big data working group comments march 7, 2018 example: -very comprehensive -used examples that made...

$: Big Data Working Group Comments March 7, 2018 example: -very comprehensive -used examples that made the processes very clear -clarified something that was previously a mystery W:\National$
Big Data Working Group Comments

March 7, 2018

Alabama (Daniel Davis and Jerry Workman)

California (Ken Allen)

Connecticut (Wanchin W. Chou)

Texas (Rachael A. Cloyd)

American Insurance Association (Lisa Brown) and National Association of Mutual Insurance Companies (Paul T. Tetrault)

American Academy of Actuaries (Bob Beuerlein)

Chubb Insurance (John R. Marlow)

Insurance Services Office, Inc (Stephen C. Clarke)

Property Casualty Insurers Association of America (Deirdre Manna)

Risk & Regulatory Consulting (John Humphries, Dave Heppen, and Debbie Rosenberg)

ALABAMA (DANIEL DAVIS AND JERRY WORKMAN)

Here are the comments on these 3 documents from myself and Jerry Workman, Deputy Commissioner at the AL DOI.

Responding to the “Regulatory Framework Document”,

Here are our comments:

Are there adequate privacy standards for the use of consumer data? Probably not, and the Committee should strive to set some. Do consumers have data ownership rights? They should have. Should insurers be required to notify consumers regarding the use data about them, similar to notification requirements of the Fair Credit Reporting Act? We believe they should tell insureds which variables effect their rates, without going into too much specific model detail. Should consumers have the right to contest data and request corrections to data? Yes.

Should there be specific levels of correlation and/or causality for rating variables? “Causality”?? -- No. For example, poor credit scores don’t cause people to have accidents but they are correlated and quite predictive/useful to insurers. “Specific Levels of correlation”?? – it would be very hard to define this. If it improves the model, it should be included. That is the test.

Are regulators seeing additional risk segmentation Yes -- and is this having a positive or negative impact on consumers? There is less risk sharing across the various risk tiers. The best drivers (for example) are paying even less. The worst ones are paying even more. This fairer in the mathematical/economic sense. However it is likely that the uninsured motorist population will grow as a result of less risk sharing across the risk tiers.

Is there a need for additional regulatory oversight of data vendors? We believe so.

Consider passing Legislation that requires insurers to only use variables in their rating that can be verified and checked by an effected insured. This would put the onus on the big insurers to pressure data vendors to make themselves transparent to insureds.

REGULATORY REVIEW OF COMPLEX MODELS:

We should set up a centralized (confidential) NAIC Library of answer/explanations from insurers of their GLM/predictive analytics processes. Individual states can send the best, clearest explanations to the NAIC with a header paragraph that describes why the actuary in question felt the company response was so good:

1

For example: -very comprehensive -used examples that made the processes very clear -clarified something that was previously a mystery

W:\National Meetings\2018\Spring\TF\Innovation\Big Data\Comments\Alabama.docx

2

CALIFORNIA (KEN ALLEN)

With regard to the fifth bullet under Action 1, we recommend that the language be revised to clarify how or the extent to which this software, databases and other technology will be used to assist in the analysis of models. For instance, is the expectation that the NAIC, either by developing or purchasing a license for a pre-existing model, will be able to replicate the output of an insurer’s model? Or is it rather the intent that any such model will serve as a “reasonableness” check of the insurer’s output? Or both? W:\National Meetings\2018\Spring\TF\Innovation\Big Data\Comments\California.docx

3

January 12, 2018 Ms. Elizabeth Kelleher Dwyer, Vice Chair Big Data (EX) Working Group National Association of Insurance Commissioners 1100 Walnut Street, Suite 1500 Kansas City, MO 64106-2197 Re: Comments on Big Data Working Group-State Regulatory Review of Complex Models Dear Ms. Dwyer: On behalf of the State of Connecticut Insurance Department (CID), we appreciate the opportunity to provide comments on the proposal that was exposed by the Big Data WG (EX) on December 19, 2017. We should respect and protect companies’ intellectual property and encourage them to innovate. Only when we can assure appropriate confidentiality protection of proprietary information of the insurers and data vendors will we be able to receive and validate the newly developed models and manage the predictive modeling risks/impacts properly. On behalf of the CID, I would like to comment as follows Attachment A - State Regulatory Review of Complex Models

(1) Summary paragraph: The following describes the principles and structure for a mechanism to assist state regulatory review of complex models:

• State regulators will maintain their current rate regulatory authority. • State regulators will work to share information that aids speed to market. • State regulators will share expertise and discuss technical issues regarding complex

predictive models. • State regulators will seek legal assistance to assure each state’s confidentiality

provisions apply.

4

In order to protect the intellectual property of the insurer and data vendor and to encourage innovation, CID would suggest a change to replace “apply” (underlined above) with the words: “are applied properly.”

(2) Attachment A Item 1: The Big Data (EX) Working Group will request the Casualty Actuarial and Statistical (C) Task Force to appoint a Predictive Analytics (C) Working Group with following 2018 charges: o Facilitate discussion among regulators regarding rate filing issues of common

interest across states through the scheduling of regulator-to-regulator conference calls. Seek legal assistance to ensure the states can share confidential information during these discussions and that states’ confidentiality protections apply. [Note: The first sentence is a current Casualty Actuarial and Statistical (C) Task Force charge.]

For the above paragraph, the discussion among regulators should focus on the issues of common interest, and higher standards of confidential agreements should be applied to protect the intellectual property of the insurers and data vendors. Connecticut suggests that the term common interest be emphasized and that the sentence “Seek legal assistance to ensure…,” underlined above, be strengthened and replaced with “Proprietary information of a filing company or data vendor will not be discussed or exchanged unless and until there is legal assurance that each state’s confidentiality protections apply to such discussion(s).” Thank you for considering our comments. We look forward to discussing our comments with you. Wanchin W. Chou, FCAS, MAAA, CPCU Chief Insurance Actuary State of Connecticut Insurance Department Mail: PO Box 816, Hartford, CT 06142-0816 Del: 153 Market St., Hartford, CT 06103 Phone: 860-297-3943 CC: Timothy Curry, Deputy Commissioner George Bradner, Director – PC Product Wanchin Chou, Chief Insurance Actuary

5

TExAs DEPARTMENT OF INSURANCE333 Guadalupe, Austin, Texas 78701 * P0 Box 149104, Austin, Texas 78714-9104(512) 676-6000 (800) 578-4677 TDI.texas.gov I @TexasTDl

January 12, 2018

Superintendent Elizabeth Kelleher Dwyer, Vice Chair VIA EMAIL ONLY:NAIC Big Data (EX) Working Group [email protected]: Tim Mullen, NAIC Director of Market Regulation1100 Walnut St, Suite 1500Kansas City, MO 64106

RE: Regulatory Review of Complex Models, Proposed Framework, and Data Needs

Dear Superintendent Kelleher Dwyer and working group members:

Thank you for this opportunity to comment on the group’s efforts to aid state regulatory review ofcomplex models used in support of personal auto and homeowners insurance rate filings, theproposed framework, and the assessment of regulatory data needs. TDI appreciates the workinggroup’s efforts in working toward implementation of Charges A, B, and C.

Most of TDI’ s comments and concerns focus on seeking clarification of statements made in eachattachment. TDI submits the following comments and suggested improvements to the documentsfor the working group’s consideration:

Attachment A — Regulatory Review of Complex Models

TDI has concerns about portions of Attachment A, similar to concerns TDI previously raised inour July 28, 2017 comments to the working group. Charge B for the working group states that theproposed mechanism for review of complex models “shall respect and in no way limit states’regulatory authority.” Although we believe it is not the intent, the wording and tone of AttachmentA appears to mandate participation in shared, multi-state review of complex predictive models.For example, bullets 3 and 4 both state that regulators “will share expertise and discuss technicalissues” and “will seek legal assistance.” (emphasis added).

Thus, TDI reiterates its July 2017 comments: Ratefiling review rests with the individual states;thus, state participation in any review mechanism should be optional. We make the followingsuggestions to address our concerns:

1. The word “will” in bullet 3 should be replaced with the word “may.”

2. We request revision and clarification on bullet 4, which mandates each state to seek legalassistance regarding confidentiality. TDI’ s concerns with a shared mechanism for reviewof complex models stretches beyond confidentiality. As discussed in our July 2017comments, because insurance companies are entitled to due process should a rate filing bedisapproved by a state, a shared review mechanism may draw any and all reviewers intoany participating state’s rate litigation, as witnesses, whether intended or not. In addition,

6

NAIC Big Data (EX) Working GroupJanuary 12, 2018

Page 2 of 5

should some mechanism ultimately be adopted that contemplates states sharingconfidential portions of rate filings, it may be appropriate for the NAIC to seek legalassistance to review the scope of the NAIC Sharing Agreement, and determine if its scopeprovides any protections for states participating in such a mechanism, or if the agreementneeds expansion or amendment to encompass the sharing of confidential informationduring complex model reviews.

Absent any such protections in the NAIC Sharing Agreement, TDI suggests replacementof bullet 4 to state:

If state regulators opt to participate in multi-state regulatory review ofcomplex models, each state is individually responsible for seeking its ownlegal assistance regarding: protection of confidential information sharedduring multi-state reviews of rate filings; assurance that the participatingstate is authorized to share confidential information with participatingstates under both civil and criminal laws; and, risks related to civillitigation resultingfrom shared, multi-state review of rate filings.

3. Under No. 1, bullet 2, we request clarification on the suggestion that the PredictiveAnalytics Working Group (PAWG) recommend filing requirements. Because each statehas different filing requirements, is the intent to charge PAWG with drafting a model lawsetting rate filing requirements for complex models? If so, then perhaps, that should bemore clearly expressed.

4. Under No. 1, bullet 3, TDI again expresses concern with the second sentence, and requestsclarification on the mandate to seek legal assistance. Choosing to participate in a multistate discussion of a specific rate filing with a complex model necessitates considerationof legal issues beyond confidentiality. TDI again recommends revision of this sentenceconsistent with our suggestion in number 2 above.

In addition, the first sentence of No. 1, bullet 3, acknowledges the current charge of the CasualtyActuarial and Statistical Task Force (CASTF). These regulator-to-regulator calls discuss ratefiling issues of common interest, and while the topic of complex models may be raised onoccasion, the calls are not specific to sharing expertise and discussing technical issues regardingcomplex predictive models. These calls of broad and common interest are helpful to regulators.Perhaps the working group should consider whether PAWG should facilitate and schedule callsto discuss complex predictive models separate and apart from the CASTF calls being presentlyconducted?

7


Page 3 of 5

Attachment B — Proposed Framework

TDI requests two changes in Attachment B:

1. On page 1, under the “Issues Raised by Regulators,” TDI suggests insertion of this questionas number 6:

Is the data provided by vendors to insurers trustworthy and reliable, andwhat assurances do insurers getfrom vendors, by contract or otherwise, toensure that the data provided is trustworthy and reliable?

2. On pages 4-5, TDI requests correction of an out-of-date citation to Commissioner’’i OrderNo. 2691, dated August 7, 2013, as an example of a state law that addresses filingrequirements to adjust for the use of new data and predictive modeling.

Order No. 2691 adopted amendments to the Texas Basic Manual ofRules, Classificationsand Experience Rating Plan for Workers’ Compensation and Employers’ LiabilityInsurance. Those amendments included the addition of Rule VI-M concerning the use ofmodeled rating factors to calculate premium in the Texas Basic Manual. Later, in OrderNo. 3142, the Commissioner adopted the National Council on Compensation InsuranceBasic Manual for Workers Compensation and Employers Liability Insurance with TexasExceptions, and the national and Texas-specific endorsements and forms in the NCCIForms Manual of Workers Compensation and Employers Liability Insurance. The NCCImanuals must be used for Texas workers’ compensation policies with an effective date onor after 12:01 a.m., October 1, 2014. Thus, Order No. 2691 and Rule VI-M is considereda historical reference for policies issued before that date.

Notwithstanding the change in manuals, the NCCI Basic Manual with Texas Exceptionsdoes contain and allows use of the modeled rating factor (MRF) as an optional factor whencalculating premium, and predictive modeling may be used to determine the MRF. Toaddress the change in manuals and correct the citation to the order adopting the NCCI BasicManual, TDI provides the enclosed Supplement containing a mark-up of our suggestedrevisions to pages 4-5.

Technicalities

Lastly, TDI also suggests correction of the following grammatical errors:

• In Attachment A, No. 1, bullet 5, we suggest inserting the word “with” in the sentence asfollows: “to assist fflh analysis of predictive models.”

• In Attachment B, under Issues Raised by Industry, no. 2, we suggest inserting the word“of’ in the question as follows: “barriers to the use f data by insurers.”

8


Page 4 of 5

• In Attachment B, under Part I: Rating, third paragraph, we suggest inserting the word “to”in the sentence as follows: “A rate in a competitive market is presumed not be excessive

• In Attachment C, second paragraph, first sentence, we suggest revision to clarify whocollects the data identified: “To begin this discussion, the Working Group requested NAICstaff to identify the data it currently collects current data collected and how regulators .

Again, TDI appreciates the working group’s efforts. We hope our comments are productive inreaching clarification on issues that are significant to TDI and all states’ regulatory efforts.

Thank you for your time, attention, and consideration.

Respectfully,

Rachel A. Cloyd, JD, CP UDirector, Regulatory Analysis OfficeEnforcement Section I Legal & Enforcement DivisionTexas Department of InsuranceDirect: (512) 676-6349 I Fax: (512) [email protected]

Enclosurecc: J’ne Byckovski, Director and Chief Actuary, Property and Casualty Actuarial Office

9


Page 5 of 5

SUPPLEMENT: REQUESTED REVISIONS TO A TTACHMENT B, PAGES 4-5

Texas Regulation (Commissioner’s Order No. 3142 2691, dated March 21, 2014 August 7, 2013)

The amendments to Rule VI add section M (Modeled Rating Factor), which provides anexplanation of the MRF, describes its application, and lists the insurance carrier’s requirements touse the lv[RF. The commissioner has deleted the word “factor” after “MRF” in Rule VI, new sectionM.2.c, as proposed, because it is duplicative.

The Texas Exceptions to the NCCI Basic Manual allow use of a Modeled Rating Factor (MRF).The MRF is an optional factor that insurance carriers can file with TDI and apply when calculatingworkers’ compensation premium. The MRF takes into consideration individual risk characteristicsand loss experience of an insured. Insurers may use predictive modeling to determine the MRF.The term MRF can include &tier rating factor and other similar terms.

Under the amendments to Rule III E Texas Workers Compensation Premium Algorithm, an insurerwill apply its MRF to the policy in a multiplicative manner, after the application of the experiencerating modification, and before application of schedule rating, premium discounts, and the expenseconstant. an4 The insurer must not apply or use the MRF in a way that duplicates other ratingfactors, such as schedule and experience rating factors. Once determined, the MRF applies wil4apply during the entire policy period. Insurance carriers are will be required to evaluate eachpolicy’s characteristics and experience at each renewal to determine the MRF for the renewalpolicy. The MRFs filed by a carrier with TDI must apply to all policies for the carrier.

The amendments to Rule VI require insurance carriers Carriers are required to make a filing withTDI under Title 28, Texas Administrative Code, Chapter 5, Subchapter M (Filing Requirements)before using an MRF. The filing must include the MRFs; the characteristics, variables, or criteriaused to determine the MRFsr; actuarial support and other documentation supporting fef the MRFsand other supporting documentation.

The commissioner has determined that the amendments to the manual are necessary for insuranceciers to use MRFs in calculating workers’ compensation rates or premiums. The proposed filingrequirement is necessary to promote transparency and accountability in the use of MRFs.Including an MRF in premium calculations allows an insurance carrier to tailor premiums moreprecisely to each insured by including an insured’s specific risk characteristics and loss experience.With a more precise risk assessment, the insurance carrier can come closer to charging theappropriate premium for the risk each insured actually presents.

10

January 26, 2018 Elizabeth Kelleher Dwyer (RI) Vice Chair of the Big Data (EX) Working Group National Association of Insurance Commissioners Via Email to: Tim Mullen, NAIC Director of Market Regulation [email protected]

Re: Big Data (EX) Working Group Exposed Documents for Comment

Dear Superintendent Dwyer:

On behalf of the American Insurance Association and National Association of Mutual Insurance

Companies, we appreciate your consideration of these comments regarding the following Big Data (EX)

Working Group exposures circulated via email by NAIC staff on December 19, 2017:

1. State Regulatory Review of Complex Models 2. Discussion of Regulatory Framework 3. Assessment of Regulatory Data Needs

1. State Regulatory Review of Complex Models

As a general matter, we believe it would be helpful to have some clarification regarding what exactly

this document is meant to be. It appears to us to be something of a response to concerns that the

industry raised regarding the Predictive Analytics Team proposal circulated last summer. We would like

to know whether the new document is meant to essentially replace the previous document or to

somehow augment it.

The document’s first section states it describes “principles and structure” for a mechanism to assist state

regulatory review of complex models. In fact, however, the four following bullet points describe

principles and perhaps intent only but do not address structure. The bullet points are largely

unobjectionable, but we suggest minor, but clarifying changes be made to address the concern

regarding potential establishment of a regulatory role for the NAIC. We suggest the first bullet be

changed to:

• State regulators will retain regulatory authority and regulatory decision-making.

With regard to bullets two and three, we suggest adding language to indicate that regulators share

information and expertise “if permissible under state laws.” We suggest that the final bullet should be

changed to:

• State regulators will consult with their legal division to assure each state’s confidentiality provisions apply.

11

mailto:[email protected]

Regarding section 1 and the proposal for the Casualty Actuarial and Statistical (C) Task Force (CASTF) to

appoint a Predictive Analytics (C) Working Group, we would note the recent move to consolidate

committees, task forces and working groups and ask whether the activities contemplated could be

completed by an existing NAIC entity.

We note that the first bullet point under section 1. refers to the Product Filing Examiners Handbook but

were unable to identify such work product and suggest the reference may be to the Product Filing

Review Handbook. We would be interested to review potential changes to the Handbook after they are

developed and suggest that the next bullet point should probably be left for one of those proposed

changes, if warranted. Further, we suggest adding language to the first bullet to acknowledge that

insurers may require NDA or other confidentiality agreements before sharing information with

consultants.

The next two bullet points, regarding facilitating discussion among regulators and training and sharing of

expertise, are as noted already being pursued by CASTF so we are not clear why there is a need to

propose them here. We suggest, however, that these activities by CASTF be the subject of regular

reports to the Big Data Working Group. In particular with regard to any regulator-only conference calls

we believe it would be appropriate for CASTF to report on matters such as the quantity and scope of

items discussed without compromising confidentiality.

Regarding section 2 we would request more development of the concept proposed before any action is

taken. It is not clear for instance what research might be conducted or what the “needs of the NAIC

Membership” are regarding model review.

2. Discussion of Regulatory Framework

The second paragraph of the document is a single line stating, “The Working Group wants to balance

consumer protection with industry innovation.” We suggest deleting this sentence as it seems to imply

that consumer protection may have to be eased in some way in order to allow for innovation and we do

not believe that to be the case. Rather, we think that innovation in expanding the use of data has and

will continue to produce significant benefits for consumers and is not in conflict with consumer

protection. In fact, some innovations such as those enhancing anti-fraud efforts, may be thought of as

providing greater consumer protection.

Next, we have concerns regarding how some of the content of “Part III: Consideration of Specific

Variables” is presented. We do not necessarily agree, for instance, that the second and third questions

in the first paragraph, “Does the variable have a correlation to a prohibited factor, such as race or

income?” and “Does the variable lead to unfair discrimination?” are “generally touch[ed] upon” in

“[p]olicy discussions on whether a rating variable should be allowed.” Certainly those questions have

been posed, but that does not make them valid in the context of insurance. Absent discriminatory

treatment or failing to match price to the risk, the issue is whether they are even appropriate inquiries

to apply to insurance rating. This is especially the case since some states prohibit even asking about the

applicant’s or policyholder’s race or some other protected class status. As a result, the rating for a

particular risk is truly color blind.

12

We take specific issue with the way in which number 4, Education and Occupation, is presented, with a

vague reference to “concerns” that “have been expressed” followed by the observation that “few, if any

states have prohibited” their use. This section seems to imply both the validity of the “concerns” and a

failure on the part of states to act in response. In fact, however, these factors have been closely

examined by at least two states, New Jersey1 and Maryland, both of which found their use to be

appropriate. We suggest revising this section to provide a factual account of those reviews in response

to raised concerns, as well as the recently adopted New York regulation2 requiring demonstration that

such use does not violate insurance rating law.

We are also uncertain as to the basis and potential relevance of the last paragraph in this section and

suggest it be deleted.

3. Assessment of Regulatory Data Needs

In our view this document does a commendable job of laying out the scope and volume of data that is

collected by and available to state insurance regulators. It illustrates the numerous ways in which the

system of state insurance regulation is supported by the delivery and analysis of useful data. The

breadth of data collected, in our view, would suggest there is not a need for greatly expanding the

collection of data for regulatory purposes. If this assessment does result in any additional collection of

data we urge that any actions be subject to a rigorous cost-benefit analysis in recognition of the fact that

data reporting invariably involves costs that affect the price of insurance purchased by consumers. We

would also like to take this opportunity to note for the Working Group that the National Conference of

Insurance Legislators on November 19 of last year adopted a Resolution Encouraging the Adoption of

Voluntary Data Call Principles3 which we believe provides an excellent framework for regulators to apply

when considering the use of a data call.

We appreciate your consideration of these comments and look forward to the continued deliberations

of the Working Group.

Sincerely,

Lisa Brown, JD, MCM Assistant General Counsel & Director, Compliance Resources American Insurance Association

Paul T. Tetrault, JD, CPCU, ARM, AIM State and Policy Affairs Counsel National Association of Mutual Insurance Companies

1 http://www.state.nj.us/dobi/division_insurance/pdfs/ed_occ_april2008.pdf 2 http://www.dfs.ny.gov/insurance/r_finala/2017/rf150a2txt.pdf. 3 http://ncoil.org/wp-content/uploads/2017/11/Data-Call-Resolution-Final-1.pdf

13

http://www.state.nj.us/dobi/division_insurance/pdfs/ed_occ_april2008.pdf

http://www.dfs.ny.gov/insurance/r_finala/2017/rf150a2txt.pdf

http://ncoil.org/wp-content/uploads/2017/11/Data-Call-Resolution-Final-1.pdf

1850 M Street NW Suite 300 Washington, DC 20036 Telephone 202 223 8196 Facsimile 202 872 1948 www.actuary.org

1

January 12, 2018

Commissioner Elizabeth Kelleher Dwyer

Vice Chair, Big Data (EX) Working Group

National Association of Insurance Commissioners (NAIC).

Via email: [email protected]

RE: State Regulatory Review of Complex Models, Discussion of Regulatory Framework, and

Assessment of Regulatory Data Needs (December 19, 2017)

Dear Commissioner Dwyer,

On behalf of the Big Data Task Force of the American Academy of Actuaries,1 I would like to

offer the following comments on the three documents (State Regulatory Review of Complex

Models, Discussion of Regulatory Framework, and Assessment of Regulatory Data Needs) that

were exposed by the NAIC’s Big Data (EX) Working Group on December 19, 2017.

The Academy’s Big Data Task Force carefully reviewed the exposures and we believe the

Academy can support regulators with their work concerning big data on a number of levels. We

believe the incorporation of actuarial principles and techniques is critical in advancing and

implementing the frameworks outlined in the current exposures and in the Working Group’s efforts

more generally. To this end, we suggest specifically adding the following to the Working Group’s

charges:

“Work with the American Academy of Actuaries to provide information, training, and

education to regulators as a continuation of the Working Group’s collaboration with the

Academy in 2017.”

The Academy’s Big Data Task Force is available to discuss some or all of the following,

particularly as they relate to processes for reviewing advanced models:

Identifying risk drivers with data analytics;

Checklists for regulators to use;

Strengths and weaknesses of various models that are being used;

Financial data reconciliation;

Reasonable timeframes for reviewing models;

1 The American Academy of Actuaries is a 19,000-member professional association whose mission is to serve the public and the

U.S. actuarial profession. For more than 50 years, the Academy has assisted public policymakers on all levels by providing

leadership, objective expertise, and actuarial advice on risk and financial security issues. The Academy also sets qualification,

practice, and professionalism standards for actuaries in the United States.

14


1850 M Street NW Suite 300 Washington, DC 20036 Telephone 202 223 8196 Facsimile 202 872 1948 www.actuary.org

2

Analysis of quality of responses to regulators’ inquiries;

Data requests; and

Insurtech and Regtech.

Furthermore, the Academy welcomes the opportunity to collaborate with the Working Group in

planning for the NAIC’s 2018 Insurance Summit, particularly should there be interest in the

development of a day-long big data training session for regulators.

*****

We appreciate your time and attention to our comments. If you have any questions or would like

to further discuss this topic, please contact Nikhail Nigam, the Academy’s policy analyst for risk

management and financial reporting issues, at 202-223-8196 or [email protected].

Sincerely,

Bob Beuerlein, MAAA, FSA, FCA, CERA

Chairperson, Big Data Task Force

American Academy of Actuaries

15


Attachment B Request for Comments – Deadline of Jan. 12, 2018

Draft: 11/28/17

Regulatory Framework Proposed Structure and Issues for Discussion

Charge A of the Working Group: “Review current regulatory frameworks used to oversee insurers’ use of consumer and non-insurance data. If appropriate, recommend modifications to model laws/regulations regarding marketing, rating, underwriting and claims, regulation of data vendors and brokers, regulatory reporting requirements, and consumer disclosure requirements.”

The Working Group wants to balance consumer protection with industry innovation.

The document “Background Information for Discussion of Regulatory Framework” distributed at the NAIC Summer National Meeting provides a summary of NAIC models addressing the use of data in rating and claims for Property and Casualty Insurance. The document also summarizes some unfair trade practice considerations and consideration regarding the use of specific data variables. Finally, the document primarily focused on the standards from the following three NAIC models: (1) NAIC Property and Casualty Model Rating Law – File and Use Version (#1775); (2) NAIC Unfair Trade Practice Act (#880); and (3) NAIC Unfair Claims Settlement Practices Act (#900).

Based upon the Working Group’s review of the document, there was a general recognition for the need to first identify the scope of issues to be addressed. After the scope of issues is identified, a matrix could be created to (1) identify the regulatory issues; (2) identify the regulatory framework/standard applicable to each issue; and (3) identify whether revisions to the regulatory framework/standard need to be made.

The following issues have been identified through the Working Group’s prior discussions and comments submitted by interested parties.

Issues Raised by Consumers 1. Should consumers have the right to contest data and request corrections to data?2. Are there issues specific to a particular line of insurance?

Issues Raised by Industry 1. Do insurers and data vendors have appropriate confidentiality protections of intellectual property when submitting

models to regulators? 2. Are there regulatory standards that are barriers to the use data by insurers? 3. Are there issues specific to a particular line of insurance?

Issues Raised by Regulators 1. Do regulators have appropriate access to insurers’ models through the current rate filing process?2. Are there any data variables that should be prohibited?3. Should there be specific levels of correlation and/or causality for rating variables?4. Are regulators seeing additional risk segmentation and is this having a positive or negative impact on consumers? 5. Is there a need for additional regulatory oversight of data vendors? 6. Are there issues specific to a particular line of insurance?

7. Are appropriate predictive models based upon data desirable because they assure fair and equitable treatment ofconsumers, and avoid subjective underwriting?

Preliminary Discussion Draft

Deleted: <#>Are there adequate privacy standards for the use of consumer data?¶<#>Do consumers have data ownership rights?¶<#>Should insurers be required to notify consumers regarding the use data about them, similar to notification requirements of the Fair Credit Reporting Act? ¶

Comment [HIG1]: This is already being addressed in other working groups. Addressing here as well could lead to cross-purpose conflicting regulations.

Comment [HIG2]: See above comment.

Comment [HIG3]: We do not believe that this is appropriate. We are unclear as to why the use of enhanced underwriting information would create notice requirements. Underwriting is all about the use of all available information, including publicly available information. We believe that this could have unintended consequences and make the underwriting of risks more difficult for consumers.

Comment [HIG4]: Not sure how this would work – are we saying that a consumer could protest how the company handles underwriting? For example, if the structure of the person’s home is fire resistant or not? A consumer can do this anyway today, so not sure why there needs to be a change.

Comment [HIG5]: Agree that these need to be enhanced – If an insurer spends time and money developing a model, we do not believe it is fair that such model becomes a matter of public record in the rate filing and the regulators should take steps to ensure the confidentiality of such information

Comment [HIG6]: There appears to be regulatory distrust of predictive modeling

Formatted: List Paragraph, Space After: 0 pt,Add space between paragraphs of the samestyle, Line spacing: single, Numbered + Level:1 + Numbering Style: 1, 2, 3, … + Start at: 1+ Alignment: Left + Aligned at: 0.25" +Indent at: 0.5"

Formatted: Font: 10 pt

© 2017 National Association of Insurance Commissioners 1

Comments from Chubb

16


Big Data (EX) Working Group Background Information for Discussion of Regulatory Framework

Charge A

The first charge of the Big Data (EX) Working Group is to “review current regulatory frameworks used to oversee insurers’ use of consumer and non-insurance data. If appropriate, recommend modifications to model laws/regulations regarding marketing, rating, underwriting and claims, regulation of data vendors and brokers, regulatory reporting requirements, and consumer disclosure requirements.”

To begin this discussion, the Working Group requested NAIC staff to identify existing state authority, NAIC model laws, model regulations, and guidelines addressing insurers’ use of consumer and non-insurance data. The initial focus of this review is on insurance companies’ use of data for rating and claims in personal lines Property and Casualty insurance.

Part I: Rating

General Insurance Rating Framework

Pursuant to the established regulatory framework found in the NAIC’s Model Rating Laws (Property and Casualty Model Rating Law – File and Use Version (#1775); Property and Casualty Model Rate and Policy Form Law Guideline (#1776); Property and Casualty Model Rating Law – Prior Approval Version (#1780), rates shall not be excessive, inadequate or unfairly discriminatory. With the exception of prohibiting any risk classification from being based upon race, creed, national origin, or the religion of the insured, the models do not prescribe what data cannot be used for rating.

The following guidance is set forth in Model 1775.

A rate in a competitive market is presumed not be excessive and a rate in a noncompetitive market is considered excessive if it is likely to produce a profit that is unreasonably high for the insurance provided or if expenses are unreasonably high in relation to services rendered.

A rate is not inadequate unless the rate is clearly insufficient to sustain projected losses, expenses and special assessments in the class of business to which it applies and the use of such rate has or, if continued, will have the effect of substantially lessening competition or the tendency to create monopoly in any market.

A rate is unfairly discriminatory if, after allowing for practical limitations, price differentials fail to reflect equitably the differences in expected losses and expenses. A rate is not unfairly discriminatory if it is averaged broadly among persons insured under a group, franchise or blanket policy or a mass marketed plan.

In determining whether a rate is excessive, inadequate or unfairly discriminatory, the following criteria are to be applied: 1. Basic factors in rates. Due consideration shall be given to past and prospective loss experience within and outside this

State; to the conflagration and catastrophe hazards; to a reasonable margin for profit and contingencies; to dividends,savings, or unabsorbed premium deposits allowed or returned by insurers to their policyholders, members or subscribers; to past and prospective expenses both countrywide and those specially applicable to this State; and to provisions forspecial assessments and to all other relevant factors within and outside this State.

2. Classification. Risks may be grouped by classifications for the establishment of rates and minimum premiums.Classification rates may be modified to produce rates for individual risks in accordance with rating plans which establishstandards for measuring variations in hazards or expense provisions, or both. Such standards may measure anydifferences among risks that can be demonstrated to have a probable effect upon losses or expenses. No riskclassification, however, may be based upon race, creed, national origin or the religion of the insured.

3. Expenses. The expense provisions included in the rates to be used by an insurer shall reflect the operating methods of theinsurer and its anticipated expenses.

4. Profits. The rates may contain provision for contingencies and an allowance permitting a reasonable profit. Indetermining the reasonableness of the profit, consideration shall be given to all investment income attributable to the lineof insurance.

After setting forth the standards for whether or not a rate is excessive, inadequate or unfairly discriminatory, Model 1775 sets


Comments from Chubb

17


forth information that should be included in a rate filing as follows:

Every insurer, except those insurers who met the commercial deregulation requirements, shall file with the commissioner, except as to inland marine risks which are not written according to manual rates or rating plans, every manual, minimum premium, class rate, rating schedule or rating plan and every other rating rule, and every modification of any of the foregoing which it proposes to use.

Every insurer, except as noted above, shall file or incorporate by reference to material which has been filed with or approved by the commissioner, at the same time as the filing of the rate, all supplementary rating and supporting information to be used in support of or in conjunction with a rate. The information furnished in support of a filing may include or consist of a reference to: (a) the experience or judgment of the insurer or information filed by the advisory organization on behalf of the insurer; (b) its interpretation of any statistical data it relies upon; (c) the experience of other insurers or advisory organizations; or (d) any other relevant factors. In addition, insurers utilizing the services of an advisory organization must provide with their rate filing, at the request of the commissioner, a description of the rationale for such use, including its own information and method of utilization of the advisory organization’s information.

Within this framework, insurance companies have been permitted to use rating variables that are predictors of risk based upon actuarial judgement, which includes “assumptions on the input and assessments on the accuracy of the results.” (NAIC Price Optimization White Paper – Page 4)

Examples of State Laws

Provided below are some examples of state laws that address filing requirements to adjust for the use of new data and predictive modeling.

New Hampshire Law (55:8 Property and Casualty Insurance; Rate Filings) Every insurer shall file with the commissioner every manual, predictive models or telematics models or other models that pertain to the formulation of rates and/or premiums, minimum premium, class rate, rating schedule or rating plan and every other rating rule, and every modification of any of the foregoing which it proposes to use. Personal lines filings shall include underwriting rules used by insurers or a group of affiliated insurers to the extent necessary to determine the applicable rate and/or policy premium for an individual insured or applicant. An insurer may file its rates by either filing its final rates or by filing a multiplier and, if applicable, an expense constant adjustment to be applied to prospective loss costs that have been filed by an advisory organization on behalf of the insurer as permitted by RSA 412:23. Every such filing shall state the effective date, and shall indicate the character and extent of the coverage contemplated. Information contained in the underwriting rules that does not pertain to the formulation of rates and/or premiums shall be identified by the filer as proprietary and shall be kept confidential by the department and shall not be subject to the provisions of RSA 91-A.

Nevada Regulation (Regulatory Activity Bulletin 17-001) The Division issues this bulletin to remind insurers that any mathematical model used in underwriting or rating of any personal line of property and/or casualty insurance, or other line of property and/or casualty insurance subject to regulation of rates pursuant to NRS 686B.030, must be filed with the Division for prior approval pursuant to NRS 686B.110.

Among the information required to be filed with the Division pursuant to NRS 686B.070(1) is “Supplementary rate information,” which is defined in NRS 686B.020(4) as including any “rating rule” or “rule of underwriting relating to rates.” By definition, any underwriting rule or model used in underwriting that affects the premium that any insured would pay is a “rule of underwriting relating to rates.” Calling a model an underwriting model rather than a rating model does not affect the applicability of this requirement. The following are examples of underwriting rules and predictive models that must be filed with the Division and are subject to the Division’s prior-approval authority:

• Models and rules that determine placement of an insured within a tier where the tier placement isconsidered as a variable within the insurer’s rating plan. Tiering is considered to be rating bydefinition since tiering is merely an intermediate step between the underlying characteristics of therisk and the rating treatments ultimately assigned based on those characteristics.

Comment [HIG7]: Unless commercially deregulated…


Comments from Chubb

18


• Models and rules that determine placement of an insured within one of several affiliatedcompanies within a group, where each company would have a different rating plan and wouldpossibly charge different rates to otherwise identical risks. Because company placement directlydetermines the insured’s premium, such underwriting models are also necessarily considered to berating models.

• Models that compute any manner of score or index used as either a direct rating variable or adeterminant of eligibility or company placement, in whole or in part, if there is a possibility forsuch models to affect the premium that the insured is charged. Examples include, but are notlimited to, models based on credit information, geographical location, peril-specific riskestimation, or any demographic information. Any model that utilizes a mathematical algorithm tocalculate a score or index for eligibility purposes, and that is capable of being utilized for rating by any insurer, is also considered a rating model since the decision to reject a risk based on a score orindex is considered to be a more extreme variant of a decision to surcharge that risk based on thesame score or index. Rejecting any risks based on numerical indices would also affect thecomposition of the ultimately insured risks and thus would have an impact on the insurer’s lossexperience and actuarially indicated rates. Furthermore, the Division is concerned that rejectingrisks solely based on certain location-based indices would constitute a prohibited form of imposedunavailability of insurance in some areas of Nevada, and would thus be unfairly discriminatory.Use of particular location-based indices for the purposes of territorial rating may be approved ifappropriately filed and justified by relevant supporting data as determined in the course of theDivision’s review.

• Models that determine the extent to which an insurer relies on an actuarially indicated change to abase rate or relativity. These would include any “price optimization” models that an insurer mightuse to determine the extent to which a selected relativity moves toward the indicated relativity.Such models may not utilize any non-risk-based attributes such as price elasticity of demand orconsumer tendency to complain or shop for insurance. All risk-based attributes that such modelsuse must be fully disclosed to the Division along with the specific quantitative treatments of eachof those attributes.

The Division considers all of the aforementioned to fall under the purview of long-standing statutes and precedents. However, the proliferation of complex predictive models that some insurers have termed “underwriting models” has led to the necessity to reiterate such requirements. Nevada’s filing and prior-approval requirements continue to apply irrespective of the complexity of the algorithms utilized by insurers or the labels given to those algorithms.

Texas Regulation (Commissioner's Order No. 2691, dated August 7, 2013) The amendments to Rule VI add section M (Modeled Rating Factor), which provides an explanation of the MRF, describes its application, and lists the insurance carrier's requirements to use the MRF. The commissioner has deleted the word "factor" after "MRF" in Rule VI, new section M.2.c, as proposed, because it is duplicative.

The MRF is an optional factor that insurance carriers can file with TDI and apply when calculating workers' compensation premium. The MRF takes into consideration individual risk characteristics and loss experience of an insured. Insurers may use predictive modeling to determine the MRF. The term MRF can include tier rating and other similar terms.

Under the amendments to Rule III E, an insurer will apply its MRF to the policy in a multiplicative manner, and must not apply or use the MRF in a way that duplicates other rating factors, such as schedule and experience rating factors. Once determined, the MRF will apply during the entire policy period. Insurance carriers will be required to evaluate each policy's characteristics and experience at each renewal to determine the MRF for the renewal policy.

The amendments to Rule VI require insurance carriers to make a filing with TDI under Title 28, Texas Administrative Code, Chapter 5, Subchapter M (Filing Requirements) before using an MRF. The filing must include the MRFs; the characteristics, variables, or criteria used to determine the MRFs; actuarial support for the MRFs; and other supporting documentation.


Comments from Chubb

19


The commissioner has determined that the amendments to the manual are necessary for insurance carriers to use MRFs in calculating workers' compensation rates or premiums. The proposed filing requirement is necessary to promote transparency and accountability in the use of MRFs. Including an MRF in premium calculations allows an insurance carrier to tailor premiums more precisely to each insured by including an insured's specific risk characteristics and loss experience. With a more precise risk assessment, the insurance carrier can come closer to charging the appropriate premium for the risk each insured actually presents.

Part II: Unfair Trade Practice Considerations

The NAIC Unfair Trade Practices Act (#880) prohibits the following practices if committed flagrantly and in conscious disregard of the Act, or committed with such frequency to indicate a general business practice:

1. Making or permitting any unfair discrimination between individuals of the same class and of essentially the same hazardin the amount of premium, policy fees or rates charged for any accident or health insurance policy or in the benefitspayable thereunder, or in any of the terms or conditions of such policy, or in any other manner.

2. Making or permitting any unfair discrimination between individuals or risks of the same class and of essentially the same hazard by refusing to insure, refusing to renew, canceling or limiting the amount of insurance coverage on a property orcasualty risk solely because of the geographic location of the risk, unless such action is the result of the application ofsound underwriting and actuarial principles related to actual or reasonably anticipated loss experience.

3. Making or permitting any unfair discrimination between individuals or risks of the same class and of essentially the same hazards by refusing to insure, refusing to renew, canceling or limiting the amount of insurance coverage on theresidential property risk, or the personal property contained therein, solely because of the age of the residential property.

4. Refusing to insure, refusing to continue to insure, or limiting the amount of coverage available to an individual becauseof the sex, marital status, race, religion or national origin of the individual; however, nothing in this subsection shallprohibit an insurer from taking marital status into account for the purpose of defining persons eligible for dependentbenefits.

5. Terminating, or to modifying coverage or to refusing to issue or refusing to renew any property or casualty policy solelybecause the applicant or insured or any employee of either is mentally or physically impaired.

6. Refusing to insure solely because another insurer has refused to write a policy, or has cancelled or has refused to renewan existing policy in which that person was the named insured.

The model does not address the use of other data in the personal lines Property and Casualty insurance rating or claims settlements.

Part III: Consideration of Specific Variables

Policy discussions on whether a rating variable should be allowed generally touch upon three questions: Is the variable actuarially measurable and sufficiently related to actual or expected risk of loss? Does the variable have a correlation to a prohibited factor, such as race or income? Does the variable lead to unfair discrimination?

1. Credit Information: With the exception of California, Hawaii, Massachusetts, and Maryland, insurance companies areallowed to use credit-based insurance scores but are prohibited from basing rates solely on a credit score. Some statesalso prohibit the use of income, gender, ethnic group, religion, marital status, address, ZIP code, and nationality asfactors in a credit-based insurance score.

2. Domestic Abuse: The majority of states prohibit insurance companies from refusing to insure or limiting coveragebecause an individual is a victim of domestic violence. For claims settlements, some states may prohibit insurancecompanies from denying payment to an innocent coinsured who did not cooperate or contribute to a loss arising out of anintentional act of domestic violence where the perpetrator of the loss is criminally prosecuted for the act causing the loss.

3. Consumer Inquiries: While generally focused on homeowners insurance, the majority of states prohibit insurancecompanies from increasing premium rates, cancelling a policy, or refusing to issue or renewing a policy solely on thebasis of a policyholder inquiring about making a claim if the policyholder does not actually submit a claim.

4. Education and Occupation: While concerns have been expressed that education and occupation in rating have asignificant correlation to prohibited rating variables, few, if any states have prohibited the use of education andoccupation.


Comments from Chubb

20


Because of the increased complexity of rating models with sophisticated algorithms, it has become increasingly difficult to identify the impact of any specific rating variable. For example, an algorithm will apply varying relativity to various rating variables and may specify the order in which rating variables should be considered.

Part IV: Claim Settlements

The NAIC’s Unfair Claims Settlement Practices Act (#900) establishes a regulatory framework that requires insurer companies to adopt and implement reasonable standards for the prompt investigation and settlement of claims and prohibits insurance companies from compelling insureds or beneficiaries to institute suits to recover amounts due under its policies. There are no other NAIC models that address insurance companies’ use of consumer and non-insurance data in the settlement of property and casualty insurance claims.

Insurance regulators have generally permitted insurance companies to use data from third party vendors and corresponding algorithms for the settlement of claims as long as the use complies with the standards of fair claims settlements.

W:\National Meetings\2018\Spring\TF\Innovation\Big Data\Reg Framework.docx


Comments from Chubb

21

1

Tim Mullen

Director of Market Regulation

National Association of Insurance Commissioners

110 Walnut St., Suite 1500

Kansas City, MO 64106

Re: Comments on the Big Data Working Group's proposals

Dear Mr. Mullen:

As a licensed advisory organization, ISO provides advisory property casualty forms, prospective loss

cost data, and related products. ISO submits thousands of filings per year in all fifty states, D.C., Guam,

Puerto Rico, and the U.S. Virgin Islands. We appreciate the Big Data Working Group's ("Working

Group") continued collaboration with regulators, consumer representatives, and industry representatives.

We also thank you for the opportunity to comment on the State Regulatory Review of Complex Models,

the Discussion of Regulatory Framework and the Assessment of Regulatory Data Needs, as distributed

by the Working Group on December 19, 2017.

State Regulatory Review of Complex Models (Attachment A)

The initial focus of the proposal, as outlined therein, is on development of a mechanism for review of

models used in rate filings for Auto and Homeowners insurance. The Charge of the Working Group

reads as follows:

Propose a mechanism to provide resources and allow the states to share resources to facilitate their

ability to conduct technical analysis of, and data collection related to, the review of complex models

used by insurers for underwriting, rating and claims. Such a mechanism shall respect and in no way

limit the states' regulatory authority.

Although at a high level it might seem reasonable to create a shared resource for states to utilize in the

review of complex models, many questions and important details should be answered and discussed

before moving forward with the proposal.

Insurance Services Office, Inc.

545 Washington Boulevard

Jersey City, NJ 07310-1686

www.iso.com

Stephen C. Clarke, CPCU

Vice President

Government Relations

t 201.469.2656

f 201.748.1760

[email protected]

22

2

As a filer in fifty-four jurisdictions, the proposal continues to raise the following concerns:

• Not all states protect the confidential trade secret status of proprietary models. In these states,

filers have developed processes to allow regulators to achieve a comfort level with models

without exposing the models to public access. A one size fits all approach to model review

contemplated by the Working Group may not work in these situations.

• Industry has long had a problem with the potentially unreliable nature of confidential protection

in SERFF, especially in the many states that publish filings on the internet through SERFF Filing

Access (SFA). There have been instances where confidential information was inadvertently

released for public access. Additional confidentiality protections should be implemented.

• The proposal continues to contemplate potential use of NAIC staff to assist in review of complex

models. Currently state employees in many states are prohibited from disclosing confidential

information obtained in the course of their state employment. This prohibition typically extends

beyond the term of state-employment. However, NAIC staff are not state employees and thus

conceivably could be hired away from the NAIC to disclose trade secrets of proprietary models.

How will nondisclosure of confidential information amongst NAIC staff be handled both while

employed by the NAIC and thereafter?

• The Predictive Analytics Working Group (PAWG) will be charged with drafting potential changes

to the Product Filing Examiners Handbook to address best practices for review of predictive

analytics and models used by insurers to justify rates and to recommend filing requirements (e.g.,

information, data) for rate filings that are based on complex predictive models. What safeguards

can be used to ensure a particular actuary's state-based perspective doesn’t impact the review for

other states? Could the actuary be in a position to influence a country-wide checklist based on their

state's preferences? Further, could a uniform checklist potentially conflict with state rating laws?

We recommend that the standard setting process be conducted in an open and transparent manner

with input sought by all stakeholders,

• There is concern that creation of a technical review team, housed at the NAIC, could affect speed

to market in competitive rating states or result in unintended delegation of regulatory authority to

the NAIC.

• States approve rating plans and/or loss costs/rates that utilize the models. Where required, model

details are provided as supporting information. As written, this proposal seems to potentially add

an additional layer of regulatory approval of the model itself.

• Is the reference to "…models used by insurers to justify rates…" intended to cover widely

accepted CAT models, basic trend and loss development models, and/or third party models?

• What data is anticipated to be required under the PAWG checklist? Where third-party

proprietary data sets are used, will the model review team be authorized to execute non-

disclosure agreements directly with the provider?

• This proposal should be prospective and not retroactively touch models used in previously

approved filings.

23

3

Discussion of Regulatory Framework (Attachment B)

The Charge of the Working Group reads as follows:

Review current regulatory frameworks used to oversee insurers' use of consumer and non-insurance

data. If appropriate, recommend modifications to model laws and/or regulations regarding

marketing, rating, underwriting and claims, regulation of data vendors and brokers, regulatory

reporting requirements, and consumer disclosure requirements.

As laid out in the detailed document provided by the NAIC staff, regulators have many tools today to

oversee the use of large datasets in the insurance market, including:

• Rating laws which require that rates and rating factors not be excessive, inadequate or

unfairly discriminatory.

• Actuarial Standards of Practice which deal directly with the appropriateness of data and risk

characteristics.

• The rate filing process and the Market Conduct process, as well as consumer complaints,

which inform regulators of insurer practices and provide opportunities for regulatory

oversight.

• The Unfair Claims Practices Act and Unfair Trade Practices Act which regulate insurer's

claims handling, marketing, underwriting and advertising practices.

• The Fair Credit Reporting Act which requires that insurers provide transparency and

guarantee accuracy of certain data used in rating and underwriting.

• Gramm-Leach-Bliley privacy laws which regulate transparency and confidentiality of

policyholder's data.

A common thread in these tools is that they are, for the most part, adopted by regulation or enacted as

legislation. The NAIC has been a champion of the successful state based system of insurance regulation,

recognizing that it is the state legislatures making the laws and the state insurance departments enforcing

the laws that oversee the business of insurance. We are reminded of the extensive efforts undertaken

several years ago by the NAIC to encourage states to identify all regulatory requirements with no basis

in regulation or law, and attempt to repeal or discontinue such "desk drawer rules". We encourage a

measured, thoughtful and fully transparent approach to any consideration of further model law revision

or introduction. We also support continued efforts to encourage innovation while maintaining the

appropriate consumer protections which have for decades been the hallmarks of the state-based system

of insurance regulation.

Assessment of Regulatory Needs (Attachment C)

The Charge of the Working Group reads as follows:

Assess data needs and required tools for state insurance regulators to appropriately monitor the

marketplace and evaluate underwriting, rating, claims and marketing practices. This assessment shall

include gaining a better understanding of currently available data and tools, as well as

recommendations for additional data and tools, as appropriate. Based on this assessment, propose a

means to collect, house and analyze needed data.

As laid out in the detailed attachment provided by the NAIC staff, today regulators collect extensive

data from both individual companies and the market in general. Insurers and statistical agents expend

24

4

extensive resources to collect, quality check, edit, aggregate and deliver this data. ISO, for example, as a

licensed statistical agent annually collects and delivers large amounts of data to state regulators as well

as the NAIC. Further, in the review of rate filings, regulators have authority to request any supporting

information at the level of detail they deem necessary in order that regulators may properly review

filings.

It is our belief that regulators currently have the authority, and exercise that authority on a regular basis,

to collect or request sufficient data to oversee the insurance marketplace. We are concerned that the

current charge assumes otherwise and that the NAIC will pursue additional data collection that will

increase the regulatory burden on insurers, increasing costs that are ultimately borne by policyholders.

We would be happy to discuss our comments further should you have any questions. We look forward to

continuing to work with you on this critical issue.

Respectfully Submitted,

Stephen C. Clarke, CPCU

CC: Big Data Working Group Members

25

Deirdre K. Manna Vice President Political Engagement & Regulatory Affairs January 12, 2018 Tim Mullen Director of Market Regulation National Association of Insurance Commissioners 110 Walnut St., Suite 1500 Kansas City, MO 64106 Re: Comments of the Property Casualty Insurers Association of America Dear Mr. Mullen: PCI appreciates the opportunity to submit comments on the Proposal to Aid State Regulatory Review of Complex Models Used in Support of Personal Auto and Homeowner Insurance Rate Filings (“Attachment A”), the Discussion of Regulatory Framework (“Attachment B”) and the Assessment of Regulatory Data Needs (“Attachment C”), as distributed by the Big Data Working Group (the “Working Group”) via the NAIC website on December 19, 2017. Over the past few months, members of the Working Group shared their thoughts and concerns with us and, in turn, gave us an opportunity to share our thoughts and concerns as well. We have appreciated the opportunity to meet with as many regulators as we could during this time. The Working Group’s recent releases demonstrate a deliberate approach to identifying the scope of issues and potential solutions, and we appreciate these efforts. We provide our comments on each release. A. Proposal To Aid State Regulatory Review of Complex Models Used in Support of Personal Auto and Homeowner Insurance Rate Filings (the “Proposal”)(Attachment A) PCI first acknowledges that the current version of the Proposal appears to address many of our concerns about the original “Shared Resources for Complex Model Review”. We appreciate the efforts of the Working Group to continue refining the Proposal. PCI’s comments on the June 19, 2017 version of the Proposal remain on record to the extent the terms of the Proposal remain unchanged. Our primary goal since the original Proposal has always been to support state regulators while maintaining the integrity of the state-based regulatory system and encouraging innovation. With these objectives in mind, we present some additional comments on the recent Proposal that we hope will be fully explored in coming discussions.

(1) Best Practices for Review of Complex Rating Models PCI supports the concept of encouraging upfront dialogue between companies and regulators that might include checklists and guidance for insurance companies that they might follow in submitting rate filings. To be effective, this effort should enhance uniformity and speed to market, while also providing appropriate assistance to regulators and the industry as technological innovations bring advances to our industry. We respectfully ask the Working Group and the CASTF to conduct future work in developing checklists and guidance in an open and transparent manner, reflecting the experience of states that have been successful in using these tools for speed to market. If the expectation is that “best practices” developed during this process will become uniform standards for rate models and for regulatory review of models, then each state should adopt such standards by statute or rule. Standards should be based upon accepted actuarial principles. Standards that become regulatory requirements and could be the basis for disapproving a rate filing must be standards that exist in state law.

26

2

(2) State Regulatory Authority and Review Technology The Proposal indicates that a key principle for the Working Group’s analysis is that “State regulators will maintain their current rate regulatory authority.” PCI strongly agrees with this principle, and we note that each state’s rate regulatory authority is fundamental to a state’s ability to regulate its markets and the solvency of the insurance companies operating in its markets. One of the Proposal’s revised recommendations is for CASTF to work with NAIC technical staff to identify software, databases and technology to assist in the review of predictive models. While PCI generally supports regulators’ efforts to use technology to assist in performing their regulatory duties, that technology must not make core regulatory determinations. We would like to see the NAIC develop and expose for interested party input, both at the NAIC and in any state that proposes to use it, standards for regulators’ use of technology (“RegTech”), before any technology is embedded into rate reviews. Filers will need a clear understanding of how RegTech will align with statutory standards for rate reviews and companies’ due process rights. The use of RegTech should be consistent with the insurance laws and regulators’ authority, and must not have the effect of indirectly enlarging such authority or delegating authority to a third party. Because of the direct impact on insurers, we would ask that the process of identifying and selecting RegTech be open and transparent, conducted in accordance with state procurement laws, and subject to appropriate regulatory boundaries. PCI expects RegTech may eventually assist regulators with the rate review process, but no RegTech should supplant the decision-making role of state regulators or constrict or limit a regulator’s review of filings. The effects of RegTech should be directed and controlled in all ways by the governing state, not by third parties, in order to assure the integrity of our state-based system. We look forward to discussing these issues with the Working Group and others as the Proposal is further refined. (3) State Regulatory Authority and NAIC Review Team. The Proposal includes a recommendation that:

“NAIC management will be asked to conduct research into the appropriate skills and the potential number of resources required for the organization to address the needs of the NAIC Membership in conducting their reviews of models and make appropriate recommendations to the NAIC Executive Committee and Internal Administration (EX1) Subcommittee.”

PCI members have a number of concerns with this recommendation and would like to ensure that this review does not move the proposal back to its original form. Our first concern is that this recommendation is vague and seems to be premature. Over the past few months, PCI has visited with many regulators and encountered decidedly mixed messages. Regulators from some states said they do not need assistance because they have actuarial talent or use outside actuaries. A few said that they need assistance and cannot or will not use outside actuaries. We observed that most states have legal authority to recoup their expenses for skilled actuarial staff or services. A few said they do not have sufficient resources or the ability to pass on these expenses to industry. Many states indicated they handle complex filings through direct communications with the filer. Some states are seeking guidance regarding the information they should request and about how to best review large and complex filings. Other states commented on the need for greater uniformity. PCI has significant concerns about a single-source review mechanism staffed and operated by the NAIC. At best, this creates a very real potential for bottlenecking rate filings and hindering speed to market efforts. At worst, the mechanism could be an unlegislated delegation of state authority to the NAIC. Certainly, under recent PBR legislation, an NAIC team reviews life products. But the PBR process and the NAIC’s involvement are authorized by specific state legislation. Similar analogies exist with the financial statement filing process, producer data and licensing, as well as life-related product filings reviewed by the Interstate Insurance Rate and Product Commission. The key difference is that each of those regulatory initiatives required legislative authority. We respectfully suggest to the Working Group that other avenues in the revised proposal would be far more beneficial. If the mechanism is to be revisited at some point in the future, we urge the NAIC to first survey

27

3

states to determine the actual need for NAIC staff to conduct reviews of rating models or provide other assistance, and identify the states that need help. It would be appropriate to address these issues before discussing legal authority, staffing, staff qualifications or funding issues. Once the scope of the actual need is determined, then states and interested parties can discuss appropriate solutions and the legal boundaries of such options. PCI and its members strongly believe that there are other, less drastic options to help those states in need of assistance, and we are willing to work on developing solutions. To date, however, there has not been any demonstrable, widespread need or justification for NAIC employees to be directly involved in any part of rate reviews, a regulatory function. We believe it is premature to advance such a mechanism at this time. (4) Confidentiality. PCI agrees with the comments in the recent Proposal regarding Confidentiality and we would expand on these comments. Any discussions about a review mechanism or process to share information about rating models among states or others needs to include a discussion about confidentiality. Confidentiality is a critical issue for the industry. PCI members would appreciate understanding what information might be shared among states and with NAIC employees. The sharing of such highly sensitive, proprietary information and trade secrets has different potential consequences than the sharing of company information in a setting such as a MAWG or FAWG meeting. Given the significant differences among state laws governing personal lines, PCI encourages regulators to consider what information about a rating model filed in a particular state would be comparable and relevant if shared with other states. B. Regulatory Framework Analysis (Attachment B) (the “Framework Analysis”) PCI is a strong supporter of state regulation and believes the framework of our regulatory system contains well-established standards and processes that are every bit as appropriate and important today as when they were initially developed. The Framework Analysis identifies several important issues that warrant discussion. The Framework Analysis raises important issues, such as the extent of protections afforded by state credit scoring laws and consumer rights with respect to credit data. These debates are not new and have been largely resolved by state insurance regulators and legislatures. We request that the current discussions remain focused on new models and new data, rather than re-open old discussions. Other issues such as data ownership and regulation of data vendors are important issues and may be beyond the authority of insurance regulators to resolve. Conversely, key sources of consumer protection available under federal law, including the Fair Credit Reporting Act and the Gramm-Leach-Bliley Act, should be included in the Framework Analysis to provide a fulsome view of currently available consumer protections associated with the use of data in underwriting and rating. PCI requests the Working Group to revise Attachment B to include these key federal protections. The well-established, uniformly recognized standards for ratemaking – that rates be adequate, not excessive and not unfairly discriminatory (collectively, “rate standards”) – are as critical to a well-functioning market today as when the rate standards were first established in the United States. The rate standards help assure solvency and, thus, consumer protection. In turn, a solvent insurance industry contributes to economic stability for the country. In PCI’s view, the rate standards are bedrocks of state insurance regulation and should define future work on the Proposal and the Framework Analysis. PCI is not in favor of any action or analysis that would undermine, weaken, or discard such fundamental standards. Indeed, in their recent and effective response to the price optimization issue, regulators and the NAIC re-affirmed the fundamental importance of such rate standards. Rating variables, the information used to support rates and classifications, must be risk-based and predictive of loss. The insurance laws of most states expressly prohibit the use of certain, specific variables, and do not authorize regulators to broaden the scope of those statutory prohibitions. We are concerned that any focus on the degree of correlation between permissible and prohibited variables could suggest regulators have the ability to decide which variables may be used, despite the provisions of state law. It is PCI’s position that such important public policy decisions are to be made by state legislatures.

28

4

PCI is also concerned that the phrase “unfair discrimination” has been used in Working Group discussions without defining it. In the context of state insurance laws, “unfair discrimination” has a different meaning than the type of discrimination prohibited under the U.S. Constitution or federal anti-discrimination laws. It is important not to conflate the two. In insurance, “unfair discrimination” means treating similar risks differently. It does not mean that the use of certain data is necessarily unfair or improper. C. Assessment of Regulatory Data Needs (Attachment C) Attachment (C) provides a comprehensive review of data that is currently available to regulators via industry filings with the NAIC. As the memorandum notes, regulators have a tremendous amount of market and company data to use in regulating the industry generally and in regulating individual companies. The industry commits significant resources to gathering, compiling and delivering an extensive amount of data to regulators via the NAIC. If there are questions about the industry’s use of data in rating, that information should be available in rate filings. The comprehensive analysis of available data in Attachment (C) demonstrates that regulators have sufficient data, are well informed, and have the tools necessary to protect consumers. We anticipate, however, that some will argue regulators need more data to review rates and rating models. PCI disagrees. We understand that regulators may need additional information from a particular company about a particular filing, but we have not seen any evidence that broad categories of new data are needed to review rates and rating models. Regulators make determinations every day that rates are adequate, not excessive and not unfairly discriminatory, without the need for a data call. In conclusion, as a strong supporter of state-based regulation and the laws that have made it the most effective system in the world, PCI urges the Working Group to evaluate carefully the issues we raise. PCI commits to continuing to work with you to develop realistic, practical solutions, where needed, within the existing state-based regulatory framework. Thank you for the opportunity to comment on these important issues. Sincerely, Deirdre Manna

29

Memo To: Tim Mullen, NAIC Director of Market Regulation

From: John Humphries, Dave Heppen, and Debbie Rosenberg - Risk & Regulatory Consulting, LLC

Date: January 12, 2018

Subject: RRC Response to the NAIC Big Data (EX) Working Group regarding the Regulatory Framework – Proposed Structure and Issues for Discussion

Risk & Regulatory Consulting, LLC (“RRC” or “we”) supports the efforts of the Big Data (EX) Working Group (“Working Group”) to outline a process for determining whether changes to the current regulatory framework are needed to address the use of consumer and non-insurance data, and if so what form these changes should take. We have reviewed the discussion document and offer the following comments for your consideration. We would be glad to answer any questions, and we appreciate the opportunity to offer our comments.

We appreciate the efforts of the Working Group to thoughtfully consider the far reaching implications of new data sources that are available for policy rating. The data that is now available is far beyond that envisioned when current rating laws were drafted, and as time progresses there will be other available data that has not been considered to date. As such, innovative regulatory approaches are needed to address the impact of new sources of rating data on consumer insurance pricing. The work of this group will impact insurers, regulators and, most importantly, consumers, for decades to come. We strongly endorse the Working Group’s desire to, “balance consumer protection with industry innovation.” Both consumer protections and industry innovation are critically important in any regulatory approach that is developed. With this background we suggest:

• The Working Group should identify all variables currently being used in rating models to ensure that these variables are not only in compliance with current state law and statutory guidance but will also represent good public policy for the future. Special attention must be paid to rating variables that could serve as a proxy for variables already prohibited or not in keeping with desired public policy. For instance, some variables such as occupation could be used in a manner to serve as a proxy for race, creed, national origin, or the religion of the insured. The Working Group may need to consider not only proposed rating variables at face value but also require analysis to confirm there is not significant correlation to prohibited factors or outcomes. One potential approach is a periodic survey of the rating variables that are currently in use. Such a survey ideally would be as broad as possible. The survey could assist regulators in gaining an understanding of changes in commonly used rating variables as well as less frequently used variables. This in turn could suggest areas for further study by the Working Group.

30

RRC Response to the Big Data (EX) Working Group

January 12, 2018

Page 2 of 2

2

• In addition, as the ability to segment data continues to grow, there is risk that the resulting classes will be defined in narrow terms that could also inadvertently become proxies for prohibited rating characteristics. Understanding the changing characteristics of rating groups as segmentation refinement continues to develop will be an important challenge for regulators. Analysis of the variables being used, both now and in the future, could help regulators better understand this potential issue.

• Similar to Nevada Regulatory Activity Bulletin 17-001 included in Attachment B, statutory guidance should include language that broadly defines underwriting rules and rating models based upon the ultimate effect on premium paid by an insured. Tier rating, company placement within a group of affiliates, or other similar rating structures should not be allowed to circumvent rating prohibitions or guidance.

• Provisions to ensure the accuracy of input data for rating models will be critically important, and should be considered as part of the regulatory process. The risk of inaccurate internal policy data is already mitigated by a series of internal controls and processes that are reviewed by both internal/external audit and confirmed during periodic statutory examinations. As new rating variables from new data sources including external vendors become key factors in determining the premium charged to consumers, then the accuracy of this data should also be protected by appropriate controls and readily available for review during audits and statutory examinations.

With these suggestions in mind, we strongly support regulatory facilitation of improvements to rating methodologies. The regulatory framework should encourage research into continued development of modeling techniques and rating variables that allow for:

• Improvements in the accuracy of pricing • Reductions in subsidizations of individual rating groups • Encouragement of safer behavior through the use of improved ability to reflect such behaviors in pricing

The Working Group has succeeded to date in providing a forum for input from regulatory, consumer, and industry perspectives in furthering these goals. We encourage continuation down this path.

31