 Standards Reference: Module 1, Section I, C.1.b Reference: Module 3, Section V, #2 Reference: Module 3, Section V, #5 Reference: Module 3, Section VII The loss costs produced by the ARA model show no illogical relations with respect to risk. 1. Loss costs produced by the model are positive and non-zero for all Florida zip codes. 2. Color coded zip code maps will be provided to the professional team during their visit. 3. Loss costs decrease as the surface roughness increases, all other factors held constant. 4. Loss costs decrease as “quality” of construction increase. 5. Loss costs decrease as wind-resistive fixtures and/or construction techniques are applied. 6. Loss costs decrease with increasing deductibles, all other factors held constant. 7. The model does not explicitly treat “building code enforcement”. To the extent that building code enforcement is quantified in specific structural or building parameter terms, the model shows that loss costs decreases for stronger/higher quality construction, all other factors held constant. 5.4.8 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. Reference: Module 1, Section I, B.3 Reference: Module 3, Section IV, #1-2 Reference: Standard 5.6.2 The model produces statistical distribution of losses that are treated mathematically to reflect deductibles and coinsurance. The details of this approach and validation will be presented to the professional team during their visit. 5.4.9 Contents The model shall provide a separate mathematical representation of contents loss costs, and the modeler shall demonstrate its actuarial soundness. Reference: Module 3, Section IV, #5 Reference: Module 3, Section IV, #7 Reference: Standard 5.6.2 The ARA model produces direct estimates of damage to contents. The model has been validated through comparisons with actual loss data. Figure 4 shows a comparison of modeled and observed content loss as a function of the loss to the building.

Applied Research Associates, Inc. 17 February 2002 2001 Standards Observed Model Content Damage Ratio Damage Content

Building Damage Ratio

Figure 4. Comparison of Modeled and Observed Mean Content Damage vs. Mean Building Damage. 5.4.10 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. Reference: Module 3, Section IV, #6 Reference: Standard 5.6.2 Additional living expenses are estimated using a model which estimates the time required to rebuild a damage structure and includes a component for damage to infrastructure due to storm surge and waves. The model does not initiate the computation for additional living expenses associated with wind induced damage until the physical damage sustained to the building is significant enough such that the building is unlivable. ALE losses associated with storm surge and wave damage to the infrastructure can occur when there is no damage to the structure. Figure 5 shows a zip-code comparison of modeled and actual ALE costs.

Observed Model ALE Damage Ratio

Building Damage Ratio Figure 5. Comparison of Modeled and Observed Mean ALE Damage vs. Mean Building Damage.

Applied Research Associates, Inc. 18 February 2002 2001 Standards 5.4.11 Building Codes Information upon which building code quality and enforcement is assessed, if incorporated in the model, shall be objective and reasonably accurate and reliable. Reference: Module 1, Section 1, C.1.b Reference: Module 3, Section III, #3 Reference: Standard 5.6.2 ARA’s model does not treat building code “quality and enforcement” in the context of a single parameter model input. Building construction parameters are considered in the development of vulnerability functions that are input to the loss projection model. Building construction information is based on multiple data sources and are reasonably accurate and reliable. 5.4.12 Hazard Mitigation Data or information upon which differences in loss costs due to fixtures, design features, or construction techniques designed for hazard mitigation are derived, if incorporated in the model, shall be objective and actuarially reasonable. Reference: Module 1, Section I, A.6 Loss costs projections that reflect hazard mitigation wind-resistive design, fixtures, and construction techniques are based on objective information and actuarially reasonable modeling approaches. 5.4.13 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. Reference: Module 3, Section IV, #9 Reference: Module 3, Section V, #2 Reference: Standard 5.6.3 Figure 6 presents a comparison of modeled and actual total losses by storm and company for residential coverage. The comparisons indicate reasonable agreement between the observed and modeled losses. Additional, and more detailed, comparisons of modeled and actual incurred losses are given in our response to Module 3. Additional detail of actual and modeled zip-code level losses will be presented to the professional team.

Applied Research Associates, Inc. 19 February 2002 2001 Standards Comparison of Actual Company Losses to Modeled Losses

100

10

1

Actual Loss Model Loss 0.1

0.01

0.001

Figure 6. Comparison of Modeled and Observed Losses for Homeowner Policies. 5.4.14 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 standard 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. Reference: Module 3, Section I, #7 Reference: Module 3, Section I, #10 Reference: Module 3, Section V, #2 Reference: Module 3, Section V, #4 Reference: Standard 5.6.3 The ARA hurricane model calculated historical average annual loss for the FHCF aggregate exposure database is $2.39 billion per year. The ARA hurricane model calculated simulated average annual loss for the same exposure database $2.79 billion per year. The differences between the historical and simulated average annual statewide loss costs are statistically reasonable as seen through a t-test performed to examine the equivalence of means.

Applied Research Associates, Inc. 20 February 2002 2001 Standards 5.4.15 Output Ranges Any model previously found acceptable by the Commission shall provide an explanation suitable to the Commission concerning 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 from prior submission of greater than ten percent in maximum or minimum loss costs for any county shall be specifically listed and explained. 2. Differences from prior submission in the relativities between loss costs for building and the corresponding loss costs for contents shall be explained. 3. Differences from prior submission in the relativities among corresponding deductibles shall be explained. Reference: Module 3, Section V, #4-5

All counties have at least one value of the high, low or weighted averages that changed by at least 10%. Changes in the output ranges from the prior year submission are a result of changes in the hurricane storm set (updated to include 2000 storms), the use of the 2001 zip codes (vs. 2000 zip codes), a shift from a geographic centroid to a population centroid for zip codes, and changes to the terrain, damage, and building distribution models. The changes to the building stock distribution are based on new analysis and some additional data. The updated building stock model results in “stronger” buildings in Southeast Florida, and generally “weaker” buildings elsewhere. These changes affect the distribution of the building vulnerability and hence the loss costs. In addition, the relativities among deductibles and the relativities between content loss and building loss are also affected. 5.4.16 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. Reference: Module 1,Section II, C.2 Reference: Standard 5.6.3 Figure 7 shows the variation in mean loss costs as a function of the number of years of simulated storms for Alachua and Dade counties. As seen in these comparisons, the estimates of loss costs change little as the simulation approaches 100,000 years, indicating that the contribution to any errors associated with the sampling process are negligible.

Applied Research Associates, Inc. 21 February 2002 2001 Standards Alachua County 0.8 d e r

u 0.7 s n 0.6 0.5 $1000 I / L 0.4 A 0.3 de A

Wi 0.2 y t

n 0.1 u o

C 0 0 20000 40000 60000 80000 100000 Number of Simulated Years

Dade County 9 d e r

u 8 s n 7 6 $1000 I

/ 5 L A 4 A e

d 3

Wi 2 y 1 ount

C 0 0 20000 40000 60000 80000 100000 Number of Simulated Years

Figure 7. Variation in Loss Costs vs. Number of Simulated Years.

5.4.17 Total Estimated Losses The modeler shall demonstrate through the information submitted in Form B and Form D (Module 3, Section VII) that the model produces reasonable relationships among the total estimated losses produced by the model for building, appurtenant structures, contents, and additional living expense. Reference: Module 3, Section VII Information for Form B is given in the output files “ARA_2001FormB.xls” and “ARA_2001FormB.pdf”. Information for Form D is given in the output files “ARA_2001FormD.xls” and “ARA_2001FormD.pdf”.

5.5 Computer Standards 5.5.1 Primary Document Binder

Applied Research Associates, Inc. 22 February 2002 2001 Standards 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. Reference: Module 1, Section I Reference: Module 1, Section II The primary document binder contains fully documented sections, based on accepted software engineering practices, for each computer standard. 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. Reference: Module 1, Section I Reference: Module 1, Section II Reference: Module 3, Section VI, #2 All software requirements are specified and documented for each of the major components of the model. 5.5.3 Software 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 (that individual who can explain the functional behavior of the component and respond to questions concerning changes in code, documentation, or data for that component) shall be identified and documented. 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). Reference: Module 1, Section I Reference: Module 1, Section II Detailed control and data flow diagrams, interface specifications, and data file specifications are included in the primary document binder and the supplemental documentation binders for each of the major components of the model. Model component custodians are identified in the primary binder. Revision control for each component is maintained using Microsoft Visual Source Safe. 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

Applied Research Associates, Inc. 23 February 2002 2001 Standards level components shall incrementally be translated into a larger number of components until the code level is reached. Reference: Module 1, Section I Reference: Module 1, Section II Each major component of the model has been incrementally translated into a hierarchy of sub-components, which have been implemented in FORTRAN and C++. Documentation is provided in the source code to trace the functionality implemented within each module back to the design specifications. 5.5.5 Software Verification 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. Reference: Module 1, Section I Reference: Module 1, Section II ARA's software verification procedures include: (a) use of logical assertions and error checking throughout the code to ensure that data values are within valid ranges, (b) reviews of completed code by engineers on the team who were not responsible for developing that code, (c) walkthroughs of the code to ensure proper flow, (d) inspections of internal variables to ensure proper input, output, and processing, (e) reviews of outputs to ensure proper trends, and (f) validations against historical events, experimental data, and/or independently-computed results. 5.5.6 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. All components when interfaced shall function correctly. Reference: Module 1, Section I Reference: Module 1, Section II Reference: Standards 5.6.4 and 5.6.5 Verification and validation tests for the software components are documented through extensive comparisons of intermediate and final outputs produced by the code against historical, experimental, and/or independently-computed results. Results of these tests are documented, typically in the form of graphs or tables, in the component documentation binders. 5.5.7 Software Maintenance and Revision The modeler shall specify all policies and procedures used to maintain code, data, and documentation. The modeler shall use tracking software to track all errors, as well as modifications to code, data, and documentation. Reference: Module 1, Section I

Applied Research Associates, Inc. 24 February 2002 2001 Standards Reference: Module 1, Section II ARA uses Version 6.0 of the Microsoft Visual Source Safe revision control system to track all changes to the software. Only authorized users are allowed to check in changes to the archive. Project check points are used to document the individual revision numbers of the source code files, header files, workspace files, data files, and documentation files in each new release of the model. ARA uses Version 4.4 of Softwise’s PR-Tracker problem report tracking system to document, prioritize, and assign all problems and enhancements to the software, data, and documentation. 5.5.8 User Documentation The modeler shall have complete user documentation including all recent updates. Reference: Module 1, Section I Reference: Module I, Section II User documentation is maintained which specifies the content and format of all input and output files, the operation of all command line interfaces, and the operation of all graphical user interfaces. The user documentation is maintained along with the source code in the project revision control system. 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. Reference: Module 1, Section II, B.12 Reference: Module 3, Section I, #7 The use of historical data in developing the hurricane model has been demonstrated to be reasonable through publications in the scientific literature. 5.6.2 Comparison of Historical and Modeled Results The modeler shall demonstrate the agreement between historical and modeled results for hurricane frequencies, tracks, intensities, and physical damage using accepted scientific and statistical methods. Reference: Module 1, Section II, A.1 Reference: Module I, Section II, B.7 Reference: Module 1, Section II, C.1 Reference: Module 1, Section II, C.3 Reference: Module 1, Section II C.5-6 Reference: Module 3, Section III, #4-5 Reference: Module 3, Section IV, #3-6 Goodness of fit tests, comparing modeled to historical data, have been performed using t tests, F tests, Chi-squared tests and the Kolmogorov-Smirnoff test, for all

Applied Research Associates, Inc. 25 February 2002 2001 Standards historical data used to define the hurricane hazard. Example comparisons are given in Figures 8 for the distribution of storm central pressure deficit for all tropical cyclones passing within 155 miles of milepost 1450 (approximately Miami). Figure 9 shows a comparison of the modeled and historical distributions of storm heading of all tropical storms passing within 155 miles of milepost 1450. Results of all of the statistical tests performed will be available to the professional team during their on-site visit.

Mile post 1450 K-S Te st:Pa ss C-S Te st 1:Pa ss C-S Te st 2:Pa ss 24

21 Observed 18 Model

15 t n

u 12 Co 9

6 3

0 7.5 22.5 37.5 52.5 67.5 82.5 97.5 112.5 127.5 142.5 Central Pressure Deficit (Minimum within 250 km of Milepost)

Figure 8. Comparison of Historical and Modeled Distributions of the Central Pressure Deficit of all Tropical Storms Passing Within 155 miles of Milepost 1450.

Mile post 1450 K-S Te st:Pa ss C-S Te st 1:Pa ss C-S Te st 2:Pa ss 25 Observed 20 Model

t 15 oun C 10

5

0 -170 -130 -90 -50 -10 30 70 110 150 Heading (Degrees, Clockwise from North)

Figure 9. Comparison of Historical and Modeled Distributions of Storm Heading of all Tropical Storms Passing Within 155 miles of Milepost 1450.

Applied Research Associates, Inc. 26 February 2002 2001 Standards 5.6.3 Uncertainty Characterization The modeler shall provide an assessment of uncertainty using confidence intervals or other accepted scientific characterizations of uncertainty. Reference: Module 1, Section II, B.9 Estimates of uncertainty are not a standard output of the model. Estimation of uncertainty will be shown to the professional team. 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 parameters 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. Reference: Module 1, Section I, A.5 Reference: Module 1, Section II, B.13-15 The model has been tested and sensitivities estimated for cases of simultaneous variations of input parameters. The sensitivity study results will be presented to the professional team. 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 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. Reference: Module 1, Section I, A.5 Reference: Module 1, Section II, B.9 Reference: Module 1, Section II, B.13-15 An uncertainty analysis has been performed and the results will be presented to the professional team.

Applied Research Associates, Inc. 27 February 2002 2001 Standards MODULE 1

I. General Description of the Model (Standards 5.5.1-5.5.8 for all items in this Section) A. In General 1. Specify the model and program version number reflecting the release date. (Standard 5.1.3) Our current submittal is HurLoss Version 3.0 with release date February 2002. 2. Provide a complete and concise description of your model, with a one- page introductory summary. Include a description of your methodology, particularly the wind components, the damage components, and the insured loss components. Indicate where probability distributions have been fit to historical data and demonstrate their agreement. Describe sensitivity and uncertainty analyses used in the development of your model. Describe the computer language/code in which your computer program is written and what type of computer hardware is needed. Specify the details of translation from model structure to program structure. ARA’s hurricane model has been developed using wind engineering principles to enable detailed estimates of damage and loss to buildings and their contents due to wind storms. The model uses a peer reviewed (reviewed by both meteorologists and wind engineers) hurricane hazard model that enables the modeling of the entire track of a hurricane or tropical storm. The hurricane windfield model has been more extensively validated than any other published hurricane windfield model. The results of ARA’s hurricane hazard model are used directly in the design wind speed map given in ASCE-7- 98. Since the hurricane model simulates the entire hurricane track, whether the storm makes landfall or not, computation of loss does not require a storm to make landfall, but simply requires that the wind speed produced by the storm at any point exceed a predefined minimum value. The model has the ability to treat storms that make multiple land falls. ARA’s physical damage model is based on load and resistance analysis of building components. The physical damage model estimates the damage to the building in terms of failure of building envelope components. Insured loss is estimated from the building damage states using empirical cost estimation techniques for building repair and replacement. Contents loss is based on an empirical model that relates contents damage to building envelope performance. The building, appurtenance, contents, and loss of use components have been validated with insurance loss data. For portfolio assessments, a fast running loss function is developed for each building class. These functions are used to estimate losses for each coverage type and deductible in each simulated storm. The computational procedure is straightforward simulation and, typically, 100,000 years of storms are computed. The model can produce losses by coverage, policy, site, zip code, county, state, and portfolio levels. The storm-by-storm and year-by- year output data can be easily postprocessed to obtain average annual loss (AAL) as well

Applied Research Associates, Inc. 28 February 2002 Module 1 as the loss distribution statistics (Probable Maximum Loss, etc.) on a per occurrence or per year basis. Hurricane Hazard Model. The two key components of the hurricane hazard model are: (i) probabilistic models describing the occurrence rates, storm tracks, and intensities, and (ii) the hurricane windfield model. The probabilistic potion of the hurricane hazard model is described in detail in Vickery, et al. (2000). The key features of the storm track model are the coupling of the modeling of the central pressure with sea surface temperature, and the ability to model curved tracks that can make multiple landfalls. The entire track of a storm is modeled, from the time of storm initiation over the water, until the storm dissipates. The starting times (hour, day and month) and locations of the storms are taken directly from the HURDAT data base. Using the actual starting times and locations ensures that any climatological preference for storms to initiate in different parts of the Atlantic Basin at different times of the year is maintained. The coupling of the central pressure modeling to sea surface temperature ensures that intense storms (such as category five storms) cannot occur in regions in which they physically could not exist (such as the New England area) and, as shown in Vickery, et al. (2000b), the approach is able to reproduce the variation in the central pressure characteristics along the U.S. coastline. In the hurricane hazard model, the storm’s intensity is modeled as a function of sea surface temperature, until the storm makes landfall. At the time of landfall, the filling models described in Vickery and Twisdale (1995a) are used to model the intensity of the weakening storm. Over land, following the approach outlined in Vickery, et al. (2000b), the storm size is modeled as a function of central pressure and latitude. If the storm exits land into the water, the storm intensity is again modeled as a function of sea surface temperature, allowing the storm to possibly re- intensify and make landfall again elsewhere. The validity of the modeling approach for storms near the coastal United States is shown through comparisons of the statistics historical and modeled key hurricane parameters along the North American coast. These comparisons are given in Vickery, et al. (2000), where comparisons of occurrence rate, heading, translation speed, distance of closest approach, etc., are given. These comparisons are made using the statistics derived from historical and modeled storms that pass within 250 kilometers of a coastal milepost location. The comparisons are given for mileposts spaced 50 nautical miles apart along the entire United States Gulf and Atlantic coastlines. Hurricane Windfield Model. The hurricane windfield model used in our simulation model is described in detail in Vickery, et al. (2000a). The model uses the results of the numerical solution of the equations of motion of a translating hurricane. The asymmetries in a moving storm are a function of the translation speed of the storm and the non-linear interactions between the wind velocity vectors and the frictional effects of the surface of the earth. The numerical solutions of the equations of motion of the hurricane have been solved separately for a storm translating over the ocean and for a storm translating over land. The separate solutions were developed for the over land case and the over water case, since in the over water case, the magnitude of the surface drag coefficient is a function of the wind speed itself, whereas in the over land case the magnitude of the surface drag coefficient is wind speed independent. The outputs of the numerical model represent the integrated boundary layer averaged wind speeds, representative of a long duration average wind, taken as having an averaging time of one hour. The mean one hour average, integrated wind speeds are then combined with a

Applied Research Associates, Inc. 29 February 2002 Module 1 boundary layer model to produce estimates of wind speeds for any height and averaging time. The boundary layer model, described in Vickery, et al. (2000a) and Vickery and Skerlj (2000) is based primarily on the ESDU (1982, 1983) models for the atmospheric boundary layer. The boundary layer model can deal with arbitrary terrain conditions (any surface roughness) changing both the properties of the mean flow field (i.e. the mean wind speed at a given height decreases with increasing surface roughness) as well as the gustiness of the wind (i.e. the gust factor increases with increasing surface roughness). The gust factor portion of the ESDU based model has been validated through comparisons to gust factors derived from hurricane wind speed traces as described in Vickery and Skerlj (2000). The entire hurricane wind field model (overall flow field, boundary layer model and gust factor model) has been validated through comparisons of simulated and observed wind speeds. These wind speed comparisons have been performed through comparisons of both the peak gust wind speeds and the 10 minute average wind speeds. The comparisons have shown the windfield model reproduces observed wind speeds well, matching both the gusts and the long period average winds. The model has been validated separately at offshore, coastal and inland stations, taking into account the effects of local terrain and anemometer height on the measured and simulated wind speeds. Damage and Loss Model. ARA’s modeling approach for damage and loss employs two separate models. A building performance model, using engineering-based load and resistance models, is used to quantify physical damage. The building physical damage model takes into account the effects of wind direction changes, progressive damage and storm duration. Economic loss, given physical damage, is estimated using repair and reconstruction cost estimation methods. This process is therefore similar to how an insurance adjuster would estimate the claim, given observed damage to the building. Through direct simulations of thousands of storms with representative buildings, the building performance model produces outputs that are postprocessed into loss functions for building, contents, and loss of use. The damage and loss modeling methodology is shown schematically in Figure 10. The loss projection model uses these fast running loss functions (vulnerability functions) with the hurricane model to produce insured loss. Both the physical damage model and the loss model developed by ARA have been validated through comparisons of modeled and observed damage data collected after hurricane events. Separate models have been developed to estimate the financial losses to the building, the building’s contents, additional living expenses, and losses to appurtenant and exterior structures. Software, Hardware, and Program Structure. ARA’s detailed hurricane, building performance, and loss analysis models that are used to develop the fast-running hurricane loss modules have been developed in FORTRAN and C++. The databases supporting the models include binary data, flat text files, and MS Access files. These tools are run in- house and are not licensed to users. A Hurricane Risk Analysis Portfolio Tool is being developed separately in Microsoft Visual C++ 6.0.

Applied Research Associates, Inc. 30 February 2002 Module 1 Advanced Hurricane Model Building Physical Building Hazard Risk Performance Windfield Engineering Models Damage Losses & Track Model Model Loads Fast Running Resistances Loss Functions Failures

Validation (individual component and end-to-end models)

(a) Individual Buildings and Building Class Performance Model

Portfolio Type Policy Exposure Inputs Terrain Fast-Running Information Module Loss Functions

Hurricane Loss Ground-Up Insured Data Set Analyzer Losses Losses

Statistical End-to-End Validation

(b) Multiple Site – Multiple Building Loss Projections

Figure 10. Overview of Hurricane Damage and Loss Modeling. The applications require a PC with a Pentium II Intel-based processor, running a Windows NT operating system, a CD-ROM drive, and approximately 12.5GB of disk storage. Translation from Model structure to Program Structure. ARA uses a modular design approach in converting from model structure to program structure. The analysis process is broken into distinct phases, or modules: i.e., Wind, Damage, and Loss. Each of these modules is a stand-alone program. The architecture and program flow of each module are established in a structured design process through the creation of flow charts and/or “pseudo-code”. Each model element is translated into subroutines or functions on a one-to-one basis. Changes to the models are reflected by one-to-one changes in the software code. Although this approach is not a pure object-oriented design approach, it does incorporate some of the features, such as data encapsulation, that make the object- oriented approach successful. 3. Describe the theoretical basis for your model. Provide precise citations to or, preferably, copies of, the representative or any primary technical papers which help describe the theory underlying your model and which you relied on as to any particular component of the model. Theoretical Basis. As indicated earlier, the hurricane hazard model is developed using the historical database of storms given in HURDAT. ARA’s model has been peer

Applied Research Associates, Inc. 31 February 2002 Module 1 reviewed and accepted for publication in the American Society of Civil Engineers publication, The Journal of Structural Engineering. The model in its entirety was used to develop the design wind speeds given in ASCE-7-98 for the hurricane prone coastline of the United States. ARA’s hurricane model is routinely used by two of the three major North American Boundary Layer Wind Tunnel Laboratories to determine the site-specific hurricane climate risks in terms of directionally dependent speed frequencies for determining design wind loads for major buildings along the United States coastline, Caribbean, and Asia. The physical damage methodology has been developed using a load and resistance modeling approach. Load and resistance analysis methods are fundamental to an evaluation of building performance to extreme winds and are used extensively in the analysis and design of structural systems, including codified design. Some important theoretical and application references are given by Twisdale and Vickery (1995) in Chapter 20, “Extreme Wind Risk Assessment,” of the Probabilistic Structural Mechanics Handbook. Key references are highlighted in the reference list with an asterisk. 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 your computer program. Explain all interfaces and coupling assumptions. The main program analysis modules used in the multiple building model are shown in Figure 11. More detailed, proprietary flow-charts of each module will be available for the professional team, on-site visit.

Policy Data Zip Code Mapping

Terrain Portfolio Building Class Mapping Mapping

Hurricane Loss Fast-Running Data Estimation Loss Functions

Portfolio Ground-Up Information Loss Deductible, etc.

Insured Loss Figure 11. High Level Flowchart of Portfolio (Multiple Buildings – Multiple Site) Computer Model.

Applied Research Associates, Inc. 32 February 2002 Module 1 5. Provide a list and a description of the model variables and the outputs from your 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. (Standards 5.6.4 and 5.6.5) The model variables are given in Table 1. Model Outputs. Model outputs vary as requested by the client. Typical outputs include county wide losses by coverage, line of business, deductible type and event. This type of output can be readily used for PML estimates. Finer levels of outputs, such as at the policy level and zip code level are usually given as estimates of AAL by coverage, line of business, etc. Loss Costs from Historic Storms. Loss costs from historic storms can be produced using either the fast running loss functions developed for use in the portfolio model or with the building performance model (which takes into account storm duration, change of wind direction, etc.). The computation of loss costs from historical or stochastic storms is fundamentally the same. The losses are summed by territory, line of business, etc., and divided by the appropriate exposure and the number of years of storms.

Table 1. Model Variables Variable Critical Quantitative Range of Variables Estimation of local wind speed Central Pressure Yes Yes Continuous Function, Varies with Location – Difference Maximum Value of 150 mbar Translation Speed Yes Yes Continuous Function – Maximum Value of 70 mph (much less in Florida) Radius of Max Winds Yes Yes Continuous Function – Maximum Value of 155 miles Storm Direction No Yes Continuous Function –Bounded between 0 and 360 degrees Central Profile Yes Yes Continuous Function – Bounded between 0.5 Parameter and 2.5 Local Terrain Yes Yes Discrete Function Varies with land use Roughness category and region Prediction of damage Insured Value Yes Yes N/A Deductibles and Limits Yes Yes N/A Construction Class Yes Yes N/A Variables

Applied Research Associates, Inc. 33 February 2002 Module 1 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. (Standards 5.4.2 and 5.4.12) The model does not include any modification factors that are applied to any of the actuarial functions. Mitigation factors are incorporated only if mitigation studies are required. The details of the mitigation model will be shown to the professional team. 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.)? (Standard 5.3.4) The number of categories of different vulnerability functions used in a loss projection study depends on the objective of the study. For a basic analysis, wall construction is a common way to analyze and report results since insurers have wall construction classes. On the other hand, to develop a classification for wind vulnerability, many key building variables may be evaluated. In summary, the vulnerability functions may be based on either fine grained or coarse grained representations of the construction parameters. Examples of the number of categories or building classes considered in loss projection studies will be reviewed with the professional team. The basis for differentiation is the building performance model, which uses engineering analysis, empirical data, and judgement. The building categories used in the model are built up from detailed engineering load and resistance models that take into account building shape, roof shape, roof cover, garage doors, roof-wall connection, sheathing attachment, etc. For each building, damage and loss are estimated for the building, appurtenances, contents, and loss of use. Once the losses have been generated, fast running vulnerability (loss) functions are developed for the different coverages, etc., for each class. In this manner, the appropriate building vulnerability information is captured in the respective vulnerability functions used in the loss projection. 8. What are the primary or representative documents used or the research results which developed the model’s vulnerability functions (damage ratios)? (Standard 5.3.1) ARA engineers and scientists with significant experience in wind engineering, building performance, and post storm damage surveys developed the vulnerability functions. Examples of some of the reports used directly in the development of the ARA damage and loss models include Stathopoulos (1979), Meecham (1988), FEMA (1992), Ho (1992), Crandall, et al. (1993), Cunningham (1993), Sparks, et al. (1994), Monroe (1996), Reed, et al. (1996, 1997), Uematsu and Isyumov (1998), and Twisdale, et al. (1996, 1998). ARA engineers have performed post storm damage surveys following Hurricanes Andrew, Erin, Opal, Bertha, Fran and Bonnie. ARA engineers have also participated on FEMA BPAT teams. In part because of our proven expertise in damage model development and validation, ARA was selected by a panel of wind engineering and meteorology experts to develop FEMA’s HAZUS model for wind loss estimation.

Applied Research Associates, Inc. 34 February 2002 Module 1 9. What efforts have been made to update or revise your model or specific parts of the model? How many times have revisions been made? Discuss which changes you consider substantive and which you consider technical. When did the revisions occur? What specific revisions were made? (Standard 5.1.3) ARA’s advanced hurricane models have been under development since 1992 and have been used in hurricane risk and building performance projects since 1994. ARA’s models have been used for various insurance research and loss analysis studies since 1996. ARA’s individual building performance model has been used to evaluate individual buildings and building classes since 1997. The major changes to our model from Version 2.0 to 3.0 include updates to the hurricane model to include the year 2000 storms, updates to the zip code database, updates to the terrain database, and updates to the default building stock models. Potential future changes to the model will depend on outcomes of ongoing research and model validation studies performed at ARA. These updates are expected to be made on an annual basis. 10. Describe methods and procedures available to the model user so that the user may incorporate modifications into the model. (Standards 5.4.2 and 5.4.4) Modifications to the model are made only by ARA personnel. 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 loss costs each time it is run with the same input information. 2. What is the highest 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. (Standard 5.2.6) Loss costs can be produced at the coverage, policy, site, zip code, county, state and portfolio levels. The highest resolution is for a site and an individual building located at a latitude-longitude point. The resolution for the reported output ranges in this document is zip codes. The model explicitly considers the location and terrain roughness of each area, such as beach/coastal, inland location, etc. For each simulated storm, the effects of terrain and location relative to the storm, distance from coast, and time since landfall are treated. 3. How does the model handle deductibles (both flat and percentage), policy limits, replacement costs, and insurance-to-value when estimating loss costs? (Standard 5.4.8) ARA has developed fast-running loss functions that enable direct computations for deductibles within the simulations. Since loss estimates are performed on a policy-by- policy basis, the actual loss experienced by the building is simply equal to the ground-up

Applied Research Associates, Inc. 35 February 2002 Module 1 loss minus the deductible. Deductibles that are expressed as a percentage are converted to dollar equivalent and then applied. Unless otherwise indicated and documented, the insured value of the structure is assumed to be equal to the replacement value of the structure. In the estimation of loss costs, the modeled losses are not allowed to exceed the policy limits without separate documentation and justification. 4. Are annual aggregate loss distributions available? What review or tests have been done on these? (Standards 5.4.1, 5.4.3, and 5.4.4) Annual aggregate loss distributions can be produced directly from the outputs since ARA simulates 100,000 years of hurricane experience and all of the individual storms generated within each year. The validation tests for the hazard and loss models are discussed in Module II.C. 5. How are loss adjustment expenses considered within the loss cost estimates? Loss adjustment expenses are not considered. 6. Can your model distinguish among policy form types, for example, home-owners, dwelling property, mobile home, renters, condominium owners etc., and if so, what are your assumptions? Does your 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. ARA’s model can distinguish policy form types. Loss costs are produced separately for homeowner policies, mobile home policies, condominium and rental policies. Loss costs are produced separately for structure, contents, loss of use, mobile home, and commercial residential (condominiums). C. Other Considerations 1. Describe how your 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; (Standards 5.4.3 and 5.4.5) Socio-economic effects, such as cost inflation of labor and materials, or FEMA mitigation, can be considered only as a separate study. These effects would require a separate study with appropriate documentation. b. Building code and enforcement differentiation; (Standards 5.4.7 and 5.4.11) While ARA’s building performance modeling approach allows treatment of certain building code and enforcement issues, the loss projection model does not require separate inputs on building codes/enforcement. c. Specific construction characteristics (e.g., use of hurricane shutters); (Standard 5.4.2) Specific construction characteristics are treated the same way as described above for building code issues, that is, by evaluating the effects through the building performance model. The building performance model is run with specific

Applied Research Associates, Inc. 36 February 2002 Module 1 building class inputs to generate a loss function for that class (e.g., hip, hurricane straps, no shutters, etc.). The building classes can reflect a fine grained characterization or a very rough characterization (frame, masonry, …) of the exposures. Loss functions for one or more classes of buildings are used in the hurricane loss projection model. d. Storm surge and flood damage to the infrastructure. The impact of storm surge and flood damage is treated in losses associated with ALE. Storm surge and flood damage to the infrastructure does not impact building, content or appurtenant structure losses. 2. List your input variables for all of the categories in 1 above. (Standard 5.1.6) a. The inputs for “socio-economic” effects would be developed on a case-by- case basis and run with our building performance model. For example, these might include increased labor and material unit costs in our explicit building repair and replacement cost model. FEMA mitigation efforts, such as residential mitigation in Project Impact communities, would be treated by developing loss functions that represented the level of mitigation performed (shutters, roof deck, roof to wall connections, improved roof cover, etc.). Such analyses are performed on a case-by-case basis. b,c. The input variables for addressing building code and construction characteristics include building performance parameters that affect damage and loss in hurricanes, such as roof cover, roof deck attachment, roof-to-wall connection, roof shape, windows, doors, garage openings, etc. The effects of these variables are analyzed in the building performance model to produce a loss function that can then be used in loss studies for a site, territory, county, state, etc. d. The only input required for storm surge and flood damage to the infrastructure is the ALE limit. II. Specific Description of the Model (Standards 5.5.1-5.5.8 for all items in this section)

A. Model Variables 1. Using the list of model variables provided in response to I.A.5 above, describe the source documents and any additional research which was done 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 how you deviate, if at all, from the Commission’s hurricane set. Describe intensities used for these hurricanes. (Standard 5.6.2) The primary sources of information used to develop the statistical models for describing hurricane risk in the United States are (i), the HURDAT database, used for developing the models describing storm tracks, heading, occurrence rates and central pressure, (ii) the data given in the publication NWS-38, used for developing the models

Applied Research Associates, Inc. 37 February 2002 Module 1 describing central pressure and radius to maximum winds, and (iii) a database of upper level aircraft measurements provided by the Hurricane Research Division for developing statistical models defining the Holland Profile parameter, B. The development of our hurricane model used all storms in the HURDAT database whether they made landfall in Florida or not, and thus our data set inherently includes all storms in the Hurricane Commissions storm set. 2. List the current primary databases used by your model and the aspects of the model to which they relate. Indicate which databases are “public” and which are “proprietary”. The primary databases are listed in Table 2.

Table 2. Primary Databases Used by Model Primary Availability Model Use Database Historical Storms Public Used to derive hurricane climatology and to estimate losses from historical storms. Stochastic Storms Proprietary Estimates of Loss Costs and PML Surface Public as Raw data. Estimates of reduction in wind speed due Roughness Proprietary as used in model. to surface friction. Vulnerability Proprietary Estimates of Damage and Loss 3. What are your assumptions in the following areas: a. Meteorological The major assumptions related to meteorology are that : (i) The overall hurricane wind field can adequately be defined with the four key storm parameters (∆p, Rmax, translation speed, and Holland profile parameter B) coupled with the interaction between storm speed and the surface roughness. (ii) The boundary layer and gust structure within the storm can be adequately modeled with the ESDU based boundary layer models. (iii) The limited 100 year record of hurricane activity near the US coastline is representative of the hurricane activity expected to be experienced in the near future. Comparisons of measured and observed hurricane wind speeds suggest that assumptions (i) and (ii) noted above are reasonable. a. Damageability Building stock representation and terrain roughness effects. b. Insurance Coverage No assumptions are made. How does your model address the issue of demand surge? (Standards5.4.1 and 5.4.4)

Applied Research Associates, Inc. 38 February 2002 Module 1 Loss costs projections do not include demand surge. Separate analyses would be required to treat demand surge. 4. Are there other major or significant assumptions not listed above? If so, describe. (Standards 5.4.1 and 5.4.4) None. 5. Describe the nature and extent of actual insurance claims data which 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. (Standards 5.4.1 and 5.4.6) ARA has used data for seven storms with losses from four different companies for development and validation. The details will be provided to the professional team at the on-site visit. B. Methodology 1. Specify the wind speed(s) (e.g., one-minute sustained, peak gusts, etc.) used for loss estimation. (Standards 5.2.4 and 5.2.7) Wind induced losses are modeled as a function of the peak gust wind speed. 2. How is the asymmetric nature of hurricanes considered? (Standard 5.2.4) Asymmetries are modeled in the hurricane wind field as described in Vickery, et al., (2000) and in our response to question I.A.1. 3. Describe the nature of the filling rate function used. (Standards 5.2.4 and 5.2.9) The filling model used in our model is described in detail Vickery and Twisdale (1995a) and in Module 3, Section I, Question 6. 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. (Standards 5.2.4 and 5.2.8) Given a hurricane with fixed values of central pressure, heading, position, etc., and a site located on land, the variable which has the greatest impact on wind speed is the local ground roughness. Other variables having minor effects include distance from the coast, distance from the center of the storm, and the air-sea temperature difference. The database used for determining the surface roughness was obtained from the five Florida Water Management Districts (FWMD) that serve Florida, as noted in Figure 12. The FWMD Land-Use Land Cover databases from these agencies are considered superior sources for Florida terrain information due to higher spatial and land use resolution than the USGS LULC data. In addition, the FWMD data are based on more recent imagery. Given the LULC information, the aerodynamic surface roughness length, zo, associated with a given LULC category has been assigned a value, based on wind engineering expertise and judgement, coupled with sample aerial photography.

Applied Research Associates, Inc. 39 February 2002 Module 1 SRWMD

SJRWMD NWFWMD

SWFWMD

SFWMD

Figure 12. The Five Florida Water Management Districts. 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. (Standards 5.2.4 and 5.2.8) The characteristics of the hurricane that are needed to produce an estimate of wind speed at a site are as follows: (i) Central Pressure Difference (ii) Holland Pressure Profile Parameter (iii) Radius to Maximum Winds (iv) Storm Translation Speed (v) Storm Track (vi) Latitude and Longitude of Site (vii) Surface Roughness at the Site (viii) Distance of the Site from the Coast. This is applied statewide as discussed in Part I of Module 1. 6. Which variables in the wind speed component are dependent, and how is this dependence incorporated in the model? (Standard 5.2.4) The storm central pressure is modeled as a function of sea surface temperature. The storm track, which defines storm heading, translation speed, and distance to site, is a function of latitude and longitude. The radius to maximum winds and the Holland pressure profile parameter are a function of central pressure and latitude, as described in Vickery, et al. (2000b).

Applied Research Associates, Inc. 40 February 2002 Module 1 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. (Standards 5.2.3, 5.2.4, 5.2.6, 5.2.7, and 5.6.2) The model does not use a coastline segment approach to define hurricane parameters. The modeling process is described in detail in Vickery, et al. (2000). The hurricane frequency distribution by region is given in Module 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? (Standards 5.2.3, 5.2.4, and 5.2.7) The hurricane simulation model generates a time series of storms in the Atlantic basin simulating a period of 100,000 years. The storm generation and track modeling technique used to generate this time series of storms is described in the answer to the general description of the model and in Vickery, et al. (2000b). 9. Does the model produce confidence intervals for: a. Wind speed estimates given a set of hurricane parameters? The model does not automatically produce confidence estimates for wind speed given a set of hurricane parameters. b. Damage estimates given a wind speed estimate? The model does not automatically produce confidence intervals for damage estimates given wind speed. The model does produce the data that could be used for confidence interval determination. c. Annual loss costs? The model does not automatically compute confidence intervals for loss costs. Characterize the uncertainties in your model, for example, with an uncertainty analysis or suitable equivalent. Uncertainty refers both to possible model misspecifications and inherent random variation. (Standards 5.6.3 and 5.6.5) The major modeling uncertainties include hurricane, terrain, building stock, and vulnerability. An example of our analysis of hurricane modeling uncertainty is provided in Twisdale, Vickery and Hardy (1993), where it is shown that the uncertainty in wind speed as a function of return period is relatively large (CoV~10%). Additional uncertainty analysis results will be discussed with the professional team. 10. Describe the method or methods used to estimate annual loss costs needed for ratemaking. Identify any source documents used and research performed. Using a straightforward simulation of N years of storms, expected annual losses are computed simply as the sum of all losses, net of policy conditions, divided by N. Loss costs for a given territory are computed by dividing the average annual loss by the appropriate exposure base.

Applied Research Associates, Inc. 41 February 2002 Module 1 11. What functions or variables does your model consider to be independent? On what are the other functions or variables dependent (including latitude)? Are there limitations on the functions or variables that are a function of latitude? If so, describe. What are the intermediate (endogenous) variables which are part of the calculations between the inputs and outputs described in I.A.5? (Standard 5.1.4) Table 3 summarizes the relationships between variables.

Table 3. Model Variables Variable Dependent on Central Pressure Sea Surface Temperature, Latitude and Longitude Difference (Over Water) Central Pressure Central Pressure Difference at time of Landfall, Location of Difference (Over Land) Landfall, Translation Speed, Time Since Landfall Radius to Maximum Central Pressure Difference, Latitude Winds Holland Profile Central Pressure and Latitude Parameter Translation Speed Latitude and Longitude Wind Speed Hurricane Parameters, Local Roughness, Height Above Ground Heading Latitude and Longitude Building Loss Physical Damage to Building, Insured Value of Building, Actual Value of Building, Deductible Content Loss Physical Damage to Building, Insured Value of Contents, Actual Value of Contents, Deductible Intermediate variables in the building performance model include representation of the physical damage to buildings. 12. Identify the form of the probability distribution used for each function or variable, if applicable. What statistical techniques are used for distributions that are estimates? What tests are used for goodness of fit? (Standard 5.6.1) The hurricane track modeling approach used does not sample from predefined distributions of central pressure, heading, etc. The model predicts the parameters (position and central pressure), of a storm at the next time step based on its position, speed and heading at the current time step and up to two prior time steps. The resulting distributions of translation speed, heading, distance of closest approach are log-normal, bi-normal, and either uniform or linear, respectively. The distribution of the central pressure is approximately Weibull. The radius to maximum winds is modeled using a log-normal distribution, dependent on the central pressure difference and latitude, and the Holland profile parameter is modeled as being normally distributed about the regression line with a mean linearly dependent on central pressure and latitude. 13. What is the most sensitive aspect of your model? Is this sensitivity based upon a) an assumption, b) an underlying datum unique to your model, or c) a technique which the model employs? Discuss fully and provide an example to illustrate how (to what degree) this sensitivity affects output results. (Standards 5.1.4, 5.6.4, and 5.6.5)

Applied Research Associates, Inc. 42 February 2002 Module 1 The single most important element in loss estimation is the wind speed since the loads and, hence, physical damage to a building are proportional to the square of the wind speed. Thus the estimation of losses is sensitive to the hurricane wind climate and the wind model. Loss results are particularly sensitive to the occurrence rate and area of the more intense hurricanes (Saffir-Simpson Category 3 or higher). Given a location (i.e., the hurricane wind climate is held fixed) another key variable is surface roughness and the manner in which the boundary layer is modeled for the sea-land transition. Loss predictions are, or course, sensitive to the building construction parameters and, hence, the vulnerability functions used. These observations of sensitivity are based on numerous detailed sensitivity analyses, theoretical considerations, as well as field observations. Examples showing the impact of the different parameters used in the model on loss cost estimates will be provided to the professional team during their visit. 14. Are there other aspects of your model that may have a significant impact upon the sensitivity or variation in output results? (Standards 5.1.4, 5.6.4, and 5.6.5) The key modeling sensitivities for loss estimation are listed above. 15. What sensitivity and uncertainty analyses have been done on the model’s variables? (Standards 5.1.4, 5.6.4, and 5.6.5) One-by-one sensitivity studies have been performed examining the effects of terrain, building characteristics, and wind speed. Sensitivity studies have also been performed examining hurricane climate modeling. A model-independent analytic study has been performed that shows the sensitivity of loss cost estimation for a point location. A conference paper (Twisdale, et al., 1993) has been published on uncertainties in hurricane wind risk estimation for a single location. C. Validation Tests 1. What were the nature and results of the tests performed to validate the wind speeds generated? (Standard 5.6.2) Wind speed validation studies have been performed using data obtained from Hurricanes Frederic (1979), Alicia (1983), Elena(1985), Hugo (1989), Bob (1991), Andrew (1992), Emily (1993), Erin (1995), Opal (1995), Bertha (1996), Fran (1996), Bonnie (1997), and Georges (1998). In all cases, quantitative comparisons of modeled and observed estimates of wind speeds are only made if the measured records of wind speeds are continuous (or contain daily maxima), the height of the anemometer is known, the averaging times of both the gust and long term average is known and the terrain surrounding the anemometer is known. Comparisons for all valid records are given in Vickery, et al. (2000) for storms up to and including those that occurred in 1996. 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 which were used and the results. Specify the number of hurricanes or trials which were used. (Standard 5.4.16) A direct statistical validation of expected loss costs by region is not possible owing to the limited hurricane loss data at any given region, and thus indirect validation

Applied Research Associates, Inc. 43 February 2002 Module 1 procedures are required. This validation process assumes that if the hurricane climatology, in terms of hurricane intensity, frequency, translation speed, filling and windspeed, is properly modeled, and if the prediction of loss given a hurricane is properly modeled, then the loss cost estimates should be valid. Comparisons of modeled and observed wind speeds were presented in the answer to question 5.1.4 of the General Standards. These comparisons showed that given information describing a storm, including the central pressure difference, ∆p, the radius to maximum winds, Rmax, translation speed, Holland pressure profile parameter, B, heading and location, the wind field model is able to provide good estimates of wind speed at a site. Figure 13 shows comparisons of the modeled (simulated) and historical (derived from HURDAT) values of storm heading, translation speed, distance of closest approach and occurrence rate along the entire US coastline. Data are given for all storms passing within 155 miles of a particular milepost, spaced at 50 nm increments along the coast.

1.6 80 HURDAT (MEAN) 1.4 60 SIMULATED (MEAN) e 1.2 40

1 20

0.8 0 HURDAT (MEAN) Heading 0.6 -20 HURDAT (STD DEV) SIMULATED (MEAN) 0.4 -40 Annual Occurrence Rat Annual Occurrence SIMULATED (STD DEV) 0.2 -60

0 -80 0 500 1000 1500 2000 2500 3000 0 500 1000 1500 2000 2500 3000

Milepost Milepost

40 20 HURDAT (MEAN) ) HURDAT (MEAN) 35 HURDAT (STD DEV) 10 S MULATED (MEAN)

) SIMULATED (MEAN) 30 SIMULATED (STD DEV) 0 25 -10 20 -20 15

-30

Translation Speed (mph Speed Translation 10

5 (miles Distance Approach Minimum -40

0 -50 0 500 1000 1500 2000 2500 3000 0 500 1000 1500 2000 2500 3000 Milepost Milepost

Figure 13. Comparison of Simulated and Observed Key Hurricane Statistics along the Gulf and Atlantic Coasts of the United States. Figure 14 shows a comparison of the simulated and modeled landfall rate of hurricanes as a function of storm intensity in Florida. This comparison shows the combined effect of intensity and frequency modeling, indicating that the model is performing well in its ability to reproduce the statistics of landfall counts of storms as a function of intensity in Florida. The landfall counts given in Figure 14 represent the first landfall data for entering storms (upper plot) and the all storm data, which includes first, second, third landfall, etc. for entering storms as well as all exiting storms.

Applied Research Associates, Inc. 44 February 2002 Module 1 FLORIDA - Entering Storms

35.0

s 30.0 Model ear

Y Data 25.0 01 1 n i 20.0 s t n e

v 15.0 E of

er 10.0 b m u

N 5.0

0.0 12345 Storm Category

FLORIDA - All Storm s 60.0 Model s r Data

a 50.0 e

01 Y 40.0 1

in s t 30.0 en v E

of 20.0 ber m

u 10.0 N

0.0 12345 Storm Category

Figure 14. Comparison of Modeled and Observed Landfall Rates of Hurricanes as a Function of Intensity in Florida. Figure 15 shows a comparison of the historical and modeled annual rate of hurricane landfalls in Florida. In the comparisons given in Figure 15, each hurricane that makes landfall in Florida is counted as a single event for both the modeled and historical storms. The agreement between the annual rate historical and modeled hurricane landfalls is very good.

Applied Research Associates, Inc. 45 February 2002 Module 1 0.7 Model 0.6 Historical 0.5

0.4

0.3 Probability 0.2

0.1

0 012345678910 Number of Landfalling Hurricanes/Year

Figure 15. Comparison of Modeled and Observed Annual Number of Landfalling Hurricanes in Florida. The ability of the model to produce reasonable estimates of loss as a function of wind speed is shown, for example, in Figure 16. As seen in this example, the model is able to reasonably reproduce the observed losses as a function of the peak wind speed in a storm. Figure 17 shows separate examples of total losses experienced by different companies for various events. 3. What were the nature and results of the tests performed to validate the damage estimates generated? (Standard 5.6.2)

Figures 16 and 17 present example comparisons of simulated and observed losses from recent hurricanes. Other intermediate results for damage estimation will be shown to the professional team. 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 which was done on these calculations. Insured losses from ancillary perils were not included in any loss cost calculations.

Applied Research Associates, Inc. 46 February 2002 Module 1 Actual (%)

Model (%)

Peak Gust Wind Speed in Open Terrain (mph)

Figure 16. Comparison of Modeled and Actual Losses as a Function of Peak Gust Wind Speed in Open Terrain.

Comparison of Actual Company Losses to Modeled Losses

100

10

1

Actual Loss Model Loss 0.1

0.01

0 001

Figure 17. Comparison of Total Losses for Different Company (A, B, C, and D) and Storms for Homeowner Policies.

Applied Research Associates, Inc. 47 February 2002 Module 1 5. What were the nature and results of any validation tests on any other aspects of the model? (Standard 5.6.2) All relevant validation tests have been discussed in the answers to previous questions and/or will be discussed in more detail in the answers to questions in Module 3. 6. Provide documentation of all validation tests performed. (Standard 5.6.2) Validations are discussed in Module 3.

Applied Research Associates, Inc. 48 February 2002 Module 1 MODULE 2

Background/Professionalism

1. Company Background

A. Describe the ownership structure of your company. Is your company affiliated with any other company? If so, describe the nature of the relationship. Applied Research Associates, Inc. is an employee-owned company. ARA is not affiliated with any other company. B. How long has your firm been in existence? Applied Research Associates, Inc. was founded in 1979. C. In what year was your model developed? ARA staff developed a hurricane risk model for nuclear power plant risk analysis in 1982-1983. This model was similar to Russell’s (1971, 1974) in that it used coast crossing segments, but included a probabilistic wind field model. New research to develop an advanced hurricane wind risk model was begun in 1992 under funding by the National Science Foundation. The current version of the hurricane model, including the damage and loss analysis methodology was developed beginning in 1995. D. How long have you been using your model for ratemaking purposes? Our model was approved in 2000 by the FCHLM and has been used since that time for ratemaking and reinsurance applications. Our model was used in technical support of a ratemaking filing in 1999. Our building performance model has been used for individual buildings and building classification loss estimation studies for insurance companies since 1997. E. In which states have you attempted to use your model for ratemaking purposes? Has your 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 you have previously used your model for ratemaking purposes. (The Commission may contact these persons to discuss your work.) Our model is currently accepted for use in Florida. F. Describe generally your company’s services and the percentage of the company’s annual income derived from each. ARA is a diversified engineering and applied science research, consulting, and software development firm. ARA has approximately 640 employees with 13 major offices throughout the United States, as shown in Figure 18. The risk analysis, wind engineering, and building performance capabilities are located in Raleigh, NC and Orlando, FL. Our building inspection and wind mitigation certification services are located in both Raleigh and Orlando. The breakout of services by broad classes is currently:

Applied Research Associates, Inc. 49 February 2002 Module 2 Defense Technologies 34% Civil Technologies 16% Environmental Technologies 13% Transportation 12% Tests and Measurements 8% Computer Software 8% Manufacturing 5% Systems Analysis 4% 100%

ARA’sA ’ Technical c ic Capabilities apab • Probabilistic Risk and Decision Analysis • Software Development • Wind Engineering • Environmental Engineering • Structural Engineering • Risk Management • Engineering Reliability • Facility/Security Engineering • Pavements and Geotechnical Engineering • Tests and Measurements • Explosive Effects

Figure 18. ARA Office Locations and Technical Specialties. Our hurricane risk analysis, risk management, insurance applications, and wind engineering services are included in the Civil Technologies. The approximate breakout of these services is: State Governments (Risk and Mitigation) 35% Research & Development (Hurricane and Wind) 30% Insurers and Insurance Organizations 20% Building Owners, Consultants, Wind Tunnels 15%

Applied Research Associates, Inc. 50 February 2002 Module 2 G. How long have you used your model for analyzing insurance company exposures or other such uses? Describe these uses. ARA’s model has been used in insurance related analyses and exposures since 1997. ARA’s hurricane model has also been used to perform individual residential risk/loss analysis and mitigation benefit-costs analyses in the State of Florida since 1997. About 2200 homes have been analyzed to date. Beginning in 1999, ARA’s models have been used to perform commercial building damage and loss risk assessments. 2. Professional Credentials (Standard 5.1.2 for all items in this section) A. List the names of your technical staff and consultants and indicate their highest degree obtained (discipline and University), their years of experience with hurricane modeling for ratemaking, and their credentials and years of experience in their area of expertise. Lawrence A. Twisdale, Ph.D., P.E., Civil Engineering, University of Ilinois Dr. Lawrence A. Twisdale is a Principal Engineer and Senior Vice President of Applied Research Associates, Inc. and has more then 28 years of experience directing and performing research and engineering analyses for extreme wind effects and structural response. Dr. Twisdale has 10 years experience in the application of hurricane modeling relevant to insurance loss analysis and rate making. Dr. Twisdale has been involved in developing methodology and applications to extreme wind risk analysis and wind-borne debris hazard analysis since 1974. Dr. Twisdale has performed numerous site-specific wind analyses, including tornado, hurricane, and extratropical cyclones. Dr. Twisdale was Co-Principal Investigator with Dr. Peter Vickery on the NSF research “Hurricane Wind Hazards and Design Risk in the United States.” This research has developed a physics and meteorology-based numerical windfield model as part of a wind hazard risk assessment methodology. Dr. Twisdale was PI on an NSF funded project to test a “Predictive Methodology for Building Cladding Performance Using a Sample of Hurricane Bertha Damage Data.” He was Principal Investigator on the Project Impact Study that involved performing mitigation analysis on 31 houses in Deerfield Beach, Florida. He has also led the efforts on the “Residential Construction Mitigation Program (RCMP)” in 1998 in SE Florida that involved engineering mitigation analysis of over 1100 houses. The RCMP in 1999-2000 will evaluate 1500 houses in the Panhandle and Central Florida. Dr. Twisdale is the Co-Principal Investigator for FEMA’s HAZUS Wind Loss Estimation Methodology. He was Principal Investigator on several industry funded research projects on the “Analysis of Hurricane Windborne Debris Impact Risks for Residential Structures” that influenced the SSTD-12 and ASTM Standards. He is a member of the Wind Engineering Research Council, IBHS Wind Committee, and the American Society of Civil Engineers. He has served on several ASCE technical committees on wind effects, structural reliability, and dynamic structure response. Dr. Twisdale has over 70 technical papers in wind engineering, risk analysis, response of structures to extreme wind, and wind-borne debris effects. Dr. Twisdale was an invited author to prepare the chapter on “Extreme Wind Risk Assessment” in the Probabilistic Structural Mechanics Handbook, published in 1995.

Applied Research Associates, Inc. 51 February 2002 Module 2 Peter Vickery, Ph.D., Engineering Science, University of Western Ontario Dr. Peter Vickery joined Applied Research Associates in 1988. Prior to joining ARA, Dr. Vickery completed both his Masters and doctoral studies at the University of Western Ontario, working with Dr. Alan Davenport, an internationally recognized expert and leader of modern wind engineering technology. Dr. Vickery has over 17 years of experience in wind engineering. He has 9 years experience in the application of hurricane modeling relevant to insurance loss analysis/rate making. Dr. Vickery has published six peer reviewed journal papers related to hurricane risk and additional papers related to wind loads on buildings and other structures. Dr. Vickery serves on the ASCE-7 wind load task committee developing wind loading provisions for use in the United States, and has served on the Board of Directors of the American Association for Wind Engineering. In 1996, Dr. Vickery received the Collingswood Prize from ASCE for the best paper published by a younger member. With Dr. Lawrence Twisdale of Applied Research Associates, Dr. Vickery developed the hurricane missile models used in the wind load damage tools. The hurricane missile model results were instrumental in defining the wind borne debris risk criteria as given in the SSTD-12 missile protection standard and the new ASTM wind borne debris standard. Dr. Vickery has performed post storm damage surveys following Hurricanes Andrew, Erin, Opal, Fran, Bertha and Bonnie. Following Hurricane Andrew, Dr. Vickery served as an expert witness testifying to the wind speeds produced by the storm. He was responsible for developing the wind load and damage portion of the loss model used for residential structures and played the primary role in the development of ARA’s hurricane simulation model. Dr. Vickery is Co-PI on ARA’s effort to develop the HAZUS Wind Loss Estimation methodology and software and our current Mitigation Grant Feasibility Study for the Hawaii Hurricane Relief Fund. Jason Lin, Ph.D., Aeronautical Engineering, University of Western Ontario Dr. Jason Lin joined ARA in 1997, after having worked six years as a Senior Research Engineer and Project Manager at the Boundary Layer Wind Tunnel Laboratory (BLWTL) at the University of Western Ontario, London, Canada, a world famous organization in the field of Wind Engineering. At BLWTL, Dr. Lin worked closely with several of the best known and respected wind engineers in the world: Alan G. Davenport, David Surry and Nick Isyumov. Dr. Lin’s work at BLWTL involved wind tunnel investigations, computer modeling and predictions of wind effects on various types of buildings, and computer modeling of storm wind fields and simulations of extreme wind climates. He has authored over 10 publications in professional journals, including several co-authored with Drs. Davenport, Surry (BLWTL) and Dr. Simiu (National Institute of Standards and Technology). Dr. Lin had also been invited by the Swedish National Institute for Building Research to co-lead a joint research program on building aerodynamics in Sweden in 1990-1991. Having been the Principal Engineer of the first wind laboratory sponsored by the Chinese construction industry and conducted wind studies on several of the then tallest buildings in Asia, Dr. Lin was named one of the 10 Most Outstanding Young Researchers in Engineering Sciences by the National Natural Sciences Foundation of China in 1988.

Applied Research Associates, Inc. 52 February 2002 Module 2 As a Senior Scientist at ARA, Dr. Lin is responsible for several technical components contributing to the establishment of the vulnerability functions, including building vulnerability to wind pressure effects and wind-borne debris impacts. He also contributes to the calculation of company loss estimates. He has 8 years experience in the application of hurricane modeling for insurance loss analysis. Peter F. Skerlj, M.E.Sc., Engineering Science, University of Western Ontario Mr. Peter Skerlj has over 7 years experience in wind engineering and hurricane modeling relevant to rate making. He has been involved in boundary layer wind tunnel studies including a study of mean external pressure gradients and residual pressures acting on pressure-equalized rainscreen wall systems. He has also been extensively involved in modeling the climatology of wind-driven rain across Canada. He has been involved in a NSF-funded project that developed a new approach for assessing hurricane risk in the United States. This approach has been used to develop the hurricane wind speed maps for ASCE 7-98. He has analyzed hurricane upper-level pressure and wind velocity data measured from NOAA aircraft for the development of hurricane windfield modeling parameters. He has been involved in a NSF-sponsored project that investigated the correlation between Doppler Radar measurements and the areal extent of strong winds in thunderstorms. He has been involved in a NSF-sponsored project that developed risk consistent hurricane-induced storm surge and wind modeling for the US coastline and is currently involved in a NSF-funded project on hurricane-induced wave modeling. He has been extensively involved in conceptualizing and developing a computer program to predict wind-induced damage to building structures, including prediction of damage to the external building envelope and prediction of water entering the building through failed envelope components. He has performed post storm damage surveys for Hurricane Bonnie (1998). Mr. Skerlj is a member of the American Society of Civil Engineers. Michael Young, M.E.Sc, Engineering Science, University of Western Ontario Mr. Young joined ARA in 1997 as a Scientist after earning a Master’s degree in Wind Engineering at the University of Western Ontario’s Boundary Layer Wind Tunnel Laboratory (BLWTL). Mr. Young has done physical wind tunnel modeling of low rise and high rise structures while working as a project engineer at BLWTL. He has also been employed as a project engineer at Rowan Williams Davies and Irwin, Inc. (RWDI) doing wind tunnel simulations on buildings in the US and overseas. Mr. Young has performed post-hurricane damage surveys for Hurricane Bonnie (1998). His current responsibilities include development of individual wind risk analysis tools for residential and commercial buildings, validation of existing damage and loss models, and development of software products for insurance portfolio risk assessment. Mr. Young is a member of the American Society or Civil Engineers, and also the Southern Building Code Congress, Inc. Mr. Young has 5 years of experience in the hurricane modeling field and 7 years of experience in wind engineering. Kevin Huang, Ph.D., Civil Engineering, Clemson University Dr. Huang, who has more than 5 years of experience in hurricane wind field modeling and wind-related risk analysis, joined ARA in 1999 as a Senior Engineer. Prior to joining ARA, Dr. Huang was a Research Assistant in Civil Engineering Department of Clemson University where he worked closely with Dr. Peter Sparks and Dr. David Rosowsky in the areas of wind engineering, risk assessment, and reliability analysis of residential buildings. Dr. Huang was the key researcher for the development of a GIS-

Applied Research Associates, Inc. 53 February 2002 Module 2 based hurricane hazard assessment system, including the development of a hurricane field model, damage model, and final software package. He also served as the Administrator of an Arc/Info server and assisted the establishment of the Laboratory for Emerging Technologies. Dr. Huang received his Ph.D. in Civil Engineering from Clemson University. Using information from reconnaissance aircraft, Doppler radar, data buoy and surface weather stations, Dr. Huang has performed near-real time hurricane wind field analyses for Hurricanes Bonnie (1998), Earl (1998), Georges (1998), Dennis (1999), and Floyd (1999). He also has extensive experience with historical hurricanes, such as Hugo (1989), Andrew (1992), Erin (1995), Opal (1995), Bertha (1996), and Fran (1996). His current responsibilities at ARA include analysis of wind and insurance data using geographical information systems (GIS), computer simulations of hurricane risk over the United States; development of predictive models for hurricane-induced structural damage; and development of predictive models for costs associated with hurricane- induced structural damage. Chris Driscoll, B.S., Civil Engineering, University of South Florida Mr. Driscoll joined ARA in 1999 as a Staff Engineer his bringing his 10 years of various experience in residential/commercial construction and computer programming. He earned his BS in Civil Engineering from the University of South Florida. He has been working on residential/commercial hurricane inspection and mitigation programs for the Department of Community Affairs, FEMA, and for small municipalities. Mr. Driscoll has evaluated post-hurricane wind damage surveys for Erin (1995), Opal (1995) and Georges (1998). He has performed building stock analysis using data collected from Florida Panhandle and South Florida building surveys. He is a developer of wind damage assessment computer modeling tools. Mr. Driscoll has three years experience in hurricane modeling and 4 years in structural engineering. Srinivas R. Kadasani, M.S., Environmental Engineering, Oklahoma State Univ. Mr. Kadasani, joined ARA’s wind group in July 1999. Mr. Kadasani graduated from Oklahoma State University (OSU) majoring in Environmental Engineering with emphasis in Computer Modeling. As a research assistant at OSU, Mr. Kadasani conducted experimental studies on ground water bio-remediation of TNT and p-DCB contaminated aquifers. As a part of his graduate thesis he developed a software plume model for TNT and DCB migration through different types of aquifers. His current responsibilities include engineering analysis, terrain modeling, and development of ARA’s portfolio analysis software. His areas of expertise include Object Oriented Design and Programming. Mr. Kadasani has approximately 3 years experience in the area of hurricane application modeling and software development. Francis M. Lavelle, Ph.D., Civil Engineering, Rice University Dr. Francis M. Lavelle is a Principal Engineer at Applied Research Associates with over eleven years of research, development, and project management experience in the areas of extreme loading effects and risk analysis of structures. Dr. Lavelle’s graduate research at Rice University included studies on the dynamic response of locally nonlinear structures, the seismic response of secondary systems, and retrofitting techniques for reducing earthquake-induced pounding damage in adjacent buildings. Dr. Lavelle is currently leading ARA’s software development efforts for the HAZUS Wind Loss Estimation Model. HAZUS is a joint effort between the Federal Emergency Management Agency (FEMA) and the National Institute of Building Sciences (NIBS) to produce a

Applied Research Associates, Inc. 54 February 2002 Module 2 nationally applicable software program for estimating potential losses from earthquake, flood, and wind hazards. He has been Principal Investigator on a number of government- sponsored research projects, including an $8.0M, 3-year contract involving both structural engineering research and software development. During the same period, he was also responsible for managing ARA’s Advanced Modeling and Software Systems (AMSS) group, a group of 15 civil, mechanical, aerospace, and software engineers. Before heading the AMSS group, Dr. Lavelle was responsible for developing reliability- based design safety factors for structural response to extreme loads. Dr. Lavelle is a member of the American Society of Civil Engineers and has been a registered professional engineer in the state of North Carolina since 1994. Marshall B. Hardy, M.S., Statistics, University of Kentucky Mr. Marshall B. Hardy has over eleven years experience in applied statistical analysis and probabilistic risk assessment. He has a Masters degree in Statistics from the University of Kentucky. He has approximately 5 years experience in support of hurricane modeling relevant to insurance loss analysis and rate making. Mr. Hardy performs multivariate statistical analysis and tests for ARA’s wind engineering applications. He has aided in the design of simulations, analyzed loss distributions, and performed multivariate analyses of hurricane damage and loss data sets. He has aided in the development of loss functions. He was a key developer of ARA’s TORSCR code that is now the nuclear power industry standard for PRA for extreme wind and tornado-missile hazard. He has applied the TORSCR code and statistically characterized distribution parameters for ten nuclear power plant PRAs. Mr. Hardy has also served as lead statistical analyst on numerous DoD, NSF, and EPA projects. This experience involved statistical modeling of uncertainty in nonlinear elasto-plastic response to wind, or multivariate analysis-of-variance (ANOVA) to identify dominant uncertainties in the predictive methodologies, and cluster analysis for code validation. Mr. Hardy is an expert in the SAS statistical analysis system that will be used in this work and is also an experienced FORTRAN programmer on a number of systems including the DNA Cray at Los Alamos, DEC minicomputers, and PCs. Richard W. Pearson, B.S., Computer Science, North Carolina State University Mr. Richard W. Pearson joined ARA’s wind group in July 2001. He brings with him 6 years of experience in software development and a background in object oriented programming, analysis, and design. Mr. Pearson graduated from North Carolina State University (NCSU) majoring in Computer Engineering with an emphasis on Software Architecture, Analysis, and Design. As a Software Engineer at Stingray Software, Mr. Pearson contributed to the design and development of the award-winning software tools suite, Objective Studio. During the same period, Mr. Pearson was a member of a research team tasked with identifying potential new markets and products for Stingray Software. Mr. Pearsons’ current responsibilities include the architecture and development of ARA’s portfolio analysis software. He has approximately 6 months experience in the application of hurricane modeling. B. Describe the credentials of the individuals or groups involved in the development of the following aspects of the model: 1. Meteorology Peter J. Vickery, Ph.D.

Applied Research Associates, Inc. 55 February 2002 Module 2 Peter F. Skerlj, M.E.Sc. Kevin Huang, Ph.D. Lawrence A. Twisdale, Jr. Ph.D., P.E. 2. Vulnerability Peter J. Vickery, Ph.D. Peter F. Skerlj, M.E.Sc. Jason Lin, Ph.D. Michael A. Young, M.E.Sc. Lawrence. A. Twisdale, Jr., Ph.D., P.E. 3. Actuarial Peter J. Vickery, Ph.D. Peter F. Skerlj, M.E.Sc. Marshall B. Hardy, M.S. Jason Lin, Ph.D. Srinivas R. Kadasani, M.S. Michael A. Young, M.E.Sc. Lawrence. A. Twisdale, Jr., Ph.D., P.E. 4. Computer Science Srinivas R. Kadasani, M.S. Francis M. Lavelle, Ph.D., P.E. Michael A. Young, M.E.Sc. Peter J. Vickery, Ph.D. Richard W. Pearson, B.S. 5. Statistics Peter J. Vickery, Ph.D. Peter F. Skerlj, M.E.Sc. Jason Lin, Ph.D. Marshall B. Hardy, M.S. Francis M. Lavelle, Ph.D., P.E. Lawrence. A. Twisdale, Jr., Ph.D., P.E. State whether these persons are full-time employees or outside consultants. All persons indicated above are full-time employees. 3. Multi-discipline Team (Standard 5.1.2 for all items in this section)

Applied Research Associates, Inc. 56 February 2002 Module 2 A. Indicate the different academic disciplines used to provide input and to construct your model. Wind Engineering Structural Engineering Meteorology Mathematics, Statistics Actuarial Science Computer Science B. Of the disciplines listed above, which are represented by current employees with your firm? Are other disciplines represented through consulting arrangements? The Wind Engineering, Structural Engineering, Mathematics and Statistics, and Computer Science are represented by full time ARA employees. The Meteorology discipline is represented through multiple peer reviews of published reports and papers on the ARA hurricane model. The Actuarial Science discipline is represented through consultants. C. Provide visual business workflow documentation connecting all personnel related to model design, testing, execution, and maintenance. Workflow documentation will be shown to the professional team. 4. List of Clients A. Provide a sample list of your clients in the following categories: for ratemaking, for reinsurance and capital markets, in 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 your ratemaking clients, how many clients have a U.S. aggregate annual property and casualty insurance premium of $100 million or more? Do any of your 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 your work.) ARA will provide a list of clients to the professional team for ratemaking, reinsurance, and government. B. Describe the present mix of your 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. Our hurricane catastrophe modeling/risk management work mix has been approximately: state government (35%), federal government (30%), insurers and reinsurers (20%), building owners/consultants, and wind tunnels (15%). This mix has been essentially the same for the past 3 years. C. How long have your ratemaking clients been clients of your company? As mentioned previously, we have not yet provided direct ratemaking support. 5. Independent Expert Review (Standard 5.1.2 for all items in this section) A. What independent peer reviews have been performed on the following parts of the model:

Applied Research Associates, Inc. 57 February 2002 Module 2 1. Meteorology The wind speed (windfield) model has been anonymously peer reviewed for publication in the Journal of Structural Engineering by experts in the fields of wind engineering and meteorology. The wind speed frequency model was anonymously reviewed at the same time the wind speed model was reviewed, also for publication in the Journal of Structural Engineering. The meteorological components have also been peer reviewed by the HAZUS Wind Committee. 2. Vulnerability The vulnerability function parameters have not been reviewed by personnel outside of ARA, except that the HAZUS Wind Committee has reviewed methodology and procedures used to develop the vulnerability functions. 3. Actuarial The actuarial components and resulting annualized loss costs have been reviewed by a consulting actuarial firm. 4. Computer Science None. 5. Statistics None, except through review of publications. 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. The peer reviewers for the journal articles describing the windfield and climatological modeling will be provided. The actuarial review team review letter will also be provided. C. Describe the nature of any on-going or functional relationship your company has with any of the persons performing the independent peer reviews. State which of the peer reviews described above were paid for by your firm and which were performed for no compensation. Describe any review by an independent organization, such as Standard and Poor’s, Moody’s, etc. None. The original actuarial review was paid for by our firm. D. Discuss any adversarial situations (such as a ratemaking hearing) in which your model was subjected to review. None.

Applied Research Associates, Inc. 58 February 2002 Module 2 MODULE 3 Module 3 - Section I Meteorology - Hurricane Set (Standards 5.2.3, 5.2.4, 5.2.6, 5.2.8, and 5.2.9 for all items in this section)

1. Define an “event” in your 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)? (Standard 5.2.5) An event is defined as: (i) any simulated hurricane that makes landfall in Florida, or (ii) a by-passing hurricane that produces peak gust wind speeds on land of at least 50 mph anywhere in Florida. The effect of changing the definition of an event on estimates of average annual loss will be shown to the professional team. The computation of loss occurs when the peak gust wind speed equals or exceeds 50 mph. 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 your model produces? (Standards 5.2.5 and 5.2.7) Category Maximum Sustained Wind Speed (mph) 1 95 2 110 3 130 4 155 5 >155 The maximum wind speeds produced by the model in each of the five categories are shown below. The maximum wind speeds of category 1 through 4 storms are based upon the Saffir-Simpson scale, and the maximum sustained wind speed associated with the category five storm has been derived from the maximum peak gust wind speed produced in the simulation of Florida hurricanes. 3. How does your model handle events with multiple landfalls? Are these defined as a single event or multiple events? How does this affect your frequency assumptions? (Standard 5.2.5) ARA’s hurricane model simulates the entire track of a hurricane or tropical storm, and thus storms with multiple landfalls are inherently included in the modeling process. Storms making multiple landfalls in Florida (including exiting storms) are defined as required in Module 3, Section 1, Question 12. 4. How does your model handle the definition of an event from the insurance policy perspective? In other words, does it 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? Although ARA’s hurricane model is able to compute the time of landfall, or time of loss, this information is not currently used, thus a single storm which makes multiple landfalls is treated as a single event.

Applied Research Associates, Inc. 59 February 2002 Module 3 5. Describe the hurricane tracks in your model. Discuss the appropriateness of the hurricane tracks used by your model. What historical data are your hurricane tracks based on? The modeling of the hurricane tracks in our model is described in detail in Vickery, et al. (2000b). The approach models a storm track beginning with its initial development over the ocean and ending with its final dissipation over land or in the open ocean. The approach has been validated as described in Vickery, et al. (2000b) through comparisons of simulated and observed key storm statistics along the coast of the United States. The development of the storm track model used the historical data given in the HURDAT database, encompassing the period 1886 through 2000. 6. Describe in detail the decay rates or hurricane degradation assumptions used by your 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? Once a simulated storm makes landfall, the central pressure difference, ∆p, is decreased as the storm fills (or weakens). The filling of the storm is modeled using the approach described in Vickery and Twisdale (1995a), where the central pressure difference at any time after landfall, ∆p(t), is given as: ∆ = ∆ − p(t) p0 exp( at)

where ∆po is the central pressure difference at the time of landfall and a is a filling parameter which is a function of the storm intensity at the time of landfall as well as the location (or region) of landfall and t is the time in hours since the center of the storm crossed land. Three different representations of the filling parameter, a, are used, one for the Gulf of Mexico, one for the Florida Peninsula and one for the Atlantic Coast. In performing studies for locations in the State of Florida only the first two filling model regions (different representations of the filling parameter, a) are applicable. In addition to the filling associated with the increase in the storm central pressure (or reduction in ∆p) the wind speed at ground level is immediately decreased at the coastline because of frictional effects. This immediate decrease in wind speed is followed by a gradual decrease in wind speed at the surface associated with a change in the boundary layer characteristics of the storm in the region of the eyewall. The reduction in wind speeds associated with changes in the hurricane wind field following landfall is described in Vickery, et al. (2000a). Additional changes in the characteristics of the hurricane following landfall include a decrease in the value of the Holland profile parameter, B, and an increase in the radius to maximum winds, Rmax, with both changes being brought about because both B and Rmax are correlated with central pressure and/or latitude, both of which continue to change after landfall. How far inland hurricane force winds penetrate given the land fall of a storm of different intensities will vary significantly with the storm characteristics. These storm characteristics include, translation speed, the Holland B parameter, Rmax, ∆po, orientation of the storm with respect to the coastline, air-sea temperature difference, and the location of landfall. In the example results presented in the following table, the following assumptions have been made:

Applied Research Associates, Inc. 60 February 2002 Module 3 (i) The initial wind speed represents the average value of the storm category computed for over water conditions (i.e. one minute average wind speed computed at a height of 10 meters above the water level).

(ii) The value of central pressure ∆po used to arrive at the above noted wind speeds are computed using the median value of Rmax associated with ∆po for a latitude of 28 degrees N, and the mean value of the Holland profile parameter, B, associated with the given values of Rmax and ∆po. (iii) The overland wind speeds used to define the hurricane intensity are based on the modeled one minute average wind speeds at a height of 10m above ground in open terrain. (iv) The air-sea temperature difference at the time of land fall is zero. (v) The storm is moving at a speed of 15 mph (a representative value for the Florida area). Table 4. Inland Distance of Hurricane Force Winds Distance of Hurricane Force Winds Inland (miles) Category South Florida Filling Model Gulf of Mexico Filling Model 1 40 31 2 219 115 3 259 161 4 288 184 5 288 184 7. Provide a graphical representation of the modeled degradation rates over time compared to the Kaplan-DeMaria decay rate and the +/- 20% range. Comparisons of degradation rates with Kaplan-DeMaria are given in Figure 19. Hurricane Fran - Wind Speed Reduction with Time Hurricane Erin - Wind Speed Reduction with Time

140 ) ) 120

-20% ph ph -20%

120 m m Kaplan-DeMaria Mean 100 Kaplan-DeMaria Mean (

20% d ( 20% 100

eed Fran Simulation ee Erin Simulation

p 80 p

S 80 S d d n

i 60 n 60 i d W

40 d W 40 ne i ne a i t

20 a s

t 20 u s S 0 u S 0 0 5 10 15 20 0 5 10 15 20 Time After Land Fall (hrs) Time After Land Fall (hrs)

Hurricane Alicia - Wind Speed Reduction with Time Hurricane Elena - Wind Speed Reduction with Time 140 140 ph) -20% ph) -20% m 120 m 120 Kaplan-DeMaria Mean Kaplan-DeMaria Mean 100 20% d ( 100 20% Alicia Simulation ee Elena Simulation peed ( p

80 S 80 d n nd S i i 60 60 W

d

e 40 ed W 40 n ai t ain

t 20 20 s s u u S

S 0 0 0 5 10 15 20 0 5 10 15 20 Time After Land Fall (hrs) Time After Land Fall (hrs) Figure 19. Example Comparisons of Modeled Degradation Rates to Kaplan-Demaria Decay Rates

Applied Research Associates, Inc. 61 February 2002 Module 3 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? Did you make any modifications to the data set? If so, describe in detail the modifications and their appropriateness. (Standards 5.4.14 and 5.6.1) The HURDAT data set encompassing the period 1886-2000 represents the prime source of historical information used to develop frequency distributions for storms affecting the United States and surrounding areas. The HURDAT database has been supplemented with information given in NWS-38 (Ho, et al.,1984). The prime source of information used to develop models for the radii to maximum winds is NWS-38 supplemented with data on more recent storms as given in the scientific literature. 9. Provide ranges for radius of maximum winds, radius of hurricane force winds and far field pressure used by your model for the central pressures provided in Table 5. (Standard 5.2.7) The radius of hurricane force wind is not a standard output of the model, however for the purposes of providing information on the radius to hurricane force winds, a limited number of simulations were performed for storms in the Florida region. Recall that both the radius to maximum winds and the pressure profile parameter are random variables that are correlated with one or more of central pressure and latitude, and further note that, given information on the central pressure, the radius to maximum wind and the Holland pressure profile parameter, the radius to hurricane force winds varies with the translation speed of the storm. Table 5 presents the results of simulations performed using the statistical models for Rmax and the Holland pressure profile parameter described earlier. For each central pressure value given in the following table (far field pressure = 1013 mbar in all cases), 1000 simulations were performed for a storm located at a latitude of 27 degrees North, with the translation speed sampled for a location in the mid point of the state. From each of the 1000 simulations for each central pressure value, unique values of Rmax and radius to hurricane force winds, RHur, were produced, yielding a total of 1000 estimates of Rmax and RHur for each value of central pressure. The median, mean, standard deviation and the 5th and 95th percentile values are shown below. In the case of the weaker storms, not all combinations of central pressure, B, etc. produce a storm that has hurricane force winds anywhere in the storm, and thus the last column in the table below indicates the percentage of simulated storms that yield maximum wind speeds of at least 74 mph (one minute average, over water).

Applied Research Associates, Inc. 62 February 2002 Module 3 Table 5. Simulated Wind Radius Statistics Central Radius of Maximum Wind (miles) Radius of Hurricane Force Winds (miles) Pressure Median Mean Std 5th 95th Median Mean Std 5th 95 % with (mbar) Dev Dev Hurricane Force Winds 900 13.1 14.1 5.4 6.9 24.8 48.7 52.2 19.8 25.6 87.4 100% 910 14.9 15.9 6.2 7.7 27.7 51.9 54.7 20.2 27.1 92.3 100% 920 16.3 17.7 7.2 8.6 32.4 51.6 56.3 22.3 27.1 98.5 100% 930 17.6 18.9 7.5 9.3 32.6 51.8 55.2 20.8 27.1 93.4 100% 940 19.4 20.7 8.0 10.0 35.6 50.1 53.2 19.3 26.2 89.2 100% 950 20.5 22.1 8.4 11.2 38.7 48.0 50.8 18.0 26.6 82.4 100% 955 21.0 22.8 8.8 11.4 40.7 45.8 49.0 18.1 24.9 83.6 98% 960 21.5 23.3 9.1 11.7 41.1 43.6 46.4 17.2 23.8 78.6 98% 965 21.7 23.8 9.6 12.0 41.7 40.5 44.2 17.3 22.7 77.3 96% 970 23.2 25.1 9.6 11.8 42.4 40.3 43.1 16.5 21.5 73.0 94% 975 23.8 25.7 9.9 11.7 41.8 38.2 40.5 15.9 19.2 70.4 89% 980 22.8 25.2 9.8 12.5 42.9 33.5 36.6 14.5 18.1 62.3 83% 985 23.5 26.0 9.9 11.9 43.4 31.9 34.2 13.7 15.7 60.2 68% 990 22.7 26.5 10.4 12.1 42.0 28.2 30.9 12.2 14.8 53.6 45% 10. Provide maps showing the maximum winds at the zip code level for the modeled 101 year historical storm set and also for a 101 year period from the stochastic storm set. Figure 20 shows the maximum wind speeds (one minute sustained at a height of 10m in open terrain, at the zip-code centroid) for the historical storm set. The historical storms were modeled such that the maximum sustained wind speed (10 m above ground, over water) at the time of landfall matches the maximum wind speed as given in the official storm set. If central pressure data is given in HURDAT, the central pressure as a function of time after landfall uses the HURDAT data. If central pressure data is not given in HURDAT, the central pressure after landfall is modeled using the filling models given in Vickery and Twisdale, (1995a). The radius to maximum winds is modeled using information obtained from NWS-38 or the Hurricane Research Division of NOAA. If no radius to maximum wind data is available, then Rmax is modeled using the regression model used in the stochastic storm modeling. Figure 21 presents an example of the maximum sustained windspeeds obtained from a random 101 year period taken from the full 100,000 year simulation.

Applied Research Associates, Inc. 63 February 2002 Module 3 Maximum Sustained (mph) < 40 40 - 50 50 - 60 60 - 70 70 - 80 80 - 90 90 - 100 100 - 110 110 - 120 120 - 130 130 - 140 140 - 150 150 - 160 160 - 170

Maximum Sustained Wind Speeds at Zip-code Centroid

Figure 20. Maximum One Minute Sustained Wind Speeds for the Historical Storm Set.

Sustained Wind Speed (mph) < 40 40 - 50 50 - 60 60 - 70 70 - 80 80 - 90 90 - 100 100 - 110 110 - 120 120 - 130 130 - 140 140 - 150 150 - 160 160 - 170

Maximum Simulated Sustained Wind Speed at Zip-code Centroid (Set 3)

Figure 21. Example of Maximum One Minute Sustained Wind Speeds From a Random 101 Year Period Taken From the Full 100,000 Year Stochastic Storm Set.

Applied Research Associates, Inc. 64 February 2002 Module 3 11. Provide frequency and annual occurrence rates from both the historical data set given and the data set that your 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 22. (Standard 5.4.14)

Figure 22. Region Definitions. Please see our response to Question 12. 12. Complete the Table 6 with modeled information for Florida in total and by region as defined in Figure 22. For each region, the column labeled “Hurricanes” is the number of hurricanes that made their initial landfall in that region. The Category is the category (by the Saffir- Simpson Hurricane Scale) of the hurricane at that landfall. The column labeled “Coastal X-ings” is 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 for the strength it was as 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 upon 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.

Applied Research Associates, Inc. 65 February 2002 Module 3 Table 6. Historical vs. Modeled Hurricane Frequencies Entire State of Florida Historical Modeled Historical Modeled Historical Modeled Cat Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings 1 25 37 31.4 44.3 44.6% 48.1% 54.7% 58.9% 0.248 0.366 0.311 0.438 2 12 18 12.8 16.0 21.4% 23.4% 22.3% 21.3% 0.119 0.178 0.127 0.158 3 14 17 8.8 10.2 25.0% 22.1% 15.3% 13.5% 0.139 0.168 0.087 0.101 4 4 4 3.5 3.8 7.1% 5.2% 6.0% 5.1% 0.040 0.040 0.034 0.038 5 1 1 1.0 1.0 1.8% 1.3% 1.7% 1.3% 0.010 0.010 0.009 0.010

Region A - Northwest Florida Historical Modeled Historical Modeled Historical Modeled Cat Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings 1 11 16 15.0 18.1 64.7% 66.7% 65.2% 67.5% 0.109 0.158 0.148 0.179 2 4 5 4.7 5.2 23.5% 20.8% 20.3% 19.2% 0.040 0.050 0.046 0.051 3 2 3 2.5 2.6 11.8% 12.5% 10.7% 9.9% 0.020 0.030 0.024 0.026 4 0 0 0.7 0.8 0.0% 0.0% 3.1% 2.8% 0.000 0.000 0.007 0.007 5 0 0 0.1 0.2 0.0% 0.0% 0.6% 0.6% 0.000 0.000 0.001 0.001

Region B - Southwest Florida Historical Modeled Historical Modeled Historical Modeled Cat Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings 1 8 9 7.4 10.1 50.0% 42.9% 51.7% 54.5% 0.079 0.089 0.074 0.100 2 2 4 3.2 4.1 12.5% 19.0% 22.6% 22.3% 0.020 0.040 0.032 0.041 3 4 6 2.4 2.9 25.0% 28.6% 16.7% 15.4% 0.040 0.059 0.024 0.028 4 1 1 1.0 1.1 6.3% 4.8% 7.0% 6.2% 0.010 0.010 0.010 0.011 5 1 1 0.3 0.3 6.3% 4.8% 2.0% 1.7% 0.010 0.010 0.003 0.003

Region C - Southeast Florida Historical Modeled Historical Modeled Historical Modeled Cat Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings 1 6 10 6.6 9.8 27.3% 37.0% 41.1% 47.3% 0.059 0.099 0.065 0.097 2 5 6 4.0 5.0 22.7% 22.2% 24.9% 23.9% 0.050 0.059 0.039 0.049 3 8 8 3.4 3.8 36.4% 29.6% 21.1% 18.3% 0.079 0.079 0.033 0.038 4 3 3 1.6 1.7 13.6% 11.1% 9.9% 8.1% 0.030 0.030 0.016 0.017 5 0 0 0.5 0.5 0.0% 0.0% 3.0% 2.4% 0.000 0.000 0.005 0.005

Region D - Northeast Florida Historical Modeled Historical Modeled Historical Modeled Cat Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings Hurr X-ings 1 0 2 2.4 6.2 0.0% 40.0% 59.3% 70.2% 0.000 0.020 0.024 0.061 2 1 3 0.9 1.7 100.0% 60.0% 22.3% 19.7% 0.010 0.030 0.009 0.017 3 0 0 0.5 0.9 0.0% 0.0% 13.4% 9.7% 0.000 0.000 0.005 0.008 4 0 0 0.2 0.0 0.0% 0.0% 4.1% 0.0% 0.000 0.000 0.002 0.000 5 0 0 0.0 0.0 0.0% 0.0% 0.8% 0.5% 0.000 0.000 0.000 0.000

By-Passing Storms Historical Model Historical Model Historical Model Cat Hurr Reg Hurr Reg Hurr Hurr Hurr Hurr 1 1 B 1.3 A,B,C 20.0% 13.0% 0.010 0.013 2 2C,C2.8All 40.0% 28.0% 0.020 0.028 3 1A3.3All 20.0% 33.0% 0.010 0.033 4 1B1.9All 20.0% 19.0% 0.010 0.019 5 0 0.7 A,B,C 0.0% 7.0% 0.000 0.007

Notes: 1. Number of hurricanes does not include by-passing storms. 2. Modeled by-passing storms include all hurricanes producing a hurricane force sustained windspeeds in Florida.

Applied Research Associates, Inc. 66 February 2002 Module 3 13. Complete Table 7 showing the Probability of Hurricanes by Year.

Table 7. Probability of Hurricanes by Year

NUMBER OF HURRICANES HISTORICAL MODELED PER YEAR PROBABILITY PROBABILITY 0.52 0 0.57 0.33 1 0.27 0.11 2 0.14 0.028 3 0.02 0.0055 4 0.00 0.00087 5 0.00 0.00016 6 0.00 0.00002 7 0.00 0 8 0.00 0 9 0.00 0 10 or more 0.00 Table 7 is based on the 100,000 year simulation and includes all hurricanes producing at least hurricane force winds at the Florida coast. 14. Complete Table 8 showing the Distribution of Hurricanes by Size. For the Expected Annual Hurricane Losses column, the modeler must present personal residential, zero deductible statewide loss costs based on the 1998 Florida Hurricane Catastrophe Fund’s (FHCF) aggregate exposure data found in the file named “hlpm.exe”. For the column, Return Time (Years) the modeler should indicate the return time associated with an average loss within the ranges indicated on a cumulative basis. For example, if the average loss is $4,705 million for the range $4,501 million to $5,000 million, we are looking for the return time associated with a loss that is $4,705 million or greater. For each range limit in millions ($1,001-$1,500, $1,501-$2,000, $2,001-$2,500) the average loss within that range will be identified and then the return time associated with that loss will be calculated. The return time is then the reciprocal of the probability of the loss equaling or exceeding this average loss size. The probability of equaling or exceeding the average of each range should be smaller as the ranges increase (and the average losses within the ranges increase). Therefore, the return time associated with each range and average loss within that range should be larger as the ranges increase. We are looking for return times based on cumulative probabilities. A return time for an average loss of $4,705 million within the $4,501-$5,000 million range should be lower than the return time for an average loss of $5,455 million associated with a $5,001- $6,000 million range.

Applied Research Associates, Inc. 67 February 2002 Module 3 Table 8. Distribution of Hurricanes by Size

EXPECTED ANNUAL HURRICANE LIMIT RANGE TOTAL LOSS AVERAGE NO. OF LOSSES* RETURN (Millions) (Millions) (Millions) STORMS (Millions) TIME (Years) $ - to $500 $5,928,696 $148 40024 $59.29 2.5 $501 to $1,000 $6,343,475 $714 8883 $63.43 3.7 $1,001 to $1,500 $5,860,412 $1,231 4760 $58.60 4.5 $1,501 to $2,000 $5,090,439 $1,733 2937 $50.90 5.2 $2,001 to $2,500 $5,072,685 $2,239 2265 $50.73 5.8 $2,501 to $3,000 $4,959,286 $2,739 1810 $49.59 6.4 $3,001 to $3,500 $4,690,889 $3,241 1447 $46.91 7.1 $3,501 to $4,000 $4,471,341 $3,751 1192 $44.71 7.7 $4,001 to $4,500 $4,200,833 $4,243 990 $42.01 8.3 $4,501 to $5,000 $4,111,509 $4,747 866 $41.12 8.9 $5,001 to $6,000 $7,723,583 $5,469 1412 $77.24 9.8 $6,001 to $7,000 $7,135,829 $6,493 1099 $71.36 11 $7,001 to $8,000 $6,662,932 $7,478 891 $66.63 12.3 $8,001 to $9,000 $6,592,124 $8,462 779 $65.92 13.5 $9,001 to $10,000 $6,199,558 $9,493 653 $62.00 14.9 $10,001 to $11,000 $6,106,885 $10,492 582 $61.07 16.3 $11,001 to $12,000 $5,117,719 $11,500 445 $51.18 17.7 $12,001 to $13,000 $4,795,905 $12,456 385 $47.96 18.9 $13,001 to $14,000 $4,971,960 $13,474 369 $49.72 20.3 $14,001 to $15,000 $4,230,779 $14,488 292 $42.31 21.7 $15,001 to $16,000 $4,239,253 $15,471 274 $42.39 23 $16,001 to $17,000 $4,242,932 $16,509 257 $42.43 24.4 $17,001 to $18,000 $4,228,377 $17,472 242 $42.28 26 $18,001 to $19,000 $3,628,155 $18,510 196 $36.28 27.5 $19,001 to $20,000 $3,588,746 $19,504 184 $35.89 28.9 $20,001 to $21,000 $3,230,465 $20,445 158 $32.30 30.4 $21,001 to $22,000 $4,191,967 $21,497 195 $41.92 32.1 $22,001 to $23,000 $3,374,390 $22,495 150 $33.74 33.9 $23,001 to $24,000 $2,747,579 $23,483 117 $27.48 35.5 $24,001 to $25,000 $3,158,951 $24,487 129 $31.59 37 $25,001 to $26,000 $3,138,763 $25,518 123 $31.39 38.8 $26,001 to $27,000 $3,017,636 $26,470 114 $30.18 40.6 $27,001 to $28,000 $2,807,611 $27,525 102 $28.08 42.4 $28,001 to $29,000 $3,303,313 $28,476 116 $33.03 44.5 $29,001 to $30,000 $2,798,611 $29,459 95 $27.99 46.7 $30,001 to $35,000 $13,716,827 $32,351 424 $137.17 53 $35,001 to $40,000 $11,476,769 $37,383 307 $114.77 65.4 $40,001 to $45,000 $10,156,531 $42,495 239 $101.57 79.1 $45,001 to $50,000 $9,116,010 $47,479 192 $91.16 95.2 $50,001 to $55,000 $9,297,457 $52,232 178 $92.97 115 $55,001 to $60,000 $7,823,712 $57,527 136 $78.24 140 $60,001 to $65,000 $6,925,829 $62,394 111 $69.26 170.3 $65,001 to $70,000 $5,996,899 $67,380 89 $59.97 202.9 $70,001 to $75,000 $5,147,490 $72,499 71 $51.47 243.8 $75,001 to $80,000 $5,079,526 $76,962 66 $50.80 296.4 $80,001 to $85,000 $5,371,157 $82,633 65 $53.71 361.5 $85,001 to $maximum $26,835,131 $109,085 246 $268.35 1087.5 Total $278,906,946 76657 $2,789.07 * Return Time represents the mean return period associated with a storm producing a loss equal to or greater than the average loss within a given limit range.

Applied Research Associates, Inc. 68 February 2002 Module 3

Module 3 - Section II

Hurricane Wind Field 1. What wind values (e.g., peak gust, maximum one-minute average sustained) and for what elevation is your wind field valid? Describe in detail the rationale for using the wind field chosen by your firm. ARA’s windfield model is valid for any averaging time and height above ground. The windfield model has been validated using both peak gust wind speeds and wind speeds averaged over ten minutes. In the validation studies, all measured wind speed data were adjusted (when needed) to a height of 10m above ground (in local terrain conditions) before wind speed comparisons were performed. The basic output of the model, for use in loss estimation, is the peak gust windspeed. 2. Do you need to convert the wind speeds generated in your wind field model to another form (i.e., from one-minute sustained to peak gust) for use by the vulnerability functions used by your model? If so, is there any accuracy lost by doing so? Describe in detail. (Standard 5.2.2) The damage and loss models are developed as a function of the peak gust wind speed which, as indicated above, is a basic output of the windfield model. 3. Is the duration of wind speeds at a particular location over the life of a hurricane considered in the model? If so, at what point (or wind speed level) is the damage ratio estimated for wind speeds at a location? Does your model take into consideration both damage caused by gusts of wind and damage caused by sustained winds at perhaps a lower wind speed level? Describe your answers in as much detail as possible. The detailed (loss) functions produced by the damage model include the effects of the complete wind trace, including directionality, and gusts over the duration of the storm. The damage and loss models are initiated when the peak gust wind speeds exceed 50 mph.

Applied Research Associates, Inc. 69 February 2002 Module 3 Module 3 - Section III

Vulnerability Functions Damage Estimates (Standards 5.3.1, 5.3.2, 5.3.3, and 5.3.4 for all items in this section)

1. At what one-minute average sustained wind speed does your model begin estimating loss? The damage model is initiated when the peak gust wind speed at a height of 10 meters in open terrain exceeds 50 mph. This peak gust value corresponds to a sustained wind speed (one minute average wind speed at a height of 10 meters in open terrain) of about 40 mph. 2. Describe in detail how socio-economic effects are considered (if at all) within your model. Is this applied to every event in your model or limited to select events? If for only select events, how are they selected? If this is not considered directly in your model but only at the request of the insurance company, describe your procedure for including this in the loss estimates. Describe the validation procedures to verify the results. (Standards 5.4.3 and 5.4.5) These effects are not considered in a standard run of the model. 3. Describe in detail how building code enforcement is considered (if at all) within your model. If this is not considered directly in your model but only at the request of the insurance company, describe your procedure for including this in the loss estimates. Describe the validation procedures to verify the results. (Standards 5.3.5, 5.4.2, and 5.4.11) Building code enforcement is not considered per se in the model. Specific construction characteristics are treated by evaluating the effects through the building performance model. The building performance model is run with specific building class inputs to generate a loss function for that class (e.g., hip, hurricane straps, no shutters, etc.). 4. Describe in detail how quality of construction type, materials and workmanship are considered (if at all) within your model. If this is not considered directly in your model but only at the request of the insurance company, describe your procedure for including this in the loss estimates. Describe the validation procedures to verify the results. (Standard 5.6.2) If required, quality of construction is treated through the application of resistance factors to the nominal resistance of various building components. Validations have been performed through comparisons of modeled and observed physical damage states. 5. Describe in detail how the presence of fixtures or construction techniques designed for hazard mitigation are considered (if at all) within your model. If this is not considered directly in your model but only at the request of the insurance company, describe your procedure

Applied Research Associates, Inc. 70 February 2002 Module 3 for including this in the loss estimates. Describe the validation procedures to verify the results. (Standard 5.6.2) Since its inception in 1995, the ARA damage and loss model has been developed using a load and resistance based approach that produced physical damage to a building through the application of directionally dependent wind loads applied to individual building components including such components as roof cover, roof sheathing, roof-wall connections, windows, doors, garage doors etc. Information on the resistance of various components have been obtained from a range of sources including laboratory tests, engineering analyses, and in some cases judgement based on the results of post storm damage surveys. In addition to wind loads acting on buildings during a storm, the effect of windborne debris is included using ARA’s first principals based wind borne debris trajectory model. This windborne debris model was used to help define the design criteria for missile protection in ASTM and, to an extent, in the SBC. Given the physical damage to a building, the financial damage, or loss, is then estimated using a restoration based cost model that has been developed to estimate the repair/replacement cost of the building (and contents). Because the ARA model is a first principals based load and resistance model, mitigation is readily treated by increasing the resistances of one or more of the building components and performing a new damage and loss study. The overall methodology used to model mitigation has been independently reviewed by the HAZUS wind committee. The approach has been validated through comparisons of modeled and observed building physical damage obtained from post storm states. Details of the model, as well as the results of the comparison studies will be presented to the professional team. 6. Describe in detail your “unknown” vulnerability curve used for unknown residential construction types. If you use a composite of other vulnerability functions, describe how they are derived. Cite the documentation or describe the data used as a basis for this curve. (Standard 5.3.5) The unknown vulnerability curve varies with region in the state of Florida. The vulnerability curve is a composite that reproduces the proportion of each construction type which is prevalent in a given region. The unknown vulnerability curves do not include mobile homes.

Applied Research Associates, Inc. 71 February 2002 Module 3

Module 3 - Section IV

Insurance Functions Company Loss Estimates (Standards 5.4.1 and 5.4.4 for all items in this section)

1. A given wind speed can produce a variety of damage within a given zip code. For example, a 10% average damage ratio could result from a wide variety of damages ranging from no damage up to moderate damage. Some properties may have losses that are entirely below the deductible so that total insured losses in the zip code are well below 10%. In a similar manner for more severe wind speeds, some properties within a zip code could have damages in excess of policy limits. How does your model handle this problem? (Standard 5.4.8) Ground-up losses are estimated on a per policy basis, where for a given wind event the full distribution of possible losses is generated. The deductibles are applied on a policy-by-policy basis, and all policies having estimated losses less than the value of the deductible are reported as having net losses of zero. 2. Provide an example of how insurer loss (loss net of deductibles) is calculated. Discuss data or documentation used to confirm or validate the method used by your model. (Standard 5.4.8) Table 9 is representative of a typical one story house.

Table 9. Example of Insurer Loss Calculation Building Policy Mean Loss Average Loss Average Loss Net Value Limit Deductible Ratio (Zero Deductible) of Deductible $100,000.00 $90,000.00 $500.00 2.0% $2000.00 $1712.00 For any given windspeed, the damage is comprised of a wide range of losses ranging between 0 and 100%. To compute the mean loss for a given windspeed, each possible outcome of ground-up loss derived from the building performance model is used and the deductible is subtracted from this total loss to compute the net loss. The mean loss, including the effect of deductible, is then obtained by averaging each of the computed losses. The example paid losses given above change with the characteristics of the building. 3. Describe in detail the approach used for the appurtenant structures vulnerability function (if it is a unique function). How is it dependent on the building function? Provide documentation of validation test results to verify the approach used. (Standards 5.3.1, 5.3.2, and 5.6.2) Our current version of the appurtenant structures vulnerability function is modeled as a ratio of the building vulnerability. This simplistic function is used since: (1) the true value and type of appurtenant structure is usually unknown, (2) the contribution of the appurtenant coverage to the AAL is small, and thus errors introduced through this

Applied Research Associates, Inc. 72 February 2002 Module 3 approximation is small, and (3) insurer documentation of appurtenant structure loss is usually poor. Example comparisons of modeled and observed appurtenant structure losses for those cases where reliable data are available will be shown to the professional team. 4. Describe in detail the approach used for the mobile home vulnerability function. How is it dependent upon other building functions and are there separate mobile home vulnerability functions? Provide documentation of validation test results to verify the approach used. (Standards 5.3.1 and 5.6.2) The mobile home vulnerability functions were developed separately from other building vulnerability functions, but using the same overall approach. The structure damage functions were primarily developed using the approaches and data provided in Marshall (1993, 1994), Marshall and Yokel (1995), Vasquez (1994), McDonald and Vann (1986), Vann and McDonald (1978). The loss functions were developed independently by ARA staff to arrive at the final vulnerability functions. The model has been validated using loss data from insurers. 5. Describe in detail the approach used for the contents vulnerability function. How is it dependent on the building function (e.g., is it a function of building loss or other aspect)? Is there a minimum threshold at which loss is calculated (e.g., loss is estimated when the building damage exceeds 20%)? Provide documentation of validation test results to verify the approach used. (Standards 5.3.1, 5.3.6, 5.4.9, and 5.6.2) The model used to estimate the vulnerability of contents is based on the physical damage, and the resulting possibility of wind and water entering the building following damage. Thus, while the damage to contents is a function of the damage to the building, the model is constructed in such a way that damage to contents does not occur until sufficient physical damage to the building has occurred to allow wind and/or water to enter the building causing damage to the contents. The content model has been validated/calibrated separately from the building vulnerability model. 6. Describe in detail the approach used for the time element vulnerability function. Does it consider both direct and indirect loss to the building? For example, direct loss is for expenses paid to house policyholders in an apartment while their home is being repaired. Indirect loss is for expenses incurred (e.g., food spoilage) for loss of power, heat, etc. Is there a minimum threshold at which loss is calculated (e.g., loss is estimated for building damage greater than 20% or only for category 3, 4, 5 events)? Provide documentation of validation test results to verify the approach used. (Standards 5.3.1, 5.3.6, 5.4.10, and 5.6.2) An additional living expense model was developed using a time restoration model that is used to estimate the time homeowners are unable to live in the damaged structure. The model allows for ALE losses to be incurred due to infrastructure damage caused by storm surge and waves. The model has been calibrated through comparisons of actual insurance losses as described in the general standards. 7. Some policies, particularly for contents coverage, provide for indemnity on an actual cash value basis. Identify depreciation assumptions and

Applied Research Associates, Inc. 73 February 2002 Module 3 describe in detail the methods and assumptions used to reduce insured losses on account of depreciation. Provide a sample calculation for determining the amount of depreciation and the ACV losses. (Standard 5.4.9) The loss costs estimates do not include any assumptions with respect to depreciation. The validation of the model with actual insurance data indicates that this assumption is valid. 8. Some policies cover losses that exceed the amount of insurance. Identify property value assumptions and describe in detail the methods and assumptions used to determine the true property value and associated losses. Provide a sample calculation for determining the property value and guaranteed replacement cost losses. The loss model assumes that the insured value is equal to the replacement or actual value of the insured property and contents. Unless specified with separate documentation on replacement value, no provision is made in the loss costs analysis for cases where the replacement value exceeds the insured value. 9. Provide five (5) validation comparisons of actual exposures and loss to modeled exposures and loss. These comparisons must be provided by line of insurance, construction type, policy coverage, county or other level of similar detail in addition to total losses. Include not only the loss estimates, but also loss as a percent of total exposure as well. Total exposure represents the total amount of insured values (all coverages combined) in the area affected by the hurricane. This would include exposures for policies that did not have a loss. If this is not available, provide exposures for only these policies that had a loss. Specify which is used. Also, specify the name of the hurricane event compared. (Standard 5.4.13) Five comparisons of actual and modeled losses are provided in Table 10.

Table 10. Validation Comparisons of Actual and Modeled Losses

Comparison Number 1 Total Losses by County - Hurricane Bonnie Actual Loss Modeled Loss County Exposure Loss Loss/Exposure Exposure Loss Loss/Exposure Brunswick 64,968 368 0.0057 64,968 122 0.0019 Carteret 57,860 348 0.0060 57,860 239 0.0041 Duplin 18,756 29 0.0016 18,756 24 0.0013 New Hanover 211,080 831 0.0039 211,080 1,000 0.0047 Onslow 76,489 314 0.0041 76,489 172 0.0023

Applied Research Associates, Inc. 74 February 2002 Module 3 Table 10 (continued). Validation Comparisons of Actual and Modeled Losses

Comparison Number 2 Hurricane Andrew - Losses by Construction Class Actual Loss Modeled Loss Construction Type Exposure Loss Loss/Exposure Exposure Loss Loss/Exposure Masonry 3,261,687 592,301 0.1816 3,261,687 534,877 0.1640 Wood 52,745 15,396 0.2919 52,745 11,840 0.2245

Comparison Number 3 Comparison by Coverage (Hurricane Andrew) Actual Loss Modeled Loss Coverage Exposure Loss Loss/Exposure Exposure Loss Loss/Exposure Coverage A 2,677,015 649,842 0.2427 2,677,015 593,301 0.2216 Coverage C 1,877,721 275,480 0.1467 1,877,721 207,559 0.1105 Coverage D 535,403 74,678 0.1395 535,403 60,929 0.1138

Comparison Number 4 Comparisons of Total Loss by Storm and Company Company/ Actual Loss Modeled Loss Storm Exposure Loss Loss/Exposure Exposure Loss Loss/Exposure Company A Andrew 3,342,905 611,347 0.1829 3,342,905 558,011 0.1669 Company B Andrew 5,357,841 1,000,000 0.1866 5,357,841 921,119 0.1719 Company C Andrew 467,764 50,002 0.1069 467,764 58,130 0.1243 Company A Bertha 7,157,836 3,117 0.0004 7,157,836 3,856 0.0005 Company A Opal 17,091,084 15,560 0.0009 17,091,084 28,147 0.0016

Comparison Number 5 Comparison by Line of Business Actual Loss Modeled Loss Line of Business/Storm Exposure Loss Loss/Exposure Exposure Loss Loss/Exposure Rental Andrew 73,309 4,854 0.0662 73,309 3,902 0.0532 Homeowner Andrew 3,342,905 611,347 0.1829 3,342,905 558,011 0.1669 Condominium Andrew 434,015 9,596 0.0221 434,015 10,166 0.0234 Mobile Home Georges 346,183 873 0.0025 346,183 603 0.0017

Applied Research Associates, Inc. 75 February 2002 Module 3 10. Disclose, in a model output report, the specific type of input which is required of insurers in order to use the model or model output in a personal residential property insurance rate filing. Such input includes, but is not limited to, optional features of the model, type of data to be supplied by the insurer and needed to derive loss estimates from the model, and any variables which a licensed user is authorized to set in implementing the model.

Include a copy of the input form used by insurers or others to provide input criteria to be used in the model. The information contained in the input form shall be incorporated into the model using standard actuarial and scientific methods. Methods used to include data provided in the input forms shall be disclosed and shall be based upon accepted actuarial and other disciplinary procedures.

All modifications, adjustments, assumptions, defaults, and treatments of missing values shall be fully disclosed in a form to be included with the model output. Modeler shall include in its submission, the output form that discloses any and all modifications, adjustments, assumptions, or other criteria that are included in producing the model output.

Demonstrate that the input form relates directly to the model output. As indicated elsewhere, include the model version number on the forms.

A sample “output form” is given in Figure 23. There is no “input form” since HurLoss import schemas are customized to each insurer’s data file format.

Applied Research Associates, Inc. 76 February 2002 Module 3 ARA HURLOSS MODEL STUDY DISCLOSURE SUMMARY

Input/Output Element Disclosure Discussion

Model Version: HURLOSS 2.0 Model Approved by Yes, the model used was approved by the FCHLPM FCHLPM? for the 2000 Standards on May 11, 2001. Model Options Used: None Company: Purpose of Study: Evaluate loss costs and loss costs relativities for different house construction features. Perform sensitivity analyses for loss costs relativities for multiple locations in Florida. Period of Analysis: January-February, 2001 ARA Analysts: M. Young, L. Twisdale, P. Vickery Company Supplied Data: 1. General guidance on classification variables of interest 2. Locations in Florida 3. Coverages to be considered 4. Deductible levels of 0, 2, and 5% as part of a sensitivity study 5. Contents limit is 50% of A limit 6. ALE limit is 20% of Coverage A limit Company Loss Data Analyzed: None Lines of Business: Homeowners (Single family residential only) Coverages: A, C, and D and Total (no appurtenant structures) Deductibles: 0, 2, and 5% Contents: Coverage C limit = 50% of Coverage A limit. ALE: Coverage D limit = 20% of Coverage A limit Company Exposure Analyzed? No, not applicable to this sensitivity study. Demand Surge Costs Included? No, demand surge costs are not a standard output of the model. Loss aggregation: NA. The results were not aggregated. Exposure by Zip Codes: NA. Exposure information was not used in this study. Insurance to value: Replacement value assumed to be equal to insured limit. No separate insurance-to-value assumptions have been made in this study.

Figure 23. Model Disclosure Summary

Applied Research Associates, Inc. 77 February 2002 Module 3 Module 3 - Section V

Average Annual Loss Functions Loss Costs (Standard 5.4.3 for all items in this section)

1. Provide copies of documentation and reports available to the insurer to be used to analyze loss costs or as supporting documentation in rate filings. Draft copies of the documents will be made available to the professional team. 2. In responding to the following questions, demonstrate that the results of the model are reasonably consistent with observed insurance data and other scientifically based observations. Where appropriate, explain possible inconsistencies. Document data sources. (Standards 5.4.7, 5.4.13, and 5.4.14) • Demonstrate that loss cost relationships by type of coverage (buildings, appurtenant structures, contents, time element) are consistent with actual insurance data. • Demonstrate that loss cost relationships by construction type or vulnerability function (frame, masonry, brick, mobile home, etc.) are consistent with actual insurance data. • Demonstrate that loss cost relationships between territories or regions are consistent and reasonable. Loss cost relationships by coverage and building type are compared to actual data as given in Module 3, Section IV, Question 9. The hurricane risk map for Florida derived using our peer-reviewed hurricane risk model yields 50 and 100 year return period wind speeds along the US coastline similar to results of previous studies such as (Georgiou, 1985). A comparison of the loss costs for wood, masonry, and mobile homes, shown in Figures 24 through 26, shows the predicted loss costs vary with the hurricane risk, with a maximum in South Florida where the hurricane risk is highest, to a minimum away from the coast, in North Florida where the risk is lowest. Note that the loss costs include the effects of local terrain while the windspeed map is given for open terrain. 3. Provide copies of thematic maps (with a minimum of 6 value ranges) displaying zero deductible loss costs by 5-digit zip code for frame, masonry, and mobile home. (Standard 5.1.7) The required maps are shown in Figures 24 through 26.

Applied Research Associates, Inc. 78 February 2002 Module 3 Wood Frame 0.1 - 0.5 0.5 - 1 1 - 3 3 - 5 5 - 8 8 - 12 12 - 16 16 - 20 20 - 25 No Data

Figure 24. Ground-Up Loss Cost for Wood Frame Houses.

Masonry 0.1 - 0.5 0.5 - 1 1 - 3 3 - 5 5 - 8 8 - 12 12 - 16 16 - 20 20 - 25 No Data

Figure 25. Ground-Up Loss Cost for Masonry Wall Houses.

Applied Research Associates, Inc. 79 February 2002 Module 3 Mobile Home 0.1 - 0.5 0.5 - 1 1 - 3 3 - 5 5 - 8 8 - 12 12 - 16 16 - 20 20 - 25 25 - 35 No Data

Figure 26. Ground-Up Loss Cost for Mobile Homes. 4. The modeling company shall provide to the Commission output ranges in the format shown in the file named “2001OutPut.xls” on the enclosed CD-ROM. A hard copy of the output range spreadsheets shall be included with the submission and shall appear as indicated, at the end of Module 3, Section VII, Form F. The company shall also provide the output ranges on CD-ROM in both an Excel and a PDF format as specified. The file name shall include the abbreviated name of the modeler and the Standards year. (Standard 5.4.15) Loss costs shall be provided by county in a format adopted by the Commission. Within each county, loss costs shall be shown separately per $1,000 of exposure for personal residential, renters, condos, and mobile home; for each major deductible option; and by construction type. For each of these categories using zip code centroids, the output range shall show the highest loss cost, the lowest loss cost, and the weighted average loss cost based on the Florida Hurricane Catastrophe Fund (FHCF) aggregate exposure data provided to each modeler in the file named “hlpm.exe” on the enclosed CD-ROM. A file named “99 FHCF Wts.xls” has also been provided to be used to determine the weighted average loss costs. Include the statewide range of loss costs (i.e., low, high, and weighted average). For each of the loss costs provided by the modeling company, the company shall identify what that loss cost represents by line of business, deductible option, construction type, and coverages included, i.e., structure, contents, appurtenant structures, or additional living expenses as specified in the format in the

Applied Research Associates, Inc. 80 February 2002 Module 3 file named “2001OutPut.xls” on the supplied CD-ROM. (Standard 5.4.14) Loss costs are provided on the included CD-ROM in the files named ARA_2001OUTPUT.xls and ARA_2001OUTPUT.pdf. 5. Include an explanation of the differences between the prior year and the current year submission (if applicable). (Standard 5.4.15) NOTE: If a modeler has loss costs for a zip code for which there is no exposure, then the modeler should give the loss costs zero weight (i.e., assume the exposure in that zip code is zero). The modeler should provide a list of those zip codes where this happens. If the modeler does not have loss costs for a zip code for which there is some exposure, the modeler should not assume such loss costs are zero. Instead, the modeler should use only those exposures for which it has loss costs in calculating the weighted average loss costs. The modeler should provide a list of those zip codes where the modeler does not have loss costs for a zip code for which there is some exposure. (Standard 5.4.7) Differences between the prior year and current year submission include the following: (i) Zip Codes up-dated to the year 2001. (ii) Population weighted centroids have replaced geographic centroids. (iii) Minor changes to the hurricane hazard model to include storms through 2001. (iv) A significant change in the default building stock for Florida. 6. Provide the monetary contribution to the average annual personal residential zero deductible statewide loss costs from each specific storm in the Official Storm Set. Provide the contribution from Hurricane Andrew for each affected zip code. Monetary per storm contributions to the average annual personal lines residential zero deductible statewide loss costs are given in Table 11. The contribution from Hurricane Andrew for each affected zip code (defined here as a 1% or more contribution) is given in Table 12. Zip codes in Table 12 are taken from the 1513 zip codes provided by the FCHLMP.

Applied Research Associates, Inc. 81 February 2002 Module 3

Table 11. Historical Per Storm Ground-up Losses

Year Name Loss Year Name Loss 1903 NONAME 3 $5,513,525,066 1946 NONAME 5 $115,455,946 1906 NONAME 2 $1,701,389,857 1947 NONAME 4 $19,808,396,139 1906 NONAME 8 $6,871,552,494 1947 NONAME 8 $786,349,035 1909 NONAME 9 $123,820,778 1948 NONAME 7 $3,061,390,566 1910 NONAME 4 $5,389,438,625 1948 NONAME 8 $54,751,164 1911 NONAME 1 $94,875,394 1949 NONAME 2 $10,433,347,202 1915 NONAME 4 $29,753,147 1950 Easy $3,481,301,127 1916 NONAME 13 $1,063,264,793 1950 King $5,014,200,139 1916 NONAME 14 $30,396,474 1953 Florence $225,578,828 1917 NONAME 3 $923,970,646 1956 Flossy $255,986,976 1919 NONAME 2 $523,996,040 1960 Donna $10,782,346,942 1921 NONAME 6 $4,847,426,934 1964 Cleo $2,951,101,568 1924 NONAME 4 $3,956,770 1964 Dora $1,703,952,277 1924 NONAME 7 $1,208,169,675 1964 Isbell $2,212,216,779 1925 NONAME 2 $27,752,155 1965 Betsy $1,513,516,465 1926 NONAME 10 $107,255,357 1966 Alma $247,781,255 1926 NONAME 1 $3,500,972,720 1966 Inez $7,687,827 1926 NONAME 6 $36,585,577,162 1968 Gladys $391,576,434 1928 NONAME 1 $2,497,383,535 1972 Agnes $19,751,424 1928 NONAME 4 $29,247,673,555 1975 Eloise $298,454,062 1929 NONAME 2 $4,348,338,624 1979 David $3,964,185,197 1933 NONAME 5 $217,250,189 1979 Frederic $10,670,950 1933 NONAME 12 $21,179,788,454 1985 Elena $73,228,197 1935 NONAME 2 $2,374,605,460 1985 Kate $153,315,917 1935 NONAME 4 $154,435,429 1987 Floyd $13,093,902 1936 NONAME 5 $282,285,554 1992 Andrew $9,345,360,053 1939 NONAME 2 $213,739,873 1995 Erin $436,112,434 1941 NONAME 5 $11,641,441,727 1995 Opal $790,123,070 1944 NONAME 11 $13,845,461,353 1998 Earl $8,400,355 1945 NONAME 1 $331,016,142 1998 Georges $37,398,137 1945 NONAME 9 $5,945,062,749 1999 Irene $125,935,439

Applied Research Associates, Inc. 82 February 2002 Module 3

Table 12. Percentage Contribution of Hurricane Andrew to Historical Annualized Loss by Zip Code Zipcode FIPS Zipcode FIPS Zipcode FIPS Zipcode FIPS 33008 11 1.10% 33109 25 4.80% 33156 25 24.10% 33199 25 8.40% 33009 11 1.00% 33110 25 2.30% 33157 25 36.00% 33231 25 8.20% 33023 11 1.10% 33111 25 8.00% 33158 25 32.70% 33233 25 10.40% 33025 11 1.10% 33114 25 8.80% 33159 25 6.50% 33234 25 7.90% 33027 11 1.00% 33116 25 28.80% 33160 25 1.50% 33238 25 4.00% 33083 11 1.10% 33119 25 6.30% 33161 25 2.60% 33239 25 1.40% 33925 21 1.90% 33121 25 6.60% 33162 25 1.80% 33242 25 6.20% 33926 21 2.40% 33122 25 6.30% 33163 25 1.30% 33243 25 16.30% 33929 21 2.50% 33124 25 2.40% 33164 25 1.80% 33245 25 9.50% 33969 21 2.10% 33125 25 7.70% 33165 25 10.30% 33247 25 5.90% 34138 21 2.30% 33126 25 7.40% 33166 25 5.60% 33255 25 9.80% 34139 21 2.40% 33127 25 5.70% 33167 25 3.00% 33256 25 23.60% 34141 21 2.30% 33128 25 7.70% 33168 25 2.90% 33257 25 36.60% 34145 21 2.10% 33129 25 8.10% 33169 25 1.70% 33261 25 2.40% 34146 21 2.30% 33130 25 8.10% 33170 25 34.20% 33265 25 10.20% 33002 25 3.00% 33131 25 6.60% 33172 25 7.10% 33266 25 5.70% 33010 25 5.40% 33132 25 5.00% 33173 25 16.80% 33269 25 1.80% 33011 25 5.40% 33133 25 10.80% 33174 25 8.40% 33280 25 1.40% 33012 25 3.90% 33134 25 8.70% 33175 25 10.30% 33283 25 16.80% 33013 25 4.10% 33135 25 8.50% 33176 25 28.00% 33296 25 25.20% 33014 25 2.90% 33136 25 6.90% 33177 25 34.60% 33943 25 2.90% 33015 25 1.90% 33137 25 5.30% 33178 25 5.60% 33037 87 2.70% 33016 25 3.30% 33138 25 3.80% 33179 25 1.40% 33017 25 1.90% 33139 25 6.40% 33180 25 1.20% 33018 25 3.10% 33140 25 4.60% 33181 25 2.30% 33030 25 35.20% 33141 25 3.30% 33182 25 7.30% 33031 25 34.00% 33142 25 6.10% 33183 25 15.90% 33032 25 33.00% 33143 25 16.20% 33184 25 8.40% 33033 25 33.40% 33144 25 8.30% 33185 25 11.80% 33034 25 33.80% 33145 25 9.30% 33186 25 23.80% 33035 25 37.30% 33146 25 12.20% 33187 25 31.10% 33039 25 35.40% 33147 25 4.30% 33188 25 5.50% 33054 25 2.60% 33148 25 8.70% 33189 25 35.90% 33055 25 1.80% 33149 25 17.10% 33190 25 34.60% 33056 25 1.80% 33150 25 4.10% 33192 25 7.30% 33090 25 36.10% 33151 25 5.40% 33193 25 16.20% 33092 25 33.00% 33152 25 5.90% 33194 25 8.20% 33101 25 7.30% 33153 25 3.80% 33195 25 7.40% 33102 25 6.10% 33154 25 2.50% 33196 25 24.50% 33107 25 6.50% 33155 25 9.90% 33197 25 35.20%

Applied Research Associates, Inc. 83 February 2002 Module 3 Output Range Specifications “Owners” Policy Type

Coverage A: Structure

• Coverage A: Amount of Insurance = $100,000 • Replacement Cost Included Subject to Coverage “A” Limit • Ordinance or Law Not Included

Coverage B: Appurtenant Structures

• Amount of Insurance = 10% of Coverage “A” Amount • Replacement Cost Included Subject to Coverage “B” Limit • Ordinance or Law not Included

Coverage C: Contents

• Amount of Insurance = 50% of Coverage “A” Amount • Replacement Cost Included Subject to Coverage “C” Limit

Coverage D: Additional Living Expense

• Amount of Insurance = 20% of Coverage “A” Amount • Time Limit = 12 Months

Loss Costs per $1,000 should be related to the Coverage “A” Amount.

For weighting the Coverage “D” Loss Costs, use the file named “99 FHCF Wts.xls” for distribution for Coverage “A”.

Loss Costs for the various deductibles should be determined based on “per occurrence” deductibles.

Explain any deviations and differences from the prescribed format above.

Specify the model and program version numbers reflecting the release date as a footnote on each page of the output. There are no deviations from the prescribed format.

Applied Research Associates, Inc. 84 February 2002 Module 3 Output Range Specifications “Renters” Policy Type

Coverage C: Contents

• Amount of Insurance = $25,000 • Replacement Cost Included Subject to Coverage “C” Limit

Coverage D: Additional Living Expense

• Amount of Insurance = 40% of Coverage “C” Amount • Time Limit = 12 Months

Loss Costs per $1,000 should be related to the Coverage “C” Amount.

For weighting the Coverage “D” Loss Costs, use the file named “99 FHCF Wts.xls” for distribution for Coverage “C”.

Loss Costs for the various deductibles should be determined based on “per occurrence” deductibles.

For weighting the Coverage “C” Loss Costs, use the file named “99 FHCF Wts.xls” for distribution for Coverage “C”.

Explain any deviations and differences from the prescribed format above.

Specify the model and program version numbers reflecting the release date as a footnote on each page of the output. There are no deviations from the prescribed format.

Applied Research Associates, Inc. 85 February 2002 Module 3 Output Range Specifications “Condo Unit Owners” Policy Type

Coverage A: Structure

• Amount of Insurance = 10% of Coverage “C” Amount • Replacement Cost Included Subject to Coverage “A” Limit

Coverage C: Contents

• Amount of Insurance = $50,000 • Replacement Cost Included Subject to Coverage “C” Limit

Coverage D: Additional Living Expense

• Amount of Insurance = 40% of Coverage “C” Amount • Time Limit = 12 Months

Loss Costs per $1,000 should be related to the Coverage “C” Amount.

For weighting the Coverage “D” Loss Costs, use the file named “99 FHCF Wts.xls” for distribution for Coverage “C”.

Loss Costs for the various deductibles should be determined based on “per occurrence” deductibles.

For weighting the Coverage “C” Loss Costs, use the file named “99 FHCF Wts.xls” for distribution for Coverage “C”.

Explain any deviations and differences from the prescribed format above.

Specify the model and program version numbers reflecting the release date as a footnote on each page of the output. There are no deviations from the prescribed format.

Applied Research Associates, Inc. 86 February 2002 Module 3 Output Range Specifications “Mobile Home Owners” Policy Type

Coverage A: Structure

• Coverage “A” Amount of Insurance = $50,000 • Replacement Cost Included Subject to Coverage “A” Limit

Coverage B: Appurtenant Structures

• Amount of Insurance = 10% of Coverage “A” Amount • Replacement Cost Included Subject to Coverage “B” Limit

Coverage C: Contents

• Amount of Insurance = 50% of Coverage “A” Amount • Replacement Cost Included Subject to Coverage “C” Limit

Coverage D: Additional Living Expense

• Amount of Insurance = 20% of Coverage “A” Amount • Time Limit = 12 Months

Loss Costs per $1,000 should be related to the Coverage “A” Amount

For weighting the Coverage “D” Loss Costs, use the file named “99 FHCF Wts.xls” for distribution for Coverage “A”.

Loss Costs for the various deductibles should be determined based on “per occurrence” deductibles.

Explain any deviations and differences from the prescribed format above.

Specify the model and program version numbers reflecting the release date as a footnote on each page of the output. There are no deviations from the prescribed format.

Applied Research Associates, Inc. 87 February 2002 Module 3 Module 3 - Section VI

General

1. Describe in detail how invalid zip codes are handled within your model or modeling practice. Are they deleted from the analysis, allocated, mapped back into the exposure data set, or handled in some other fashion? (Standard 5.1.5) If zip codes are unidentified, we first attempt to map the zip code to one in our historical databases. If this is not successful, we use public domain information on historical zip codes in an attempt to map the zip code. If a policy with an older zip code is found, it is mapped to the centroid of its new zip code in the most current database. Finally, if zip codes cannot be reassigned, they are removed from the analysis and reported as missing zip codes. 2. Describe what is done to prevent tampering of the computer code by users. How is the security of the model code addressed? (Standard 5.5.2) Modifications to the code can only be checked into the revision control system by authorized, in-house developers. Baseline tests are always run to ensure the model is functioning properly and reproducing known results.

Applied Research Associates, Inc. 88 February 2002 Module 3 Module 3 - Section VII

I. Data Flow Chart

Following is a data flow chart depicting the process of evaluating hurricane catastrophe simulation models:

Data Flow Chart

Hypothetical Events

Input Output Analysis Detailed Data Forms & Report Testing

Probabilistic Analysis

PHASE 1 PHASE 2

Baseline Tests (Standards 5.4.3, 5.4.5, 5.4.7, and 5.4.17 for all items in this section)

Sample Input Data

Sample input data has been provided to the modeler on the enclosed CD-ROM in the file named “inpdat01.xls”. The Commission is asking that the modeler run various scenario hurricane events (hypothetical and probabilistic) through its model on the sample input exposure data. The attached output forms must be filled out and specified loss files are to be provided to the Commission on CD-ROM in both an Excel and a PDF format. The file names shall include the abbreviated name of the modeler and the Standards year.

This data set consists of one $100,000 building for each construction type for each zip code in the state of Florida. The data set contains 6,052 records. The following is a description of the fields in the data set:

Applied Research Associates, Inc. 89 February 2002 Module 3

No. Field D escription 1. County Code Federal Information Processing Standards (FIPS) County Code - see attached description following Form F

2. Zip Code 5-digit zip code

3. Construction Type The following codes will be used: 1 = Wood Frame, 2 = Masonry, 3 = Mobile Home, 4 = Unknown

4. Deductible 1% policy deductible for all records

5. Total Insured Value $100,000 for all records - Building

6. Total Insured Value $10,000 for all records - Appurtenant Structures

7. Total Insured Value $50,000 for all records - Contents

8. Total Insured Value $20,000 for all records - Additional Living Expense

The modeler is directed to make the following assumptions with the analysis:

− Each structure is insured 100% to value − Number of stories = 1 − Occupancy type = Single Family Dwelling − Year of Construction = 1980 − Tide at landfall is 0 meters − If the model assumes different construction types other than those provided with the data, map the codes the Commission has provided to the appropriate codes. The Commission requests a copy of this mapping and proper documentation describing the reason for the mapping. In addition, the modeler is requested to provide information as to the assumptions made with the unknown construction types by the model. − Verify that only population weighted centroids were used for the location of risks within the zip code, where more specific locations were not available.

Applied Research Associates, Inc. 90 February 2002 Module 3 All other assumptions that the modeler must make with the analysis must be reviewed with the Commission staff. The intent is to keep all assumptions consistent among the modelers. All of the required assumptions listed above are made in the analysis. The unknown building is a mixture of wood and masonry that varies within different regions of the state. Mobile homes are not included in unknown building types. No other assumptions have been made.

TESTS

Zip Code Data Base - Form A

The accuracy of the model zip code database will be compared to the most current available. Complete Form A:

Form A Zip Code Data Base (Standard 5.1.5)

Zip Codes used in the model shall be weighted by population.

Describe methods used to verify accuracy of zip code data used. Describe the mapping of the construction codes provided with the data to the construction codes used by the model, if any. Describe how the unknown construction code is handled.

Model zip code database as of 2001 . Sample exposure zip codes as of most current available.

Matched Unmatched No. of Records 1513 0 % of Total Records 100 0 Total Exposure 1,089,360,000 0 % of Total Exposure 100 0

Applied Research Associates, Inc. 91 February 2002 Module 3 30 Hypothetical Events - Form B (Hypothetical Event Evaluation)

Each modeler is required to model 30 hypothetical events. These events have been specified by the Commission. These events consist of 5 hurricanes, one for each hurricane category 1-5, at 6 different landfall locations; Jacksonville, Fort Pierce, Miami, Ft. Myers, Tampa/St. Petersburg, and Panama City. The Commission is requesting the maximum estimated one- minute sustained 10-meter wind speed associated with the events as well as the estimated loss by coverage type. The purpose of this analysis is to evaluate the consistency of the wind speeds and loss estimates among the models.

A description of the events is contained in the file named “eval2.csv” on the supplied CD- ROM. Provide this information on CD-ROM in both an Excel and a PDF format. The file name shall include the abbreviated name of the modeler and the Standards year. Complete Form B using the specified file layout:

Form B 30 Hypothetical Events (Standard 5.2.5)

Estimated losses are requested in total and by coverage type for the 30 hypothetical events.

No. Field Description

1. Event ID Event identification 1-30

2. Category Saffir-Simpson Hurricane Category 1-5

3. Central Pressure Measured in millibars

4. Radius of Maximum Winds Measured in statute miles

5. Forward Speed Measured in miles per hour

6. Landfall Latitude and longitude of event at landfall location

7. Location General area of landfall

8. Direction Measured in degrees, assuming 0 degrees is north

9. Radius of Hurricane Force Winds Measured in statute miles

10. Maximum Estimated Wind Speed Maximum estimated one minute average wind speed for this event

11. Total Estimated Loss Total estimated loss summarized for building, appurtenant structures, contents and additional living expense

12. Estimated Building Loss Total estimated loss for building

13. Estimated App. Structure Loss Total estimated loss for appurtenant structures

14. Estimated Contents Loss Total estimated loss for contents

15. Estimated ALE Loss Total estimated loss for additional living expense

Modeled estimated one-minute average wind speeds produced in Form B shall be consistent with central pressure inputs. The requested data is provided on the CD-ROM in the files ARA_2001FormB.xls and ARA_2001FormB.pdf.

Applied Research Associates, Inc. 92 February 2002 Module 3 One Hypothetical Event - Form C (Hypothetical Event Evaluation)

In addition to the 30 hypothetical events, wind speeds for 336 zip codes have been provided to the modeler by the Commission. This information can be found on the supplied CD-ROM in the file named “eval3.csv”. The wind speeds* and zip codes represent a hypothetical hurricane track. The purpose is to compare the estimated damages by wind speed and construction type. The modeler is instructed to model the sample exposure data against these wind speeds at the specified zip codes and provide the Commission with damage ratios summarized by wind speed (mph) and construction type. If additional assumptions are necessary to complete this form (for example, regarding duration), the modeler shall indicate those used. Complete Form C:

Form C One Hypothetical Event

Total Loss**/ Wind speed* (mph) Subject Exposure

0.00000% 20 – 30 0.00002% 31 – 40 0.00082% 41 – 50 0.05450% 51 – 60 0.28133% 61 – 70 0.92624% 71 – 80 1.98803% 81 – 90 6.87932% 91 – 100 10.57705% 101 – 110 27.37453% 111 – 120 43.10940% 121 – 130 57.13239% 131 – 140 141 – 150 70.37171%

Total Loss**/ Construction Type Subject Exposure

1.357% Wood Fram e 1.300% Masonry 3.651% Mobile Hom e Unknown 1.344%

*Wind speeds are one-minute sustained, ten-meter wind speeds.

**Total loss is the sum of loss to all buildings in that category. For example, the total loss to all buildings affected by 50 knot winds or the total loss to all buildings with wood frame construction.

Applied Research Associates, Inc. 93 February 2002 Module 3 Loss Costs - Form D (Probabilistic Analysis)

The modeler is instructed to provide loss costs for each construction type for each zip code in the sample data set named “inpdat01.xls”. The following is a description of the requested file layout. Follow the instructions on Form D below. Note that fields 1-9 are the exposure fields from the sample data set. Fields 10-13 are for the loss costs (net of deductibles).

Form D Loss Costs

Provide this form along with expected annual loss costs by construction type and coverage for each zip code in the sample data set. There are 1,513 zip codes in the sample data set and 4 construction types; therefore, the completed file should have 6,052 records in total. If there are zip codes in the sample data set that your model does not recognize as “valid”, provide a list of such zip codes and either a) the new zip code to which the original one was mapped, or b) an indication that the insured values from this zip code were not modeled. Furthermore, provide loss cost data using all zip codes provided in the sample data set. In other words, if no losses were modeled, the record should still be included in the completed file with loss costs of zero, and, if a zip code was mapped to a new one, the resulting loss costs should be reported with the original zip code. Provide the results on CD-ROM in both an Excel and a PDF format using the following file layout. The file name shall include the abbreviated name of the modeler and the Standards year.

Order Field Name Description 1 Analysis Date Date of Analysis – YYYY/MM/DD 2 County Code FIPS County Code 3 Zip Code 5-digit Zip Code 4 Construction Type Use the following: 1 = Wood Frame, 2 = Masonry, 3 = Mobile Home, 4 = Unknown 5 Deductible 1% (of the Building Value) policy deductible for each record (i.e., 0.01*$100,000) 6 Building Value $100,000 for each record 7 Appurtenant Structures Value $10,000 for each record 8 Contents Value $50,000 for each record 9 Additional Living Expense Value $20,000 for each record 10 Building Loss Cost* Estimated expected annual loss cost for building divided by the building value modeled for each record ($100,000) 11 Appurtenant Structures Loss Cost* Estimated expected annual loss cost for appurtenant structures divided by the appurtenant structures value modeled for each record ($10,000) 12 Contents Loss Cost* Estimated expected annual loss cost for contents divided by the contents value modeled for each record ($50,000) 13 Additional Living Expense Loss Cost* Estimated expected annual loss cost for additional living expense divided by the additional living expense value modeled for each record ($20,000) *Round all loss costs to 6 decimal places

Applied Research Associates, Inc. 94 February 2002 Module 3 All deductibles are a percentage of the Building Value and are policy-level deductibles; however, for reporting purposes, the policy deductible should be pro-rated to the individual coverage losses in proportion to the loss.

Example Assume that a model analyzing Wood Frame properties in zip code 33102 (Miami-Dade County) estimated the following: Field Name Value Analysis Date 1999/11/15 County Code Miami-Dade County = 86 Zip Code 33102 Construction Type Wood Frame = 1 Deductible 1% = 0.01*$100,000 = $1,000 Building Value $100,000 Appurtenant Structures Value $10,000 Contents Value $50,000 Additional Living Expense Value $20,000 Building Loss* $10,000 Appurtenant Structures Loss* $1,000 Contents Loss* $2,500 Additional Living Expense Loss* $500 *Represents 1st dollar losses (i.e., prior to application of deductibles)

The $1,000 policy deductible would be applied as follows: Deductible 1% = 0.01*$100,000=$1,000 Building Loss $10,000-[($10,000÷$14,000)x$1,000]=$9,285.71 Appurtenant Structures Loss $1,000-[($1,000÷$14,000)x$1,000]=$928.57 Contents Loss $2,500-[($2,500÷$14,000)x$1,000]=$2,321.43 Additional Living Expense Loss $500-[($500÷$14,000)x$1,000]=$464.29

The reported Form D data are shown below: Field Name Value Analysis Date 1999/11/15 County Code Miami-Dade County = 86 Zip Code 33102 Construction Type Wood Frame = 1 Deductible 1% = 0.01 Building Value $100,000 Appurtenant Structures Value $10,000 Contents Value $50,000 Additional Living Expense Value $20,000 Building Loss Cost $9,285.71÷$100,000 = 0.092857 Appurtenant Structures Loss Cost $928.57÷$10,000 = 0.092857 Contents Loss Cost $2,321.43÷$50,000 = 0.046429 Additional Living Expense Loss Cost $464.29÷$20,000 = 0.023214 Based on the above information, the data should be reported in the following format: 1999/11/15,86,33102,1,0.01,100000,10000,50000,20000,0.092857,0.092857,0.046429,0.023214 The requested data are provided on the CD-ROM in the files ARA_2001FormD.xls and ARA_2001FormD.pdf.

Applied Research Associates, Inc. 95 February 2002 Module 3 Probable Maximum Loss (PML) - Form E (Probabilistic Analysis)

The modeler will provide estimates of loss for various probability levels using the hypothetical data set. The modeler will also provide the annual aggregate and occurrence mean, median and standard deviation for its PML distribution. Complete Form E:

Form E Probable Maximum Loss

Part A

Return Probability of Estimated Time (yea rs) Exceedance Loss

0.001% $293,496,320 Top Event $205,041,936 10,000 0.01% $189,762,352 5,000 0.02% $167,988,192 2,000 0.05% $150,594,064 1,000 0.10% $129,496,200 500 0.20% $104,057,392 250 0.40% $72,850,096 100 1.00% $47,796,976 50 2.00% $21,219,792 20 5.00% $8,382,223 10 10.00% 5 20.00% $2,287,875

Part B

Annual Ag gregate Occurrence

$4,178,906 $5,451,423 Mean $611,823 $35,605 Median Standard Deviation $16,595,663 $15,463,068

Applied Research Associates, Inc. 96 February 2002 Module 3 Hypothetical Events – Form F (Sensitivity and Uncertainty Analysis)

Form F Hypothetical Events for Sensitivity and Uncertainty Analysis

Wind speeds (in miles per hour for one minute sustained, ten-meter winds) at hourly intervals are requested over a 5 × 13 grid for the 500 combinations in the Excel file named “inputformF.xls” of the initial conditions of central pressure (in millibars), radius of maximum winds (in miles), forward speed (in miles per hour) and far field pressure (in millibars) for each of three categories of storms (1, 3, and 5) following a straight due West track passing through the point (80.2W, 25.8N). The first 100 combinations will be used in sensitivity analysis calculations, while the remaining 400 combinations will be used for uncertainty assessment. The storms themselves are similar to those in Form B, event ID 11, 13, and 15. The first worksheet in this file represents the specifications of initial conditions for a category 5 storm, the second worksheet represents the specifications of the initial conditions for a category 3 storm, and the third worksheet represents the specifications of the initial conditions for a category 1 storm. Depending on the operational model, each of the 500 simulated hypothetical events may not produce a maximum wind speed over the grid within the category given in the Saffir-Simpson scale. However, this is to be expected due to the deviation from the mean levels in a specific simulated event (for example, higher than average central pressure, lower than average far field pressure, slower than average forward speed could lead to a weak storm) and the coarseness of the grid resolution.

The “grid” of points is depicted in Figure 27 for category 5, Figure 28 for category 3, and Figure 29 for category 1. The East-West increments are 15 miles for all three storm categories, while the North-South increments vary as indicated. The North-South increment is 5 miles for category 5, 6 miles for category 3, and 8 miles for category 1. The point (0, 0) is the location of the center of the storm at time 0, and is 15 miles East of the landfall location (80.2W, 25.8N). The exact latitude-longitude location for each of the 65 vertices in the grid (5 × 13) can be deduced from this set-up.

• • • • • • • • • • • • • 15N • • • • • • • • • • • • • 10N • • • • • • • • • • • • • 5N • • • • • • • • • • • • • 0 • • • • • • • • • • • • • -5S

180W 165W 150W 135W 120W 105W 90W 75W 60W 45W 30W 15W 0

Storm Path from (0, 0) to (180W, 0)

Figure 27. Grid for Calculating Hourly Wind Velocities, Category 5 (Coordinates with Respect to Initial Storm Center at (0,0) at time 0)

Applied Research Associates, Inc. 97 February 2002 Module 3 • • • • • • • • • • • • • 18N • • • • • • • • • • • • • 12N • • • • • • • • • • • • • 6N • • • • • • • • • • • • • 0 • • • • • • • • • • • • • -6S

180W 165W 150W 135W 120W 105W 90W 75W 60W 45W 30W 15W 0

Storm Path from (0, 0) to (180W, 0)

Figure 28. Grid for Calculating Hourly Wind Velocities, Category 3 (Coordinates with Respect to Initial Storm Center at (0,0) at time 0)

• • • • • • • • • • • • • 24N • • • • • • • • • • • • • 16N • • • • • • • • • • • • • 8N • • • • • • • • • • • • • 0 • • • • • • • • • • • • • -8S

180W 165W 150W 135W 120W 105W 90W 75W 60W 45W 30W 15W 0

Storm Path from (0, 0) to (180W, 0)

Figure 29. Grid for Calculating Hourly Wind Velocities, Category 1 (Coordinates with Respect to Initial Storm Center at (0,0) at time 0) Output is to be provided on CD-ROM in both an Excel and a PDF format as shown in the file named “2001FormF.xls.” The file name shall include the abbreviated name of the modeler and the Standards year.

One sheet is used for each category of storm (Cat5, Cat3, Cat1). The columns in each sheet are: 1. Sample number (1-100) 2. Coded E-W (0, 15, 30, … , 180) 3. Coded N-S (Category 5: -5, 0, 5, 10, 15; Category 3: -6, 0, 6, 12, 18; Category 1: -8, 0, 8, 16, 24) 4. Wind speed at time 0hr 5. Wind speed at time 1hr 6. Wind speed at time 2hr 7. Wind speed at time 3hr 8. Wind speed at time 4hr 9. Wind speed at time 5hr 10. Wind speed at time 6hr

Applied Research Associates, Inc. 98 February 2002 Module 3 11. Wind speed at time 7hr 12. Wind speed at time 8hr 13. Wind speed at time 9hr 14. Wind speed at time 10hr 15. Wind speed at time 11hr 16. Wind speed at time 12hr

Successful completion of Form F demonstrates that the modeler is capable of running an insurance portfolio at a latitude/longitude level directly and at a street address level indirectly with appropriate conversion to latitude/longitude.

The required data are provided on the CD-ROM in the files ARA_2001FormF.xls and ARA_2001FormF.pdf.

Form F Comments Windspeeds are given as over water values for grid points located in the Atlantic Ocean or in the Gulf of Mexico. Overland values are given for open terrain conditions (i.e. zo=0.03m). Windspeed estimates are not able to be computed if the center of the storm is too far away from a grid point. The definition of “too far” varies with the size of the storm but is a function of the site to storm center distance divided by the radius to maximum winds. Wind speed values of zero are given in Form F when this occurs.

Applied Research Associates, Inc. 99 February 2002 Module 3

County County County County County County Code Name Code Name Code Name

001 Alachua 047 Hamilton 093 Okeechobee 003 Baker 049 Hardee 095 Orange 005 Bay 051 Hendry 097 Osceola 007 Bradford 053 Hernando 099 Palm Beach 009 Brevard 055 Highlands 101 Pasco 011 Broward 057 Hillsborough 103 Pinellas 013 Calhoun 059 Holmes 105 Polk 015 Charlotte 061 Indian River 107 Putnam 017 Citrus 063 Jackson 109 St. Johns 019 Clay 065 Jefferson 111 St. Lucie 021 Collier 067 Lafayette 113 Santa Rosa 023 Columbia 069 Lake 115 Sarasota 025* Dade 071 Lee 117 Seminole 027 De Soto 073 Leon 119 Sumter 029 Dixie 075 Levy 121 Suwannee 031 Duval 077 Liberty 123 Taylor 033 Escambia 079 Madison 125 Union 035 Flagler 081 Manatee 127 Volusia 037 Franklin 083 Marion 129 Wakulla 039 Gadsden 085 Martin 131 Walton 041 Gilchrist 087 Monroe 133 Washington

043 Glades 089 Nassau 045 Gulf 091 Okaloosa

Note: These codes are derived from the Federal Information Processing Standards (FIPS) Codes.

*The FIPS code and description for Dade County was changed to 086, Miami-Dade. The data files provided to the modelers do not reflect this change. Dade County continues to be identified as 025. Modelers should map to the old County Code 025 and if necessary, re- identify 086 to 025.

Figure 30. Florida County Codes

Applied Research Associates, Inc. 100 February 2002 Module 3

Figure 31. Florida Counties

Applied Research Associates, Inc. 101 February 2002 Module 3

OUTPUT RANGES

Applied Research Associates, Inc. 102 February 2002 Module 3 REFERENCES

AFPA/AWC (1996). “Wood Frame Construction Manual for One- and Two-Family Dwellings, 1995 SBC High Wind Edition,” American Forest and Paper Association/American Wood Council, p.71.

AFPA/AWC (1997). “National Design Specification for Wood Construction,” American Forest and Paper Association/American Wood Council, August 7.

AISC (1995). Load Resistance Factor Design, Manual of Steel Construction 2nd Edition, American Institute of Steel Construction, Chicago, Illinois, Volumes I and II, pp. 6-73–6-79 and pp. 8-98–8-130.

Alliss, R. (1992). “The Utilization of SSM/I Data in Analysis of Tropical and Extra-Tropical Cyclones,” M.S. Thesis, Dept. of MEAS, North Carolina State University, Raleigh, North Carolina.

Alliss, R., Raaman, S., and Chang, S. (1992). “Special Sensor Microwave/Imager (SSM/I) Observations of Hurricane Hugo (1989),” Monthly Weather Review, Volume 120, pp. 2723-2737.

Alliss, R., Sandlin, G., Chang, S., and Raaman, S. (1992). “Applications of SSM/I Data in Analysis of Hurricane Florence 1988,” Journal of Applied Meteorology, Volume 33, pp. 1581-1591.

Arya, S.P.S. (1988). Introduction to Micrometeorology, Academic Press.

Batts, M.E., Cordes, M.R., Russell, C.R., Shaver, J.R., and Simiu, E. (1980). “Hurricane Windspeeds in The United States,” National Bureau of Standards Report Number BSS-124, U.S. Dept. of Commerce, Washington, D.C.

Beason, W.L., Meyers, G.E., and James, R.W. (1984). “Hurricane Related Window Glass Damage in Houston,” ASCE Journal of the Structural Division, Volume 110, Number 12, pp. 2843-2857, December.

Behr, R.A., and Minor, J.E. (1994). “A Survey of Glazing System Behavior in Multi-Story Buildings During Hurricane Andrew,” The Structural Design of Tall Buildings, Volume 3, pp.143-161.

Behr, R.A., Minor, J.E., and Norville, H.S. (1993). “Structural Behavior of Architectural Laminated Glass,” Journal of Structural Engineering, Volume 119, Number 1.

*Bhinderwala, S. (1995). “Insurance Loss Analysis of Single Family Dwellings Damaged in Hurricane Andrew,” M.S. Thesis, Clemson University, Clemson, South Carolina.

Applied Research Associates, Inc. 103 February 2002 References Boissonnade, A., and Dong, W-m. (1993). “Hurricane Loss Estimation and Forecasting: Applications to Risk Management,” Proceedings 7th US National Conference on Wind Engineering, Los Angeles, California, pp. 115-124.

Brook, R.R. (1972). “Measurements of Turbulence in a City Environment,” Journal of Applied Meteorology, Volume 11, pp. 443-450.

*Canfield, L.R., Niu, S.H., and Liu, H. (1991). “Uplift Resistance of Various Rafter-Wall Connections,” Forest Products Journal, Volume 41, Number 7/8, July/August.

Case, P.C. (1996). “Wind Loads on Low Buildings with 4:12 Gable Roofs,” M.S. Thesis, University of Western Ontario, London, Ontario, Canada, May.

Cherry, N.J. (1991). “Fixing Studies for MRTI Normal Weight Tiles” SBCCI Submission, Redland Technology Limited, New Technology & Product Development Centre, Horsham, West Sussex, England, December.

*Chouinard, L.E, Chang, Liu and Cooper, C.K. (1997). “Model for Severity of Hurricanes in Gulf of Mexico”, Journal of Waterway, Port, Costal and Ocean Engineering, Vol. 123, No. 3, pp. 120-129.

Clark, K.M. (1987). “A Formal Approach to Catastrophe Risk Assessment and Management,” Proceedings November 9-11, 1986, Volume LXXIII, Number 140, pp. 69-91.

CMHC (1989). “Canadian Wood-Frame House Construction,” Canada Mortgage and Housing Corporation, p.48.

*Conner, H.W., Gromala, D.S., and Burgess, D.W. (1997). “Roof Connections in Houses: Key to Wind Resistance,” Journal of Structural Engineering, Volume 113, Number 12, pp. 2459- 2474, December.

*Crandell, J.H., Nowak, M., Laatsch, E.M., van Overeem, A., Barbour, C.E., Dewey, R., Reigel, H., and Angleton, H. (1993). “Assessment of Damage to Single-Family Homes Caused by Hurricanes Andrew and Iniki,” Contract HC-5911, NAHB Research Center, Upper Marlboro, Maryland, for U.S. Department of Housing and Urban Development Office of Policy Development and Research, September.

Cunningham, T.P. (1993). “Roof Sheathing Fastening Schedules for Wind Uplift,” APA Report T92-28, American Plywood Association, Tacoma, Washington, March.

Daneshvaran, S., and Morden, R.E. (1998). “Fast Hurricane Wind Analysis Using the Orthogonal Decomposition Method,” Journal of Wind Engineering and Industrial Aerodynamics, Volumes 77/78 Complete, pp. 703-714.

*Darling, R.W.R. (1991). “Estimating Probabilities of Hurricane Wind Speeds Using a Large- Scale Empirical Model,” Journal of Climate, Volume 4, Number 10, pp. 1035-1046.

Davenport, A.G. (1961). “The Application of Statistical Concepts to the Wind Loading of Structures,” Proceedings Institute of Civil Engineers, Volume 19, pp. 449-472.

Applied Research Associates, Inc. 104 February 2002 References Dawe, J.L., and Aridru, G.G. (1993). “Prestressed Concrete Masonry Walls Subjected to Uniform Out-of-Plane Loading,” Canadian Journal of Civil Engineering, Volume 20, pp.969-979.

Drysdale, R.G., and Essawy, A.S. (1988). “Out-of Plane Bending of Concrete Block Walls,” ASCE Journal of Structural Engineering, Volume 114, Number 1, pp. 121-133, January.

Duchêne-Marullaz, P. (1979). “Effect of High Roughness on the Characteristics of Turbulence in Cases of Strong Winds,” Proceedings 5th International Conference on Wind Engineering, Fort Collins, Colorado, p. 15.

*Emanuel, K.A. (1988). “The Maximum Intensity of Hurricanes”, Journal of the Atmospheric Sciences, Vol. 45, pp. 1143-1155.

*ESDU (1982). “Strong Winds in the Atmospheric Boundary Layer, Part 1: Mean Hourly Wind Speed,” Engineering Sciences Data Unit Item Number 82026, London, England.

*ESDU (1983). “Strong Winds in the Atmospheric Boundary Layer, Part 2: Discrete Gust Speeds,” Engineering Sciences Data Unit Item Number 83045, London, England.

Fattal, S.G., and Cattaneo, L.E. (1977). “Evaluation of Structural Properties of Masonry in Extsting Buildings,” U.S. Department of Commerce, National Bureau of Standards, p.13.

*FEMA (1992). “Building Performance: Hurricane Andrew in Florida,” Federal Emergency Management Agency, Federal Insurance Administration, FIA-22 (2/93). December 21.

Forest Products Laboratory (1987). “Wood Handbook: Wood as an Engineering Material,” U.S. Department of Agriculture, Washington D.C.

Friedman, D.G. (1987). “US Hurricanes & Windstorms,” DYP Insurance & Reinsurance Group Ltd.

*Georgiou, P.N. (1985). “Design Windspeeds in Tropical Cyclone-Prone Regions,” Ph.D. Thesis, Faculty of Engineering Science, University of Western Ontario, London, Ontario, Canada.

Georgiou, P.N., Davenport, A.G., and Vickery, B.J. (1983). “Design Wind Speeds in Regions Dominated by Tropical Cyclones,” Journal of Wind Engineering and Industrial Aerodynamics, Volume 13, Numbers 1-3, pp. 139-152.

Grimm, C.T. (1999). “Walls May Be Half as Strong as They Say!,” Southern Building, p. 19, January/February.

Gross, J.G., Dikkers, R.D. and Grogan, J.C. (1969). “Recommended Practice for Engineered Brick Masonry,” Brick Industry Association, Reston, Virginia.

Hebert, P.J., Jarrell, J.D., and Mayfield, B.M. (1996). “The Deadliest, Costliest, and Most Intense United States Hurricanes of This Century (and Other Frequently Requested Hurricane Facts),” NOAA Technical Memorandum NWS-NHC-31, p. 41.

Applied Research Associates, Inc. 105 February 2002 References *Ho, F.P., et al. (1987). “Hurricane Climatology for the Atlantic and Gulf Coasts of the United States,” NOAA Technical Report NWS38, Federal Emergency Management Agency.

*Ho, T.C.E. (1992). “Variability of Low Building Wind Loads,” Ph.D. Thesis, Faculty of Engineering Science, University of Western Ontario, London, Ontario, Canada.

Hogan, M. and Karwoski, K. (1990). “Masonry Performance in the Coastal Zone,” in Proceedings “Hurricane Hugo One Year Later,” B. L. Sill and P. R. Sparks (Eds.), Charleston, South Carolina, September.

*Holland, G.J. (1980). “An Analytic Model of the Wind and Pressure Profiles in Hurricanes,” Monthly Weather Review, Volume 108, Number 8, pp. 1212-1218.

Houston, S.H., and Powell, M.D. (1993). “Surface Wind Fields During Hurricane Bob’s (1991) Landfall in New England," Proceedings 20th Conference on Hurricanes and Tropical Meteorology, American Meteorological Society, San Antonio, Texas, May 10-14.

Houston, S.H., Powell, M.D., and Dodge, P.P. (1997). “Surface Wind Fields in 1996 Hurricanes Bertha and Fran at Landfall,” Proceedings 22nd Conference on Hurricanes and Tropical Meteorology, American Meteorological Society, Fort Collins, Colorado, May 19-23.

HUD (1980). Economic Benefit-Cost and Risk Analysis of Mobile Home Safety Research: Wind Safety Analysis, Office of Policy Development and Research, Department of Housing and Urban Development.

HUD (1994). “Final Rule - Manufactured Home Construction and Safety Standards on Wind Standards,” Office of the Assistant Secretary for Housing, Department of Housing and Urban Development, Federal Register, Volume 59, Number 10, pp. 2456-2475.

Jensen, M. (1958). “The Model-Law for Phenomena in Natural Wind,” Ingeniøren, Volume 2, pp. 121-128.

Kareem, A., and Stevens, J.G. (1985). “Window Glass Performance and Analysis in Hurricane Alicia,” ASCE Proceedings Specialty Conference: “Hurricane Alicia: One Year Later,” Galveston, Texas (August 1984), A. Kareem (Ed.), pp.178-186.

Karlsson, S. (1986). “The Applicability of Wind Profile Formulas to an Urban-Rural Interface Site,” Boundary Layer Meteorology, Volume 34, pp. 333-355.

Kind, R.J., and Wardlaw, R.L. (1984). “Behavior in Wind of Loose-Laid Roof Insulation Systems, Part I: Stone Scour and Blow-Off,” Proceedings 4th Canadian Workshop on Wind Engineering, Toronto, Canada, pp. 141-149.

Leland, K.B. (1988). “The Strength of Roof Anchorage in Unreinforced Concrete Masonry,” Masters Thesis, Clemson University, Clemson, South Carolina, August.

Lettau, H.H. (1969). “Note on Aerodynamic Roughness-Parameter Estimation on the Basis of Roughens-Element Description,” Journal of Applied Meteorology, Volume 8, pp. 828-832.

Applied Research Associates, Inc. 106 February 2002 References Lin, J.X., and Surry, D. (1997). “Simultaneous Time Series of Pressures on the Envelope of Two Large Low-Rise Buildings,” University of Western Ontario, Boundary Layer Wind Tunnel Laboratory Report, BLWTL-SS7-1997, for Dr. E. Simiu, National Institute of Standards and Technology, March.

Macha, J., Sevier, J., and Bertin, J. (1983). “Comparison of Wind Pressures on a Mobile Home in Model and Full Scale,” Journal of Wind Engineering and Industrial Aerodynamics, Volume 12, pp. 109-124.

Mark, F. (1985). “Evolution and Structure of Precipitation in Hurricane Allen (1980),” Monthly Weather Review, Volume 113, pp. 909-930.

*Marshall, R. (1977). “The Measurement of Wind Loads on a Full Scale Mobile Home,” NBSIR 77-1289, National Bureau of Standards, Washington, D.C., p. 120.

*Marshall, R.D. (1993). “Wind Load Provisions of Manufactured Home Construction and Safety Standards – A Review and Recommendation for Improvement,” NISTIR 5189, Building and Fire Research Laboratory, Gaithersburg, Maryland, for Department of Housing and Urban Development, Washington, D.C., May.

*Marshall, R.D. (1994). “Manufactured Homes – Probability of Failure and the Need for Better Windstorm Protection Through Improved Anchoring Systems,” NISTIR 5370, Building and Fire Research Laboratory, Gaithersburg, Maryland, for Department of Housing and Urban Development, Washington, DC, November.

*Marshall, R.D., and Yokel, F.Y. (1995). “Recommended Performance-Based Criteria for the Design of Manufactured Home Foundation Systems to Resist Wind and Seismic Loads,” NISTIR 5664, Building and Fire Research Laboratory, Gaithersburg, Maryland, for Department of Housing and Urban Development, Washington, D.C., August.

*McDonald, J.R., and Pennington, W.V. (1986). “Hurricane Damage to Manufactured Homes,” Presented at American Society of Civil Engineers Structures Congress ’86, New Orleans, Louisiana, September.

McDonald, J.R., Jiang, H., and Yin, J. (1997). “Full-Scale Measurements of Wind Loads on Metal Edge Flashings and Copings,” Final Report, Institute for Disaster Research, Texas Tech University, Lubbock, Texas, for Roofing Industry Consortium, February.

*Meecham, D. (1988). “Wind Action on Hip and Gable Roofs,” M.S. Thesis, University of Western Ontario, London, Ontario, Canada, August.

MHCSS (1992). “Manufactured Home Construction and Safety Standards,” 24 CFR Chapter XX, Part 3280.

Minor, J.E. (1985). “Window Glass Performance and Hurricane Effects,” ASCE Proceedings Specialty Conference: “Hurricane Alicia: One Year Later,” Galveston, Texas (August 1984), A. Kareem (Ed.), pp.151-167

Applied Research Associates, Inc. 107 February 2002 References Minor, J.E. (1994). “Wind-Borne Debris and the Building Envelope,” Journal of Wind Engineering and Industrial Aerodynamics, Volume 53, pp. 207-227.

Minor, J.E., and Behr, R.A. (1993a). “Architectural Glazing Systems in Hurricanes: Performance, Design Criteria and Designs,” Proceedings 7th U.S. National Conference on Wind Engineering, Los Angeles, California, Volume II, pp. 453-461.

Minor, J.E., and Behr, R.A. (1993b). “Improving the Performance of Architectural Glazing in Hurricanes,” Hurricanes of 1992, R. A. Cook and M. Soltani (Eds.), ASCE, pp. 476-485.

Minor, J.E., Beason, W.L., and Harris, P.L. (1978). “Designing for Wind-Borne Missiles in Urban Areas,” ASCE Journal of the Structural Division, Volume 104, Number ST11, pp. 1749-1760.

Monroe, J.S. (1996). “Wind Tunnel Modeling of Low Rise Structures in a Validated Open Country Simulation,” M.S. Thesis, Clemson University, Clemson, South Carolina, August.

NCS BCS (1988). “Manufactured Home Installations,” NCS BCS A225.1/ANSI A225.1-1987, National Conference of States on Building Codes and Standards, Inc., Herndon, Virginia, p. 49.

NCS BCS (1994). “Manufactured Home Installations,” NCS BCS/ANSI A225.1-1994, National Conference of States on Building Codes and Standards, Inc., Herndon, Virginia, p. 29.

Neumann, C.J. (1991). “The National Hurricane Center Risk Analysis Program (HURISK),” NOAA Technical Memorandum NWS NHC 38, National Oceanic and Atmospheric Administration, Washington, D.C.

NFPA (1973). “Standard for Mobile Homes,” NFPA Number 501B-1973/ANSI A119.1, p. 33.

NRC (1991). “Hurricane Elena, Gulf Coast August 29 – September 2, 1985, Natural Disasters Studies, Volume Two,” HA 286, Committee on Natural Disasters, National Research Council, Washington, DC pp. 51-58.

Parris, S. (1996). “Design of Wind Uplift Capacities and Retro-Fitted Adhesives,” 490 and 491 Project Presentation, Clemson University, Clemson, South Carolina, August.

Powell, M.D. (1987). "Changes in the Low-Level Kinematic and Thermodynamic Structure of Hurricane Alicia (1983) of Landfall," Monthly Weather Review, Volume 115, pp. 75-99.

Powell, M.D., and Houston, S.H. (1996). "Hurricane Andrew’s Landfall in South Florida − Part II: Surface Wind Fields and Potential Real-Time Applications," Weather and Forecasting, Volume 11, Number 3, pp. 329-349.

Powell, M.D., and Houston, S.H. (1998). "Surface Wind Fields of 1995 Hurricanes Erin, Opal, Luis, Marilyn, and Roxanne at Landfall," to appear in Monthly Weather Review.

Powell, M.D., Dodge, P.P., and Black, L.B. (1991). "The Landfall of Hurricane Hugo in the Carolinas: Surface Wind Distribution," Weather and Forecasting, Volume 6, pp. 379-399.

Applied Research Associates, Inc. 108 February 2002 References Reed, T.D., Rosowsky, D.V. and Schiff, S.D. (1996). “Roof Rafter to Top-Plate Connection in Coastal Residential Construction,” Proceedings International Wood Engineering Conference, pp. 4-458−4-465

Reed, T.D., Rosowsky, D.V., and Schiff, S.D. (1997). “Uplift Capacity of Light-Frame Rafter to Top Plate Connections,” Journal of Architectural Engineering, pp. 157-163, December.

Rodgers, E., Chang, S., and Pierce, H. (1994). “A Satellite Observational and Numerical Study of Precipitation Characteristics in Western North Atlantic Cyclones,” Journal Applied Meteorology, Volume 33, pp. 129-139.

Rosowsky, D.V., and Reinhold, T.A. (1999). “Rate-of-Load and Duration-of-Load Effects for Wood Fasteners,” ASCE Journal of Structural Engineering, Volume 125, Number 7, pp.719-724, July.

Roy, R. (1983). “Wind Tunnel Measurements of Total Loads on a Mobile Home,” Journal of Wind Engineering and Industrial Aerodynamics, Volume 13, pp. 327-338.

Russell, L.R. (1968). “Probability Distributions for Texas Gulf Hurricane Effects of Engineering Interest,” Ph.D. Thesis, Stanford University, Stanford, California.

Russell, L.R. (1971). “Probability Distributions for Hurricane Effects,” Journal of Waterways, Harbors, and Coastal Engineering Division, Number 1, pp. 139-154.

Russell, L.R., and Schueller, G.F. (1974). “Probabilistic Models for Texas Gulf Coast Hurricane Occurrences,” Journal Petroleum Technology, pp. 279-288.

SBCCI (1991). “Standard Building Code: Appendix H, Manufactured Home Tie Down Standards,” Southern Building Code Congress International, Inc., Birmingham, Alabama, pp. 601-606.

Shapiro, L.J. (1983). “The Asymmetric Boundary Layer Flow Under a Translating Hurricane,” Journal of Atmospheric Science, Volume 40, Number 8, pp. 1984-1998.

Shiotani, M. (1962). “The Relationship between Wind Profiles and Stabilities of the Air Layer in the Outskirts of a City,” Journal of Meteorological Society, Japan, Volume 40, pp. 315-329.

*Simiu, E., and Scanlan, R.H. (1996). Wind Effects on Structures: Fundamental and Applications to Design, 3rd Edition, John Wiley & Sons, New York.

SJI (1994). “Catalogue of Standard Specifications, Load Tables, and Weight Tables for Steel Joists and Joist Girders,” Steel Joist Institute, Myrtle Beach, South Carolina.

SJI (1998). “Structural Design of Steel Joist Roofs to Resist Uplift Loads,” Technical Digest No. 6, Steel Joist Institute, Myrtle Beach, South Carolina.

Smith, T.L. (1999). “Records of Survey on Building Damages Caused by Hurricane Andrew,” Personal Communications, TLSmith Consulting.

Applied Research Associates, Inc. 109 February 2002 References Sparks, P.R., Schiff, S.D. and Reinhold, T.A. (1994). “Wind Damage to Envelopes of Houses and Consequent Insurance Losses,” Journal of Wind Engineering and Industrial Aerodynamics, Volume 53, pp. 145-155.

Stathopoulos, T. (1979). “Turbulent Wind Action on Low Rise Buildings,” Ph.D. Thesis, University of Western Ontario, London, Ontario, Canada, February.

Steyn, D.G. (1982). “Turbulence in an Unstable Surface Layer over Suburban Terrain,” Boundary Layer Meteorology, Volime 22, pp. 183-191.

Tryggvason, B.V., Surry, D., and Davenport, A.G. (1976). “Predicting Wind-Induced Response in Hurricane Zones,” Journal of Structural Division, Volume 102, Number 12, pp. 2333-2350.

Twisale, L.A., and Vickery, P.J. (1995). “Extreme Wind Risk Assessment,” Chapter 20, Probabilistic Structural Mechanics Handbook, Van Nostrund Reinhold, New York.

Twisdale, L.A (1996). “Risk Assessment Techniques: Evaluating Mitigation Options and Cost Effectiveness,” Workshop on Recovery Through Coastal Windstorm Mitigation, Gainesville, Florida, November 14-15.

Twisdale, L.A., and Dunn, W.L. (1983). “Extreme Wind Risk Analysis of the Indian Point Nuclear Generation Station,” Final Rep. 44T-2491, Addendum to Rep. 44T-2171, Research Triangle Institute, Research Triangle Park, North Carolina.

*Twisdale, L.A., Vickery, P.J., and Hardy, M.B. (1993). “Uncertainties in the Prediction of Hurricane Windspeeds,” ASCE Conference on Hurricanes of 1992, Miami, Florida, December.

*Twisdale, L.A., Vickery, P.J., and Lin, J.X. (2000a). “Analysis of Hurricane Windborne Debris Impact Risk for Residential Structures: Part I,” to be submitted to the ASCE Journal of Structural Engineering.

*Twisdale, L.A., Vickery, P.J., and Lin, J.X. (2000b). “Analysis of Hurricane Windborne Debris Impact Risk for Residential Structures: Part II,” to be submitted to the ASCE Journal of Structural Engineering.

*Twisdale, L.A., Vickery, P.J., and Steckley, A.C. (1996). “Analysis of Hurricane Windborne Debris Impact Risk for Residential Structures,” Applied Research Associates, Raleigh, North Carolina, March.

Uematsu, Y., and Isyumov, N. (1998). “Peak Gust Pressures Acting on the Roof and Wall Edges of a Low-Rise Building,” Journal of Wind Engineering and Industrial Aerodynamics, Volumes 77 and 78, pp. 217-231.

*Vasquez, J.L. (1994). “Development of a Windspeed-Damage Correlation Model for Manufactured Housing Subjected to Extreme Winds,” M.S. Thesis, Louisiana State University and Agricultural and Mechanical College, December.

Applied Research Associates, Inc. 110 February 2002 References *Vickery, P.J., and Twisdale, L.A. (1995a). “Wind-Field and Filling Models for Hurricane Wind-Speed Predictions,” Journal of Structural Engineering, Volume 121, Number 11, pp. 1700-1709.

*Vickery, P.J., and Twisdale, L.A. (1995b). “Prediction of Hurricane Wind Speeds in The United States,” Journal of Structural Engineering, Volume 121, Number 11, pp. 1691-1699.

*Vickery, P.J., Lin, J.X., and Twisdale, L.A. (1998) “Analysis of Hurricane Windborne Debris Impact Risk for Residential Structures: Part II,” Final Rep. 5609, Rev. 1, Applied Research Associates, Inc., Raleigh, North Carolina, April.

*Vickery, P.J., Skerjl, P.F., and Twisdale, L.A. (2000b). “Simulation of Hurricane Risk in the United States Using an Empirical Storm Track Modeling Technique,” Accepted for Publication in Journal of Structural Engineering.

*Vickery, P.J., Skerjl, P.F., Steckley, A.C., and Twisdale, L.A. (2000a). “A Hurricane Wind Field Model for Use in Simulations,” Submitted to Journal of Structural Engineering.

Vickery, P.J., and Skerjl, P.F. (2000).”Hurricane Gust Factors Revisited,” Submitted to Journal of Structural Engineering.

Wieringa, J. (1992). “Updating the Davenport Roughness Classification,” Journal of Wind Engineering and Industrial Aerodynamics, Volume 41-44, pp. 357-368.

Wieringa, J. (1993). “Representative Roughness Parameters for Homogeneous Terrain,” Boundary Layer Meteorology, Volume 63, pp. 323-363.

Yersel, M., and Goble, R. (1986). “Roughness Effects on Urban Turbulence Parameters,” Boundary Layer Meteorology, Voume 37, pp. 271-284.

Young, M.A., and Vickery, B.J. (1994). “A Study of Roof Uplift Forces on Three Schools of Windsor Roman Catholic Separate School Board,” Final Report BLWT-SS29-1994, The Boundary Layer Wind Tunnel Laboratory, The University of Western Ontario, Ontario, Canada.

Zaitz, M.D. (1994). “Roof Sheathing Racking Effect on Fastener Withdrawal Capacities,” M.S. Thesis, Clemson University, Clemson, South Carolina, pp.4-5, August.

Zollo, R.F. (1993). “Hurricane Andrew: August 24, 1992, Structural Performance of Buildings in Dade County, Florida,” Technical Report No. CEN 93-1, University of Miami, Miami, Florida, March.

Applied Research Associates, Inc. 111 February 2002 References