December 2016

What is Unfair Discrimination in General Insurance Rating? A Regulator’s Perspective

The following article is an opinion piece by the author, Michael McKenney. It represents his own personal opinion and does not necessarily represent the views of the SOA, his employer (the Pennsylvania Insurance Department), any other state insurance regulator, the NAIC or the NAIC Casualty Actuarial and Statistical (C) Task Force.

By Michael McKenney

Practically everyone who works in the insurance industry can tell you that insurance rates are not permitted to be excessive, inadequate or unfairly discriminatory. But what makes a rate excessive, inadequate or unfairly discriminatory under state law?

Remarkably, even though each of Pennsylvania’s property and casualty rate regulatory acts and its Unfair Insurance Practices Act include these restrictions, there are only three instances in which they are defined.

With respect to excessive rates, Section 704(a)(2) of Pennsylvania’s Workers’ Compensation Act (77 P.S. § 1035.4(a)(2)) states:

“A rate may not be held to be excessive unless it is likely to produce a long-run profit that is unreasonably high in relation to the risk undertaken and the services to be rendered.”

Regarding inadequate rates, Section 704(a)(3) of Pennsylvania’s Workers’ Compensation Act (77 P.S. § 1035.4(a)(3)) states:

“A rate may not be held to be inadequate unless:

  1. it is unreasonably low for the insurance provided and continued use of it would endanger solvency of the insurer; or
  2. the rate is unreasonably low for the insurance provided and the use of the rate by the insurer has had or, if continued, will have the effect of destroying competition or of creating monopoly.”

Finally, Section 3(d) of Pennsylvania’s Casualty and Surety Rate Regulatory Act (40 P.S. § 1183(d)) notes the following with respect to unfairly discriminatory rates:

“No rate shall be held to be unfairly discriminatory unless, allowing for practical limitations, it clearly fails to reflect with reasonable accuracy the differences in expected losses and expenses. A rate is not unfairly discriminatory because different premiums result for policyholders with like-loss exposures but different expense factors, so long as the rate reflects the differences with reasonable accuracy. A rate is not unfairly discriminatory if it is averaged broadly among persons insured under a group, franchise or blanket policy.”

Rarely does the Pennsylvania Insurance Department run into issues with insurance companies filing property and casualty insurance rates that can be determined (within the context of a rate filing) as threatening an insurer’s financial solvency or destroying competition. On occasion, we may take issue with a filing that proposes rates we believe to be excessive, but the more common issue regards the filing of rates that we believe to be unfairly discriminatory.

When class plan changes are filed, the regulator will often be provided an exhibit displaying the current, indicated and proposed relativities. We have engaged in conversations with insurance companies and other regulators who believe that practically any selection between (and inclusive of) the current and indicated relativities is acceptable. Some believe this holds true even when different classes of risks receive different treatment so long as the selections stay within the range created by the current and indicated relativities. But is any selection (between the current and indicated relativities) not excessive, inadequate or unfairly discriminatory under Pennsylvania law?

Consider a simplistic example in which an insurer has three groups of policyholders: Average, Above Average and Below Average. In its current rating plan, average risks pay twice the rate of above-average risks and below-average risks pay twice more. Assume each group indicates the need for a rate increase of 50 percent but the insurer wishes to temper this large rate increase for its average and above average policyholders:

  Current Indicated Proposed
Above Average .50 .75 .60
Average 1.00 1.50 1.20
Below Average 2.00 3.00 3.00


If the indications are correct, none of the proposed rates are likely to produce long-run profits that are unreasonably high.

In today’s multi-state, multi-line insurance environment in which many insurers write business and the frequency with which insurers file to make rate changes, it is unlikely that the tempered rates for average and above-average risks will threaten the solvency of the insurer. Additionally, given the competitive markets in which property and casualty insurers operate, it is doubtful that the tempered rates will destroy competition.

What about unfair discrimination? After allowing for practical limitations, do the differences in the proposed rates reflect differences in expected losses and expenses with reasonable accuracy?

Both the current and the indicated rates illustrate that below-average risks should pay four times the rate of above-average risks and two times the rate of average risks. Is it therefore acceptable for below-average risks to pay five times the rate of above-average risks and two-and-a-half times the rate of average risks?

Some arguments in favor of the validity of proposed rates similar to the example provided cite laws such as Section 4(a) of Pennsylvania’s Casualty and Surety Rate Regulatory Act (40 P.S. § 1184(a)) which permit insurers to support their filings by:

“(1) the experience or judgment of the insurer or rating organization making the filing, (2) the experience of other insurers or rating organizations, or (3) any other factors which the insurer or rating organization deems relevant.”

But in permitting insurance rates to be based on “judgment” and “any other factors which the insurer or rating organization deems relevant,” does this mean that they need not be cost-based? Or are these other considerations constricted to aiding the insurer in predicting the appropriate cost-based estimates?

In Pennsylvania, we agree that insurers may use many different types of considerations to support their rates and these considerations need not (and in some cases even should not) be based on the insurer’s own experience. However, no matter what those considerations entail, differences in rates must reflect differences in expected losses and expenses with reasonable accuracy. If they do not, the rates are unfairly discriminatory.

In the above example, if there are additional analyses and/or considerations that adequately support a reasonable conclusion that the expected losses and expenses for below-average risks will be five times that of above-average risks and two-and-a-half times that of average risks, we will likely approve the filing.

As one example, if the insurer can provide a competitive analysis that shows taking the full rate indication for below-average risks will place them in line with their competition but doing so for average and above-average risks will place them far above their competition, we may accept the proposal. In this instance, the market as a whole may provide a more credible estimate of expected losses and expenses.

But if the insurer lacks information to suggest expected losses and expenses can reasonably be expected to differ to the same degree of magnitude as that which is proposed, we will likely not approve the filing.

The above was a simple example provided for illustrative purposes. In practice, today’s insurance world of “Big Data” and “Generalized Linear Models” has led to incredibly complex and segmented classification plans. Territories that were once defined at the county level (with exceptions for urban areas) are now increasingly set at the zip code level (sometimes nine-digit zip codes) or even by census block where the 100 block of North Market Street has a different territory relativity than the 200 block of North Market Street. The relativities are based not on the actual loss experience of the census block (which would lack any credibility) but instead on characteristics of the census block that statisticians can prove to be correlated with loss.

In response to the use of Big Data by insurers, the NAIC’s Market Regulation and Consumer Affairs (D) Committee has created the Big Data (D) Working Group whose 2016 charge is:

“Explore insurers’ use of big data for claims, marketing, underwriting and pricing. Explore potential opportunities for regulatory use of big data to improve efficiency and effectiveness of market regulation. If appropriate, make recommendations no later than the 2016 Fall National Meeting for 2017 charges for the Committee to address any recommendations identified by the 2016 exploration.”

With respect to the use of Big Data in pricing, the complex multivariate computer models predicting the indicated relativities underlying today’s exceptionally segmented class plans are often performed on an iterative basis and include the use of judgment throughout. Additionally, the further judgmental aspects of selecting rate relativities based on the indications are now even being modeled, a practice sometimes referred to as “price optimization.”

The NAIC’s Casualty Actuarial and Statistical (C) Task Force began drafting a white paper on the topic of price optimization after the issue was referred to it by the Auto Insurance (C/D) Study Group on Nov. 11, 2014. The white paper was adopted by the NAIC’s Property and Casualty Insurance (C) Committee on Nov. 21, 2015 and by the Executive Committee on April 6, 2016. In paragraph 1 of the white paper, the Task Force states that it “provides background research on price optimization, identifies potential benefits and drawbacks to the use of price optimization, and presents options for state regulatory responses regarding the use of price optimization in ratemaking.”

Paragraph 6 of the NAIC’s Price Optimization White Paper describes:

“In recent years, through a process or technique referred to by many as ‘price optimization,’ insurers have started using big data (data mining of insurance and non-insurance databases of personal consumer information where permitted by law), advanced statistical modeling or both to select prices that differ from indicated rates at a very detailed or granular level. Formalized and mechanized adjustments can be made to indicated rates for many risk classifications and, ultimately, perhaps even for individual insureds.”

Paragraph 9 of the NAIC’s Price Optimization White Paper further notes:

“Regulators accept some deviations from indicated rates and rating factors. However, they are concerned that the use of sophisticated methods of price optimization could deviate from traditional ratemaking, extending beyond acceptable levels of adjustment to cost-based rates and resulting in prices that vary unfairly by policyholder. Regulators in each state determine the acceptable level of adjustment allowable based on state law and regulatory judgment.”

As of the date that this opinion piece was written, approximately 20 states (including Pennsylvania) have provided official notice that the use of price optimization techniques resulting in unfairly discriminatory rates will not be tolerated. Most of these notices relate price optimization to the use of computer models to set insurance rates based in some way on how much a consumer or group of consumers may be willing to pay before shopping around. Many were based, at least in part, on a draft Bulletin included as Appendix B of the NAIC’s Price Optimization White Paper.

One of the more controversial recommendations in the NAIC’s Price Optimization White Paper is found in paragraph 48 which discusses unfairly discriminatory insurance rating practices that “adjust the current or actuarially indicated rates or premiums, whether included or not included in the insurer’s rating plan.” In providing examples of what may constitute an unfairly discriminatory practice in this regard, the paper includes:

  1. "Price elasticity of demand.
  2. Propensity to shop for insurance.
  3. Retention adjustment at an individual level.
  4. A policyholder’s propensity to ask questions or file complaints.”

These same four practices are noted in the white paper’s draft Bulletin.

Comments received by the NAIC’s Executive Committee in advance of the paper’s adoption at the 2016 NAIC spring national meeting recommended “price elasticity of demand” and “propensity to shop for insurance” be considered examples of potential unfair discrimination only when considered at the individual or granular level, but the paper was adopted without the recommended changes.

The days of reviewing a class plan variable’s loss experience on a univariate basis are long gone.  Insurers are rating at levels of segmentation that few could have imagined years ago. 

As noted in paragraph 49 of the NAIC’s Price Optimization White Paper:

“The use of sophisticated data analysis to develop finely tuned methodologies with a multiplicity of possible rating cells is not, in and of itself, a violation of rating laws as long as the rating classes and rating factors are cost-based.”

But with class plan indications derived by iterative processes using complex computer models that include judgment throughout, how does the regulator ensure that the judgmental aspects of the indication are unbiased and related to expected losses and expenses? And when the further judgmental aspects of selecting rate relativities based on these indications are also being modeled, how can the regulator keep pace?

The complexities underlying the manner with which property and casualty insurance rates are being developed and the extreme segmentation with which class plans are being administered have become significant challenges for today’s regulators. The NAIC’s Price Optimization White Paper, the bulletins that many states have issued against price optimization practices and the NAIC’s Big Data (D) Working Group are examples of recent regulatory responses to these challenges. But at the end of the day, the same standards that have applied to insurance rates for many decades continue to apply today and they remain applicable regardless of whether there are just three classes of risk (e.g., average, above average and below average) or many thousands (e.g., census blocks). Rates shall not be excessive, inadequate or unfairly discriminatory.

Resources

  1. For information on the NAIC’s Big Data (D) Working Group, including its 2016 charge, see NAIC Market Regulation and Consumer Affairs (D) Committee, Big Data (D) Working Group homepage , (last visited May 18, 2016).
  2. To access the Price Optimization White Paper, see NAIC Casualty Actuarial and Statistical (C) Task Force, “Price Optimization White Paper” (Nov. 19, 2015) available here.
  3. The NAIC’s Casualty Actuarial and Statistical (C) Task Force’s website lists various states’ bulletins on price optimization at NAIC Casualty Actuarial and Statistical (C) Task Force homepage, (last visited May 18, 2016) (found under Price Optimization Bulletins/News Releases heading).
  4. Comments on the NAIC’s Price Optimization White Paper that were received by the NAIC’s Executive Committee in advance of the paper’s adoption at the 2016 NAIC spring national meeting can be found online within the materials for the NAIC’s April 6, 2016 Executive (EX) Committee and Plenary Meeting. See NAIC, “Report of the Executive Committee” (Apr. 6, 2016) available here.
  5. Further information on the background on why the NAIC’s Casualty Actuarial and Statistical (C) Task Force began drafting the Price Optimization White Paper can be found on the NAIC’s website. See Center for Insurance Policy and Research, “Price Optimization” (Jan. 6, 2016) available at here.

Michael McKenney is the actuarial supervisor for the Pennsylvania Insurance Department, Property & Casualty Bureau. He is also currently the chair for the NAIC Casualty Actuarial and Statistical (C) Task Force.