American Academy of Actuaries: Comment Letter to DISB on their Review of Automobile Insurance for Unintentional Bias

WASHINGTON, Aug. 24 (TNStalk) -- The American Academy of Actuaries issued the following letter:

* * *

To: The Honorable Karima Woods, Commissioner, District of Columbia Department of Insurance, Securities & Banking, 1050 First Street NE, #801 Washington, DC 20002

Dear Commissioner Woods:

On behalf of the American Academy of Actuaries/1 Casualty Practice Council and Property/Casualty Racial Equity Task Force, I am pleased to provide these comments in response to the District of Columbia Department of Securities and Banking (DISB) Request for Comment dated July 22, 2022.

These comments are provided in the context of ongoing interest in the DISB efforts to study unintentional bias in automobile insurance. They are based upon original testimony previously provided on this topic, which includes the oral testimony provided on June 29 and written testimony provided on July 8 ("Academy July 8 Written Testimony").

These comments are provided in the order of the questions that the DISB included in its recent letter. To provide further assistance, a list of comments have been included which is labeled as "Further Technical Considerations" at the end of this response.

Question 1 Response:

Proposed potential outcomes the review will investigate.

As noted, the table included in this section is derived in part from the Academy's previous submissions. Specifically, the column labeled "Outcome Data" is similar to a list of potential "Measures" identified in our prior testimony. We agree that the proposed list of outcomes would be beneficial to analyze, and refers the DISB to the comments in the Academy July 8 Written Testimony.

Further, consideration of including expense and loss costs as an outcome metric in order to determine whether the costs differ across the reviewed groups is encouraged. Higher costs for a particular group may lead to higher rates for that group when standard ratemaking techniques are applied. In the case of predictive models, the higher rates may be statistically indicated through the use of risk characteristics that are correlated with the reviewed group, even if those variables appear neutral. Therefore, higher costs for protected classes could indicate potential vulnerability to unintentional bias through the use of risk characteristics correlated with the reviewed groups.

Question 2 Response:

Proposed underwriting and rating factors not considered reflective of unintentional bias.

As noted in the Academy July 8 Written Testimony, there is agreement with the DISB that risk classification is a necessary part of the insurance underwriting and rating functions. Further, there is agreement with the DISB approach to consider certain characteristics when reviewing differences in the outcomes listed in Question 1. This type of approach is consistent with a review of disproportionate impact, as discussed in the prior testimony.

The identified list (age, loss history, and driving record) is not exhaustive of all rating criteria used and approved within the District of Columbia. While we recognize that the DISB will need to balance efforts to control for underwriting and rating variables appropriately with the practical limitations of collecting rating factor information, the DISB is encouraged to consider inclusion of certain attributes known to have a material impact on expected losses. For example, it is important to consider policy coverage characteristics (e.g., coverages selected, coverage limits, and deductibles) in addition to the three elements identified by the DISB.

Other potential characteristics to consider include:

* Annual mileage driven, as well as other criteria from data collected via telematic processes,

* Vehicle characteristics (e.g., model year and symbol), and

* Driver characteristics (e.g., years of driving experience).

Regarding the assumption that driving record is not reflective of unintentional bias, the DISB is encouraged to consider the extent to which there are differences in policing practices between reviewed groups.

Question 3 Response:

Consideration of potential bias in factors that are both a protected class and comply with Actuarial Standard of Practice No. 12.

The DISB has determined that certain characteristics such as gender and marital status "typically comply" with ASOP No. 12. We do not have a position on whether particular risk characteristics would comply with ASOP No. 12. ASOP No. 12 provides guidance to the actuary in selecting risk characteristics, establishing risk classes, and testing the risk classification system. Whether a particular characteristic complies with ASOP No. 12 will depend on whether the actuary has followed the guidance in ASOP No. 12 in designing a risk classification system for the task at hand (e.g., a specific rate filing). However, actuaries are always bound to follow applicable laws and regulations, and if the DISB determined that certain characteristics must be used in a risk classification system, the actuary would be required to include them.

While we do not offer comment regarding whether gender or marital status should be included in category 2 or 4, the DISB is encouraged to consider collecting this data for use in its analysis.

Question 4: Response:

Proposed criteria to be evaluated for bias.

A clear understanding of the limitations of the inference approach to identifying race and ethnicity is recommended. For example, a Rand Corporation study states that: "Concordance between self-reported race/ethnicity and BISG [Bayesian Indirect Surname Geocoding] estimates are typically 90-96% accurate for the four largest racial/ethnic groups (Asian/Pacific Islander, Black, Hispanic and White)," but are "less accurate and not recommended for American Indians/Alaskan Natives and Multi-racial persons, due to small numbers." Further analysis and information on the collection of protected class data is provided in the Academy issue brief: Sourcing Protected Class Information in P&C Insurance. Consideration should also be given to the treatment of multi-racial or multi-ethnic individuals and households.

Further, the DISB is encouraged to consider the potential effect of individuals belonging to more than one protected class.

Further Technical Considerations:

As the DISB moves forward with the proposed study of unintended bias in auto insurance, numerous details and technical considerations will emerge. No doubt the DISB has identified a number of these, but we offer the following list of considerations in hope they will prove helpful:

* Data--Clearly, the data specifications for this study will be critical. At the outset, it will be dependent on what data the industry can provide and the timeframe under which the industry can provide it. The unique data collection and rating algorithms utilized by individual companies should be considered. The Academy is interested in staying involved and being helpful in the process of defining the data specifications.

* Time Frame--We recommend using data that was derived from policies effective before, during, and after the height of the COVID-19 pandemic.

* Coverages and Vehicle Considerations--Clear identification of policy-level data and vehicle-level data will be necessary to identify cost differences and make valid comparisons across insured customers.

The proposed approach outlined by the DISB could be useful in identifying whether unintentional bias in private passenger automobile insurance exists. We look forward to discussing the potential ways to address any bias identified as a result of this analysis.

Lauren Cavanaugh, Vice President, Casualty Practice Council, Chairperson, P/C Racial Equity Task Force