STATE OF FLORIDA
DIVISION OF ADMINISTRATIVE HEARINGS
NATIONWIDE MUTUAL FIRE, )
)
Petitioner, )
)
vs. ) Case No. 96-4989
)
DEPARTMENT OF INSURANCE, )
)
Respondent. )
)
RECOMMENDED ORDER
On April 28 through 30, 1998, a formal administrative hearing was held in this case in Tallahassee, Florida, before
J. Lawrence Johnston, Administrative Law Judge, Division of Administrative Hearings.
APPEARANCES
For Petitioner: W. Donald Cox, Esquire
Fowler, White, Gillen, Boggs, Villareal and Banker, P.A.
501 East Kennedy Boulevard, Suite 1700 Tampa, Florida 33602
For Respondent: Steve Parton, Esquire
Jeffrey W. Joseph, Esquire Department of Insurance Division of Legal Services
200 East Gaines Street 612 Larson Building
Tallahassee, Florida 32399-0333 STATEMENT OF THE ISSUE
The issue in this case is whether the Respondent, the Department of Insurance, should grant the application of the Petitioner, Nationwide Mutual Fire Insurance Company
(Nationwide), for residential property and casualty insurance premium rate increases.
PRELIMINARY STATEMENT
On April 2, 1996, Nationwide filed an application to increase its premium rates for residential property and casualty insurance by an average of 27%, effective May 31, 1996, under the "use and file" rate filing procedure. On or about September 26, 1996, the Department gave notice of intent to deny the application. Nationwide requested formal administrative proceedings, and the matter was referred to the Division of Administrative Hearings (DOAH) on October 21, 1996.
Initially, the case was set for final hearing on February 24 through 28, 1997, but the Respondent filed an unopposed Motion to Continue on February 14, 1997, and final hearing was continued until further notice while discovery and settlement discussion took place. Periodic status reports were filed until the Department filed a Motion to Reset Hearing on December 3, 1997, and final hearing was rescheduled for April 28 through May 1, 1998.
The parties filed a Prehearing Stipulation on April 17, 1998. At final hearing, Nationwide called three expert witnesses and had the following Petitioner's Exhibits admitted in evidence:
1 through 4; 7; 15; 16; and 21 through 31. The Department called one expert witness and had Respondent's Exhibit 1 admitted in
evidence. Nationwide ordered the preparation of a transcript of the final hearing, which was filed (in four volumes) on May 5, 1998.
FINDINGS OF FACT
The Petitioner, Nationwide Mutual Fire Insurance Company (Nationwide), is a foreign insurance company that has a certificate of authority issued by the Respondent, the Department of Insurance (the Department), to sell insurance in Florida.
The 1996 Rate Filing
On April 2, 1996, Nationwide filed a residential property and casualty insurance premium rate increase request with filing memorandum and supporting exhibits. Nationwide opted to "use and file" the rate increase and made the effective date of the increase May 31, 1996. To the extent not approved, the rate increase would have to be refunded.
Nationwide's 1996 rate filing requested increases for Nationwide's three single-family dwelling homeowner policies (Extended, Elite II, and Golden Blanket), as well as for its mobile home, condominium, and tenant policies. The rate filing alleged and sought to document indications that increases averaging approximately 61% were justified. Despite the alleged indications, Nationwide only requested rate increases averaging 27%.
Nationwide's rationale for seeking rate increases averaging 27% instead of 61% was that it intended to file a subsequent rate increase request averaging 27% so that, together, the two rate increase requests would result in an increase from present rates averaging 61%. There was no actuarial basis for seeking the increase in two phases; this decision was made at least in part for business reasons to make the increase more acceptable to customers.
On August 15, September 6, and November 20, 1996, Nationwide responded to Department requests for additional information and documentation in support of Nationwide's rate filing. As a result of the additional information, the rate indications alleged by Nationwide were reduced to increases averaging approximately 58%.
The 1995 Rate Filing
In 1995, Nationwide requested rate increases averaging 25%, based on indications that increases averaging 77% were justified. The increases requested in the 1995 rate filing were granted. The evidence was not clear as to the bases for either the 77% indications, the 25% request, or the Department's decision to grant the request. Without such evidence, it cannot be inferred, as the Department argues, that the pending 27% rate increase request based on indications of either 61% or 58% is too high simply because only a 25% increase was requested when rate indications were 77%.
The Hurricane Loss Component
Nationwide's rate filing is separated into non-hurricane and hurricane components. A significant portion of Nationwide's rate increase request is attributable to the hurricane component.
Since hurricanes are catastrophic events that occur relatively infrequently, insurance companies do not have the peril experience for hurricane losses that they do for most other kinds of perils covered by residential property and casualty insurance. For this reason, it is necessary to use a different methodology for the hurricane component of residential property and casualty insurance than is used for the non-hurricane component.
For the hurricane component of its 1996 rate filing, Nationwide essentially projected a dollar amount of hurricane losses for each $1,000 of insurance coverage written, times a loss projection factor (to account for changes over the years in average losses attributable to changes in average coverage value, deductibles, and changes in mix of business by rating territory), times an allocated loss adjustment expense (ALAE) ratio of 2.2%, to develop estimated hurricane losses and ALAE per policy. To arrive at the hurricane portion of the indicated premium requirement, Nationwide then divided estimated hurricane losses and ALAE per policy by 87.4% to account for agent commissions, state premium tax, profit and contingencies attributable to hurricane losses.
The ISO Excess Wind Procedure
Prior to Hurricane Andrew in 1992, hurricane losses were projected for rate purposes using a methodology developed by the Insurance Services Organization (ISO) known as the Excess Wind Procedure. Essentially, the ISO Excess Wind Procedure attempted to calculate the percentage of historical insurance losses attributable to "excess wind" from hurricanes and increased indicated non-hurricane losses by the same proportion.
The ISO Excess Wind Procedure had inherent weaknesses. The ISO Excess Wind Procedure was based on insurance loss data relating to historical hurricanes since 1960; yet, it is known that hurricanes occurred more frequently on an annual basis during the years 1900 through 1960 than during the years 1960 through 1990. In addition, the ISO Excess Wind Procedure used regional data, while it is known that Florida hurricane landfalls have been more frequent than regional hurricane landfalls.
Beyond the storm frequency issue, since it was based solely on historical data since 1960, the ISO Excess Wind Procedure did not account for the possibility of future losses in areas not hit by a hurricane since 1960. Meanwhile, the historical losses accounted for in the ISO Excess Wind Procedure occurred at times when fewer people lived in coastal areas and when property values in coastal areas were not as high, compared to inland properties, as they are now. There also have been significant changes in policy provisions (for example, provisions for payment of replacement costs rather than cash value) that would increase
insurance losses as a percentage of total losses. At the same time, safety features such as smoke detectors have been developed over the years, resulting in reductions in the non-hurricane component of total insurance losses. The ISO Excess Wind Procedure did not adjust for these changes.
Hurricane Andrew exposed the significance of the weaknesses in the ISO Excess Wind Procedure. Premium levels generally were not high enough to fund the payment of insurance losses due to Andrew. Several insurance companies were unable to pay all claims made as a result of the storm.
Since Andrew, the ISO Excess Wind Procedure has been generally discredited as inadequate due to its inherent weaknesses. Increasingly, the insurance industry and regulators have turned instead to computer models to project hurricane losses.
The AIR Model
For the hurricane component of its 1996 rate filing, Nationwide used a computer simulation model developed by Applied Insurance Research, Inc. (AIR) to project hurricane losses. AIR specializes in catastrophic risk assessment and primarily provides products and services for the insurance industry. It was formed in 1987 and has grown to approximately 60 employees,
50 of whom are technical professionals in fields such as meteorology, geophysics, seismology, structural engineering, and geotechnical engineering, statistics, mathematics, and computer software programming. AIR has about 250 clients at present,
about 100 of whom are primary insurers, and about 90 of whom are reinsurers.
AIR models hurricanes, windstorms such as tornadoes, earthquakes, and fires following earthquakes. The fundamental purpose of the AIR model is to provide AIR's clients with good information on probabilities of losses from these events.
Only a small percentage of AIR's current client base utilizes the AIR model for rate making purposes. Initially, the AIR model was used primarily by financial rating organizations to assess the solvency of insurance companies with exposure to hurricane risk. Later, it began to be used by reinsurers to determine premiums to be charged to insurance companies for hurricane risk. Since Andrew, the AIR model and similar computer models developed by other firms have begun to be used for primary insurance premium rate setting.
There are four primary components to the AIR model: event generation; estimated windfield; application of the windfield to actual properties to estimate damage; and application of the estimated damages to insurance coverage.
The event generation component of the model draws on historical information to assign probability distributions to important variables, such as frequency, intensity, landfall, and storm track. Using these probability distributions, the model then generates a probabilistic event set (or stochastic event set) consisting of a number of years of simulated storms. In the
case of Nationwide's 1996 rate filing, the AIR model simulated 10,000 years of hurricanes.
When the AIR model assigns probability distributions to variables in the event generation component, statistical "smoothing" is used to "fill the gaps" in the historical data. This allows the model to generate events in accordance with their probability of occurrence in the future instead of just their historical occurrence over the past hundred years or so.
Although the details of the statistical "smoothing" were not presented, it was clear that these statistical methods are standard and generally accepted.
In the second component of the model, windfields are estimated from the characteristics of each simulated storm. In this component, the model actually moves the simulated hurricane over land and develops a time profile of wind speeds and directions as the storm moves. The windfield generator component accounts for things such as modeled changes in central pressure, as well as dissipation from surface roughness, as the simulated storms proceeds over its track.
In the third component of the model, damage from the windfields of each simulated storm is estimated by overlaying the storm's windfields on the actual properties known to exist. In this component, the AIR model uses damage functions (also called damageability relationships). These functions relate wind speed to damage to buildings and contents, depending on construction
type, as well as resulting additional living expenses as applicable.
The last component essentially compares the damage estimates to insurance policy coverages and conditions. The result is in an insurance loss estimate for each simulated storm. To project average annual hurricane insurance losses, the estimated insurance losses from the simulated storms are annualized.
Some aspects of the AIR model, including the damage functions (or damageability relationships), are considered proprietary and were not presented in Nationwide's case-in-chief. However, it is clear that actuaries can rely on a computer model without knowing all of the proprietary information relating to the model. There are other ways to judge whether a model is worthy of reliance.
Peer Reviews of the AIR Model
One way to validate a computer model is through internal peer review. The details of Nationwide's internal peer review were not presented. However, AIR's president, Karen Clark, testified extensively, and it is clear from the evidence that AIR has validated its model to its own satisfaction. AIR's internal peer review is ongoing, and AIR continuously enhances its model as suggested by the ongoing internal review.
Another means of validating the AIR model is through external peer reviews. The AIR model has undergone several of
these. As with the internal peer reviews, the external peer reviews have resulted in enhancements to the AIR model.
One external peer review was commissioned by AIR in the mid-1980s. It addressed both AIR's hurricane model and its tornado model. While the details of the commissioned peer review were not presented, the conclusion was: "[A] scientifically defensible and realistically complete approximation of these major storm systems has been achieved. . . . On the whole, it is felt that the proposed models are scientifically well based and reflect an understanding of the current state-of-the-art of atmospheric sciences."
The North Carolina Rate Bureau also has been evaluating the AIR model since 1992. North Carolina hired various experts to review the AIR model, including an outside meteorologist and an outside statistician. Despite the length of this review, no major conclusions have been drawn yet.
One rating agency not identified by the evidence reviewed the AIR model and found a 20% variability based on all validation results. Evidence as to this external peer review was not presented.
Florida Hurricane Commission Review
In 1996, the Florida Commission on Hurricane Loss Projection Methodology (the Commission) adopted some 38 standards which computer models projecting hurricane losses should meet.
It is not clear from the evidence exactly when in 1996 the standards were adopted. The AIR model used by Nationwide in its
1996 rate filing met most but not all of the Commission's standards.
The Commission's 1996 standards required that models include and consider damage from weaker (Category 1) hurricanes storms and "bypass" hurricanes (i.e., hurricanes that do not make landfall in Florida but are close enough that winds greater than
74 miles per hour affect Florida), as well as from stronger hurricanes making landfall in Florida. The AIR model used by Nationwide in its 1996 rate filing only included and considered damage from stronger hurricanes making landfall in Florida.
The Commission's standards also required that models continue to simulate damage until modeled winds decreased to 35 miles per hour. The AIR model used by Nationwide in its 1996 rate filing only continued to simulate damage until modeled winds decreased to 50 miles per hour.
AIR revised its model to meet the Commission's 1996 standards and submitted the revised model for findings and determinations as to its "accuracy and reliability" under the Commission's 1996 standards. In May 1997, the Commission voted on certain findings and determinations regarding the revised AIR model based on the 1996 standards. While it appears that the Commission's findings and determinations were favorable to the revised AIR model (i.e., that the model was consistent with the standards), the evidence is not clear as to exactly what the Commission found or determined. AIR never received anything in writing to document the Commission's findings and determinations,
and neither any such writing nor a verbatim transcript nor the minutes of the Commission meeting were put in evidence.
If Nationwide had used the revised AIR model for its 1996 rate filing, hurricane loss projections would have been about 5% higher.
On May 29, 1997, the Commission revised its standards for computer models projecting hurricane losses. AIR again submitted its revised model under the Commission's revised standards. On October 24, 1997, the Commission determined the AIR model "to be acceptable for projecting hurricane loss costs for personal residential rate filings submitted to the Department of Insurance in Florida" and determined that the AIR model "complies with the standards adopted by the Commission on May 29, 1997."
Notwithstanding the Commission's October 24, 1997, determinations, it is not clear from the evidence whether the revised AIR model actually met each and every one of the some 38 Commission standards. The Department's actuary, Ken Ritzenthaler, who was a member of the Commission, testified that the Commission's professional team was unable to verify compliance with six or seven of the standards.
Ritzenthaler also testified that he disagreed with the Commission's determination that the AIR model met a standard requiring a model's simulation of insurance losses from Florida's
57 historical hurricanes to be "consistent" with actual insurance
losses. The AIR model simulated $1.2 billion in losses, while the "actual" losses used for the comparison were only $880,000 million.
Other Relevant Uses of the AIR Model
For a number of years, reinsurance companies have used the AIR model to determine the premiums to charge insurance companies for reinsuring their hurricane loss exposure. While only a small percentage of AIR's current client base of approximately 250 clients, not only Nationwide but also U.S.A.A., Allstate, and State Farm use the AIR model to project their hurricane losses for primary residential property and casualty insurance rate-making.
The Florida Hurricane Catastrophe Fund (the Cat Fund) was created by the Legislature after Hurricane Andrew to serve as a kind of mandatory reinsurance against catastrophic hurricane losses. For a mandatory premium payment, primary insurers of property and casualty risks in Florida are assured that the Cat Fund will pay a portion of catastrophic hurricane losses. With one minor difference, the AIR model used by Nationwide in its 1996 rate filing is the same model used by the Cat Fund to set its premiums. The only difference is that the Cat Fund inputs and output ranges are by zip code whereas Nationwide's inputs and output ranges for the 1996 rate filing were by Nationwide's territories, and some territories encompassed more than one zip code. Although the Cat Fund's rationale for using the AIR model
was not presented, the evidence was that the Cat Fund continues to use the AIR model.
Recently, some primary insurance companies, notably U.S.A.A., have turned directly to capital markets to fund a portion of its hurricane losses through the use of catastrophe bonds. To assign a rating to these bonds, it is necessary to determine the chances that the bonds would default. This required a projection of hurricane losses. In the case of U.S.A.A., hurricane losses were projected using the AIR model. Standard and Poor's, Moody's, and Fitch, Duff and Phelps conducted "due diligence" investigations of the U.S.A.A. catastrophe bonds that included validation reviews of the AIR model that lasted approximately a year before ratings were assigned for the bonds.
The Florida Residential Property and Casualty Joint Underwriting Association (FRPCJUA) uses the AIR model to project hurricane losses. The Department has required companies taking policies out of the FRPCJUA to use the AIR model to project hurricane losses in order to demonstrate the company's ability to pay on the policies.
Comparison of Projected to "Actual" Losses
Reviews of computer hurricane loss projection models usually include a comparison of losses simulated by a model for storms to actual losses from actual, historical storms. Strictly speaking, the purpose of such models is not to simulate losses from actual storms but rather to generate a complete probability
distribution. Exact comparisons between simulated and actual losses for specific historic storms are difficult, and perfect matches would not be expected.
One reason such comparisons are difficult is that a hurricane's actual overland wind speeds and other characteristics are never completely known. Surprisingly, usually relatively few overland wind speed recordings are available, and they only are accurate to within plus or minus 15%.
In addition, actual losses are not always easy to ascertain. Generally, industry-wide loss figures are less accurate than company loss figures. Real-time or other preliminary loss estimates may not be accurate. Sometimes, especially for smaller storms, more accurate after-the-fact loss figures are not available. (Now, state insurance departments get involved in surveying insurance losses from larger storms like Andrew and Hugo for years after-the-fact; for smaller storms, this is not normally done.) Also, when comparing simulated losses to actual losses from an earlier storm, it may be difficult to reconstruct the exposures, damageabilities, and policy coverages for purposes of the comparison, especially if the reconstruction is dated.
Despite these difficulties, comparisons of simulated and actual losses from specific storms are of some use in the validation of computer models. Unexplained, large divergences between simulated losses and actual losses would be cause for concern, as would a model's clear bias to either overstate or
understate losses. (Real-time simulations of losses from actual storms also can be useful to many of AIR's clients to project claims to be expected from a storm that is occurring or has just occurred.)
An industry-wide comparison of actual to simulated losses from 13 actual hurricanes since 1983 revealed that actual losses were 20% higher than the losses simulated by the AIR model. Of these 13 comparisons, seven (from 1989 through 1996) used real-time loss and exposure figures. These included Andrew in 1992, where actual losses were $16 billion compared to simulated losses of $13 billion. Significantly, actual losses were both above and below the simulated losses; the comparisons did not indicate a bias in either direction.
(These 13 storms also included Hurricane Fran in 1996, when actual losses were 2.56 times simulated losses. Ten inches of rain fell prior to Fran, especially in the area of Raleigh, North Carolina, and an unusually large number of trees toppled during the hurricane due to the pre-saturated condition of the soils.)
Losses simulated by the AIR model for certain companies also were compared to actual losses on an aggregated company basis. In this comparison, simulated losses were very close to actual losses for Hurricanes Erin, Opal, and Bertha.
AIR also compared simulated losses from the 10-year, 20-year, 50-year, and 100-year hurricanes generated by the AIR model to the losses from the 10-year, 20-year, 50-year, and 100-
year hurricanes in the "historical catalog" of Florida storms. The results were that losses from the storms in the AIR-simulated catalog approximated losses from the storms in the historical catalog.
The Department pointed out that, in six of seven examples in a previous submission by Nationwide, the AIR model simulated residential losses from particular storms that were higher than actual losses on a company basis. (The companies were not identified in the submission.) However, simulated losses often were not much higher than actual losses. For all seven examples combined, simulated losses were 17% higher than actual losses.
The previous submission by Nationwide also compared the AIR model's simulated residential losses from several storms with actual losses on an industry-wide basis. The Department pointed out that, in three of four examples, simulated losses were higher than actual losses. However, in those three examples, simulated losses were not much higher than actual losses. In addition, while the storms used in the comparison occurred in 1983 and 1985, the simulation used 1987 exposures; for that reason, simulated losses would be expected to be somewhat higher than actual losses. Meanwhile, in the example in which actual losses exceeded simulated losses, they were 19.5% higher. For all four examples combined, simulated losses were only 7% higher than actual losses on an industry-wide basis for those storms.
The previous submission by Nationwide included Florida
residential losses simulated by the AIR model for Hurricane Kate that were 34% higher than actual losses for the one (unidentified) company shown in the submission. (The Department proposed an erroneous finding that simulated losses were approximately 100% higher than actual losses for the shown in the submission.) On an industry-wide basis, simulated Florida losses for Kate were more than 100% higher than actual losses. For all states shown in the example, simulated losses were 55% more than actual losses.
As pointed out by the Department, the previous submission showed that the AIR model underestimated industry-wide Florida residential losses from Hurricane Elena by more than 300%. This anomaly was not explained. (The submission indicated that, for all states, the AIR model underestimated industry-wide residential losses from Hurricane Elena by just over 5%.)
For Nationwide's 1995 rate filing, AIR compared its model's simulated losses to Nationwide's actual losses from Hurricane Andrew. Simulated losses were approximately 50% greater than actual losses. However, this comparison was done several years after Andrew. While Nationwide's actual losses were quite accurate, AIR had to try to reconstruct Nationwide's exposures, damageabilities, and policy coverages. As mentioned, this can be difficult to do accurately.
During public hearings on its 1996 rate filing, Nationwide presented the results of a comparison between the losses simulated by the AIR model for Nationwide from Hurricanes
Hugo, Andrew, Erin and Opal and Nationwide's actual losses from those storms. The simulated losses were 13.6% higher for Hugo, 50% higher for Andrew, 1.8% higher for Erin and Opal combined, and 19.1% higher for all four storms combined. Excluding Hugo, the only one of the four that did not hit Florida, simulated losses from the other three storms were 28% higher than actual losses.
In information Nationwide provided to the Department on August 15, 1996, it was indicated that simulated losses were 12% higher for Hugo, 58% higher for Andrew, 12.5% lower for Erin, 12.7% higher for Opal, and 19.5% higher for all four storms combined. It was indicated that, excluding Hugo, simulated losses from the three storms that hit Florida were 32% higher than actual losses.
On November 20, 1996, Nationwide provided information to the Department on seven storms--Hugo, Andrew, Erin, Opal, Bob, Bertha, and Fran. This time, the simulated losses given for Hugo were $9 million lower so that simulated losses for Hugo were only 4.5% higher than actual losses, and the simulated losses given for Andrew were $11 million lower so that simulated losses for Andrew were only 30.6% higher than actual losses. Nationwide explained that the adjustment for these storms were made to correct for errors in Nationwide's coverages. The simulated losses for Bob were 9.3% higher than actual losses, and the simulated losses for Bertha were 8.2% higher than actual losses. For Fran, simulated losses were 48% less than actual losses;
Nationwide explained that unusually wet conditions immediately preceding Fran skewed the comparison. For all seven storms, including Fran, simulated losses were 13.5% less than actual losses. For just the Florida storms, simulated losses were 16.2% higher than actual losses.
The Department-Sponsored Procedure
Based primarily on the comparisons of simulated losses to "actual" losses from actual historical storms, the Department contends that Nationwide did not prove the AIR model to be completely reliable. Instead, the Department suggested the use of an alternative methodology described by the Department's witness, Ken Ritzenthaler. The Department proposed basing the hurricane component of Nationwide's rate filing on a procedure that utilized Nationwide's Cat Fund premium for a portion of Nationwide's rate and a modified ISO Excess Wind Procedure for the remainder of the rate.
Basing a portion of Nationwide's rate on the Cat Fund premium makes general sense. As it now stands, 48.5% of Nationwide's catastrophic hurricane losses will be paid by the Cat Fund. The Cat Fund premium is based on the AIR model. The
$117.74 premium Nationwide pays to the Cat Fund accounts for those hurricane losses.
For the remainder of projected hurricane losses, the Department suggested using the ISO Excess Wind Procedure with some modifications intended to address some of the procedure's inherent weaknesses. It suggested removing the regional
component of the excess wind procedure and limiting it to Florida experience, which increased the excess wind factor somewhat. It also increased the hurricane frequency in an attempt to address the higher frequency of hurricanes from 1900 to 1960 versus from 1960 to 1990. But it was not clear from the evidence how effective the Department's modifications were, and other weaknesses in the ISO Excess Wind Procedure were not addressed at all by those modifications.
It does not appear that the Department's modifications to the ISO Excess Wind Procedure were very effective in addressing the totality of the inherent weaknesses of the excess wind procedure. Even using the Cat Fund premium as the rate for almost half of projected hurricane losses, the Department developed indications that a rate increase averaging just 7.5% to 8% (assuming a 5% contingency factor) would be required, versus approximately 58% developed using the AIR model.
Since the ISO Excess Wind Procedure has been discredited, it makes no sense to combine it with a Cat Fund proxy for the AIR model. Since it has not been shown that the Department's modifications were effective in addressing the weaknesses in the ISO Excess Wind Procedure, use of the Department's combined Cat Fund/modified ISO Excess Wind Procedure is not supported by the evidence.
Creditability of AIR Model
The evidence was clear that no model is perfect, and neither AIR nor Nationwide claim that the AIR model is an
exception. While the Department's combined Cat Fund/modified ISO Excess Wind Procedure should not be credited, there are other models that may deserve credit. In addition to the AIR model, the Florida Commission has "accepted" two other computer simulation models--the RMS model, and the Equicat model. Each of these three models produce different loss cost results. On a statewide basis, the AIR model and the RMS model are fairly close for homeowner insurance; the Equicat model develops higher hurricane loss-costs for homeowner's insurance. On the other hand, the Equicat model develops lower hurricane loss-costs for mobile homeowner's insurance, and the AIR model develops the highest hurricane loss-costs for mobile homeowner's insurance.
Although Ritzenthaler testified that, in some cases, the differences in projected loss-costs among the three models "accepted" by the Commission are significant, he did not attempt to quantify the differences.
Ritzenthaler proposed developing a procedure that would somehow combine the results of all of three models "accepted" by the Commission. But he did not propose how this should be done,
and there is no evidence in the record from which it could be determined how to best do this. He also suggested giving some credence to the status quo, but he acknowledged that current rates are too low.
"Unknown" Construction Category
Over objection, the Department was permitted to introduce evidence that, in providing data for AIR to input into its model for purposes of projected Nationwide's hurricane loss- costs, Nationwide categorized the "construction type" of approximately 25% of Nationwide's insured properties as "unknown." It was later established that "unknown" represented frame/masonry veneer construction. Nationwide called this construction type "unknown" because the AIR model only had two categories--masonry and wood frame.
When given data that a percentage of properties covered by a company is of an unknown construction type, AIR would proportionately distribute those properties among the known construction types, either on an industry-wide or company-wide basis. (For Nationwide, at least, the applicable known construction types were "masonry" and "wood frame.") It was not clear from the evidence which option AIR used in the case of Nationwide's 1996 rate filing; nor were the proportions of known construction types (either masonry to wood frame, in the case of Nationwide, or whatever the known construction types are on an industry-wide basis).
Damage functions for wood frame construction are about 12% higher than for masonry construction. If AIR categorized all of the "unknown" construction types as wood frame, the AIR model could have overestimated damage by at most 4%.
Profit and Contingency
As indicated in Finding 9, supra, to arrive at the indicated premium requirement for the hurricane component of its rate filing, Nationwide divided estimated hurricane losses developed by use of the AIR model and ALAE per policy by 87.4% to account for agent commissions, state premium tax, profit and contingencies attributable to hurricane losses. In developing this percentage, Nationwide included in its financial needs model nothing for profit and 5% for contingency.
Nationwide used zero for profit factor because it was enjoying a reasonable rate of return of approximately 8.5% with just investment income on surplus and premiums collected. When confronted with this issue, Nationwide put on evidence suggesting that it actually needs a higher rate of return because its homeowner's book of business is leveraged at the rate of 2 to 1 (premium dollars to surplus dollars). But it is found that using zero for profit was reasonable for this filing and, in any event, is all Nationwide asked for.
The contingency factor is designed to account for under-projection of costs due to factors beyond a company's control. Historically, Nationwide has used a 5% contingency factor in its rate filings but still has sustained underwriting
losses of approximately 6% over the last 12 years. However, Nationwide concedes that some of this underwriting loss may have resulted from factors within Nationwide's control, such as not requesting the full indicated premium requirement in previous rate filings.
The Department contended that 5% was too high a factor for profit and contingency because the contingency factor did not address the factors under Nationwide's control. The Department concedes that a profit and contingency factor of 3.6% would have been acceptable. Using the profit and contingency factor recommended by the Department, Nationwide's rate indications would be reduced by approximately 2.5%.
CONCLUSIONS OF LAW
Rate Filings in General
Section 627.062, Florida Statutes (1997), provides in pertinent part:
The rates for all classes of insurance to which the provisions of this part are applicable shall not be excessive, inadequate, or unfairly discriminatory.
As to all such classes of insurance:
Insurers or rating organizations shall establish and use rates, rating schedules, or rating manuals to allow the insurer a reasonable rate of return on such classes of insurance written in this state. A copy of rates, rating schedules, rating manuals, premium credits or discount schedules, and surcharge schedules, and changes thereto, shall be filed with the department under one of the following procedures:
If the filing is made at least 90 days before the proposed effective date and the filing is not implemented during the department's review of the filing and any proceeding and judicial review, then such
filing shall be considered a "file and use" filing. In such case, the department shall finalize its review by issuance of a notice of intent to approve or a notice of intent to disapprove within 90 days after receipt of the filing. The notice of intent to approve and the notice of intent to disapprove constitute agency action for purposes of the Administrative Procedure Act. Requests for supporting information, requests for mathematical or mechanical corrections, or notification to the insurer by the department of its preliminary findings shall not toll the 90-day period during any such proceedings and subsequent judicial review. The rate shall be deemed approved if the department does not issue a notice of intent to approve or a notice of intent to disapprove within 90 days after receipt of the filing.
If the filing is not made in accordance
with the provisions of subparagraph 1., such filing shall be made as soon as practicable, but no later than 30 days after the effective date, and shall be considered a "use and file" filing. An insurer making a "use and file" filing is potentially subject to an order by the department to return to policyholders portions of rates found to be excessive, as provided in paragraph (h).
Upon receiving a rate filing, the department shall review the rate filing to determine if a rate is excessive, inadequate, or unfairly discriminatory. In making that determination, the department shall, in accordance with generally accepted and reasonable actuarial techniques, consider the following factors:
* * *
5. The reasonableness of the judgment reflected in the filing.
* * *
(e) After consideration of the rate factors provided in paragraphs (b), (c), and (d), a rate may be found by the department to be excessive, inadequate, or unfairly discriminatory based upon the following standards:
Rates shall be deemed excessive if they are likely to produce a profit from Florida business that is unreasonably high in relation to the risk involved in the class of
business or if expenses are unreasonably high in relation to services rendered.
Rates shall be deemed excessive if, among other things, the rate structure established by a stock insurance company provides for replenishment of surpluses from premiums, when the replenishment is attributable to investment losses.
Rates shall be deemed inadequate if they are clearly insufficient, together with the investment income attributable to them, to sustain projected losses and expenses in the class of business to which they apply.
* * *
(g) . . . The department shall issue a notice of intent to approve or a notice of intent to disapprove pursuant to the procedures of paragraph (a) within 90 days after receipt of the insurer's initial response. In such instances and in any administrative proceeding relating to the legality of the rate, the insurer or rating organization shall carry the burden of proof by a preponderance of the evidence to show that the rate is not excessive, inadequate, or unfairly discriminatory.
In this case, Nationwide opted for the "use and file" procedure under Section 627.062(2)(a)2, Florida Statutes (1997).
The Department argued at the final hearing that Nationwide's rate filing must stand entirely on its own and that no evidence other than the rate filing can be used to support it. This view was rejected on the ground that a formal proceeding under Chapter 120, Florida Statutes, is a de novo proceeding, and evidence other than the rate filing itself was received.
Similarly, the Department also seems to argue that Nationwide's rate filing can only be either approved or disapproved in its entirety and that, even if the evidence supports a larger rate increase than requested, the rate filing must be denied as being inadequate. This view is rejected as not
only illogical but also disingenuous. If current rates are inadequate and, for whatever reason, an insurer or rating organization seeks a smaller increase than the evidence would support, the Department should grant the requested increase. If the Department truly believes that the resulting increase is inadequate, the Department can initiate proceedings to increase the rates still further under Section 627.062(2)(g)-(h), Florida Statutes (1997).
Use of Models for the Hurricane Component
Section 627.0628, Florida Statutes (1995), provided in pertinent part:
LEGISLATIVE FINDINGS AND INTENT.–
Reliable projections of hurricane losses are necessary in order to assure that rates for residential property insurance meet the statutory requirement that rates be neither excessive nor inadequate. The ability to accurately project hurricane losses has been enhanced greatly in recent years through the use of computer modeling. It is the public policy of this state to encourage the use of the most sophisticated actuarial methods to assure that consumers are charged lawful rates for residential property insurance coverage.
The Legislature recognizes the need
for expert evaluation of computer models and other recently developed or improved actuarial methodologies for projecting hurricane losses, in order to resolve conflicts among actuarial professionals, and in order to provide both immediate and continuing improvement in the sophistication of actuarial methods used to set rates charged to consumers.
It is the intent of the Legislature to create the Florida Commission on Hurricane Loss Projection Methodology as a panel of experts to provide the most actuarially sophisticated guidelines and standards for projection of hurricane losses possible,
given the current state of actuarial science. It is the further intent of the Legislature that such standards and guidelines must be used by the State Board of Administration in developing reimbursement premium rates for the Florida Hurricane Catastrophe Fund, and may be used by insurers in rate filings under
s. 627.062 unless the way in which such standards and guidelines were applied by the insurer was erroneous, as shown by a preponderance of the evidence.
It is the intent of the Legislature that such standards and guidelines be employed as soon as possible, and that they be subject to continuing review thereafter.
* * *
ADOPTION AND EFFECT OF STANDARDS AND GUIDELINES.–
The commission shall consider any actuarial methods, principles, standards, models, or output ranges that have the potential for improving the accuracy of or reliability of the hurricane loss projections used in residential property insurance rate filings. The commission shall, from time to time, adopt findings as to the accuracy or reliability of particular methods, principles, standards, models, or output ranges.
In establishing reimbursement premiums for the Florida Hurricane Catastrophe Fund, the State Board of Administration must, to the extent feasible, employ actuarial methods, principles, standards, models, or output ranges found by the commission to be accurate or reliable.
With respect to a rate filing under s. 627.062, an insurer may employ actuarial methods, principles, standards, models, or output ranges found by the commission to be accurate or reliable. In reviewing the rate filing, the Department of Insurance must accept the insurer's use of such actuarial methods, principles, standards, models, or output ranges unless the department finds that the way in which such actuarial methods, principles, standards, models, or output ranges were applied by the insurer was erroneous, as shown by a preponderance of the evidence.
The commission shall adopt initial
actuarial methods, principles, standards, models, or output ranges no later than December 31, 1995. The commission shall adopt revisions to such actuarial methods, principles, standards, models, or output ranges at least annually thereafter. As soon as possible, but no later than July 1, 1996, the commission shall adopt revised actuarial methods, principles, standards, models, or output ranges which include specification of acceptable computer models or output ranges derived from computer models. As an interim measure, until the commission adopts actuarial methods, principles, standards, models, or output ranges which include specification of acceptable computer models or output ranges derived from computer
models, an insurer may use any computer model used by the State Board of Administration in setting reimbursement premiums for the Florida Hurricane Catastrophe Fund, which model shall have the same force and effect under paragraph (c) as a model adopted by the commission.
By enacting Section 6, Chapter 96-194, Laws of Florida (1996), the Legislature amended Section 627.0628(3)(c)-(d), Florida Statutes (1995), effective May 21, 1996, to read:
With respect to a rate filing under s. 627.062, an insurer may employ actuarial methods, principles, standards, models, or output ranges found by the commission to be accurate or reliable to determine hurricane loss factors for use in a rate filing under
s. 627.062, which findings and factors are admissible and relevant in consideration of a rate filing by the department or in any arbitration or administrative or judicial review.
The commission shall adopt initial actuarial methods, principles, standards, models, or output ranges no later than December 31, 1995. The commission shall adopt revisions to such actuarial methods, principles, standards, models, or output ranges at least annually thereafter. As soon as possible, but no later than July 1, 1996, the commission shall adopt revised actuarial methods, principles, standards, models, or output ranges which include specification of acceptable computer models or output ranges derived from computer models.
The AIR model used by Nationwide to support the rate filing it submitted on April 2, 1996, was the same AIR model being used by the State Board of Administration (SBA) in setting reimbursement premiums for the Florida Hurricane Catastrophe Fund (the Cat Fund) at the time. It is concluded that the purpose of Section 627.0628(3)(d), Florida Statutes (1995), was to allow Nationwide to use the AIR model in its rate filing and to require
the Department to "accept the insurer's use of such actuarial methods, principles, standards, models, or output ranges unless the department finds that the way in which such actuarial methods, principles, standards, models, or output ranges were applied by the insurer was erroneous, as shown by a preponderance of the evidence."
Although the transition between the old and new language is not smooth, the apparent purpose of the 1996 amendments to Section 627.0628, Florida Statutes (1995), was to remove the benefits of the interim measure in subsection (3). It appears that, as of the effective date of the amendments on
May 21, 1996, the Department no longer would be required necessarily to "accept" the model used by the SBA in setting reimbursement premiums for the Cat Fund in a rate filing; instead, the Department only would be required to consider the findings of the Commission as to the accuracy and reliability of actuarial methods, principles, standards, models, or output ranges, as well as hurricane loss factors derived from the application of those actuarial methods, principles, standards, models, or output ranges, as being admissible and relevant in determining hurricane loss factors in a rate filing.
The evidence indicates that the Florida Commission on Hurricane Loss Projection Methodology (the Commission) made no findings as to the AIR model used by Nationwide in its rate filing. The evidence also indicates that the Commission made no findings as to the revised AIR model until May, 1997. The
evidence is not clear as to exactly what the Commission had to say about the revised AIR model in May, 1997. But it is clear from the evidence that in October 1997 the Commission determined the revised AIR model "to be acceptable for projecting hurricane loss costs for personal residential rate filings submitted to the Department of Insurance in Florida" and determined that the revised AIR model "complies with the standards adopted by the Commission on May 29, 1997."
The Department has maintained throughout this proceeding that the Commission never "approved" the revised AIR model nor made any other findings as to the "accuracy and reliability" of the revised AIR model. But
Section 627.0628(3)(a), Florida Statutes (1995), required the Commission to "adopt findings as to the accuracy or reliability of particular methods, principles, standards, models, or output ranges." Section 627.0628(3)(b), Florida Statutes (1995), required the Cat Fund to use models "found by the commission to be accurate or reliable." Section 627.0628(3)(c), Florida Statutes (1995), allowed other insurers to use models "found by the commission to be accurate or reliable." When the statutes were amended effective May 21, 1996, Section 627.0628(3)(d), Florida Statutes (1997), required the Commission to "adopt . . . models." It is concluded that the import of the Commission's action in October 1997 was to find the AIR model to be "accurate and reliable" and to "adopt" it. Section 627.028(3), Florida Statutes (1997), requires that the Department consider those
determinations, as well as the loss factors derived from the use of the AIR model, as being admissible and relevant in determining hurricane loss factors.
Use of the AIR Model
Based on all of the evidence, and facts found, including the Commission's determinations and findings as to the revised AIR model, it is concluded that it was reasonable for Nationwide to develop the hurricane component of its 1996 rate filing using the AIR model. If the revised AIR model had been used instead of the version used in the 1996 rate filing, rate indications would have been approximately 5% higher.
It is possible that more refined rate indications could be obtained by somehow blending all three models that have been "accepted" by the Commission--the AIR model, the RMS model, and the Equicat model. But there was no evidence as to how this should be done, and it is concluded that use of the AIR model by itself does not result in projection of hurricane rates that are "excessive," as defined by Section 627.062(2)(e), Florida Statutes (1997).
In one respect, Nationwide may not have established that the way in which the AIR model was applied was proper in that masonry veneer construction was categorized as "unknown." However, at worst, this would have resulted in indications just 4% too high. At worst, such an error would fall well within the difference between Nationwide's indicated rate requirements (at
an average of 58%) and the requested rate increases (at an average of 27%).
Similarly, if the contingency factor used by Nationwide in its financial needs model was 1.4% too high, as the Department contends, the overall rate indications would be just 2.5% too high--again, well within the difference between Nationwide's indicated rate requirements and the requested rate increases.
In the worst case, assuming both a 2.5% error resulting from a miscategorization of construction type and a 1.4% error resulting from the selection of the contingency factor, the impact still would fall well within the difference between Nationwide's indicated rate requirements and the requested rate increases.
Based upon the foregoing Findings of Fact and Conclusions of Law, it is
RECOMMENDED that the Department of Insurance enter a final order approving Nationwide's request for residential property and casualty insurance premium rate increases averaging approximately 27%.
DONE AND ENTERED this 25th day of June, 1998, in Tallahassee, Leon County, Florida.
J. LAWRENCE JOHNSTON Administrative Law Judge
Division of Administrative Hearings The DeSoto Building
1230 Apalachee Parkway
Tallahassee, Florida 32399-3060
(850) 488-9675 SUNCOM 278-9675
Fax Filing (850) 921-6847
Filed with the Clerk of the Division of Administrative Hearings this 25th day of June, 1998.
COPIES FURNISHED:
W. Donald Cox, Esquire Fowler, White, Gillen, Boggs,
Villareal and Banker, P.A.
501 East Kennedy Boulevard, Suite 1700 Tampa, Florida 33602
Steve Parton, Esquire Jeffrey W. Joseph, Esquire Department of Insurance Division of Legal Services
200 East Gaines Street 612 Larson Building
Tallahassee, Florida 32399-0333
Bill Nelson, Commissioner
Department of Insurance and Treasurer The Capitol, Plaza Level 11 Tallahassee, Florida 32399-0300
Daniel Y. Sumner, General Counsel Department of Insurance and Treasurer The Capitol, Lower Level 26 Tallahassee, Florida 32399-0300
NOTICE OF RIGHT TO SUBMIT EXCEPTIONS
All parties have the right to submit written exceptions within 15 days from the date of this Recommended Order. Any exceptions to this Recommended Order should be filed with the agency that will issue the final order in this case.
| Issue Date | Proceedings |
|---|---|
| Sep. 15, 1998 | Final Order filed. |
| Jul. 24, 1998 | Nationwide`s Response to Exceptions of the Department of Insurance filed. |
| Jun. 25, 1998 | Recommended Order sent out. CASE CLOSED. Hearing held 04/28-30/98. |
| May 18, 1998 | Respondent`s Proposed Recommended Order filed. |
| May 15, 1998 | Motion to Extend Time for Filing Department`s Proposed Recommended Order filed. |
| May 15, 1998 | Nationwide`s Proposed Recommended Order filed. |
| May 05, 1998 | Notice of Filing; DOAH Court Reporter Final Hearing Transcript (Volumes 1, 2, 3, 4 tagged) filed. |
| Apr. 20, 1998 | (Joint) Prehearing Stipulation (filed via facsimile). |
| Jan. 13, 1998 | Notice of Final Hearing sent out. (hearing set for April 28 - May 1, 1998; 9:00am; Tallahassee) |
| Jan. 13, 1998 | Prehearing Order sent out. |
| Dec. 03, 1997 | (Respondent) Motion to Reset Hearing (filed via facsimile). |
| Oct. 21, 1997 | Status report (Respondent) (filed via facsimile). |
| Jun. 06, 1997 | (Respondent) Status Report filed. |
| Apr. 29, 1997 | (Respondent) Status Report filed. |
| Apr. 29, 1997 | (Respondent) Status Report filed. |
| Mar. 25, 1997 | Petitioner`s Notice of Serving Answers to Respondent`s First Set of Interrogatories; Nationwide`s Answers to Respondent`s First Set of Interrogatories filed. |
| Mar. 17, 1997 | (Respondent) Status Report (filed via facsimile). |
| Feb. 20, 1997 | Notice of Serving Response to Request for Production; Notice of Serving Answers to Interrogatories. filed. |
| Feb. 18, 1997 | (From J. Joseph) Notice of Entry of Appearance filed. |
| Feb. 17, 1997 | Order for Continuance and Status Reports sent out. (hearing cancelled; parties to file status report within 30 days) |
| Feb. 14, 1997 | (Respondent) Motion to Continue filed. |
| Nov. 18, 1996 | Notice of Final Hearing sent out. (hearing set for Feb. 24-28, 1997; 9:00am; Tallahassee) |
| Nov. 18, 1996 | Prehearing Order sent out. |
| Nov. 07, 1996 | Joint Response to Initial Order filed. |
| Oct. 28, 1996 | Initial Order issued. |
| Oct. 21, 1996 | Agency referral letter; Petition for Formal Proceedings; Notice of Rights; Agency Action letter filed. |
| Issue Date | Document | Summary |
|---|---|---|
| Sep. 15, 1998 | DOAH Final Order | |
| Jun. 25, 1998 | Recommended Order | Petitioner filed for an insurance rate increase mostly in hurricane component of rate filing. Rate indications were 58% using a computer model accepted by the Hurricane Commission. RO: Model supported 27% increase requested. |
