Similar risks should be priced similarly, and relevant distinctions based on actuarially relevant factors should be captured in the rate. Predictive analytics in insurance will allow ever more precise design of insurance products, selection of insureds and product targeting, and price for the risk exposure. However, the proliferation of “big data” that enables insurers to refine pricing, underwriting, and marketing, may involve significant regulatory and challenges for insurers, particularly against the long-standing prohibition on unfair discrimination. Several state regulators have already stopped “price optimization” for personal lines, and the NAIC is looking at the issue. Surplus lines insurers, as largely unregulated entities with full freedom of rate and form, are in an excellent position to take full advantage of the growing opportunities associated with data and analytics, although even surplus lines insurers may run afoul of unfair price discrimination prohibitions in insurance.
Predictive Analytics: Turning Data into Predictions
Predictive analytics (also called predictive modeling) allows businesses “to gain insights, foresight, and inferences from the treasure chest of raw transactional data” 1 acquired by that company’s transactions with its customers and from various data brokers and aggregators who compiled bits of information on people, such as voting registration lists, zip codes, social network websites, internet habits and views as tracked by cookies (called “data exhaust”), geolocation on smart phones, hobbies, restaurant reservations, loyalty card programs, even e-mail address domains and types of computer.
For example, a company called hiQ Labs can predict which employees might defect (“flight risk”2) and if the employer wants to retain that employee to offer a tailored solution that will keep the employee’s loyalty.3 Travel website Orbitz found Mac users spend more money on hotels than PC users, so Orbitz shows them higher-priced hotels.4 Retailer Target used predictive analytics to determine which young female customers were likely pregnant, even before the women knew, to then offer them coupons and hook them into Target for their maternity and baby clothing and care needs.5
Maisel and Cotkins explain that, “the power of analytics is to turn huge volumes of data into a much smaller amount of information and insight.” In contrast, business intelligence mainly summarizes historical data, typically in table reports and graphs.
Insurers and Predictive Analytics
Predictive analytics, of course, are nothing new in insurance. Credit scores are but one example of the opportunities and challenges associated with predictive analytics. Credit reporting agencies compile data from banks and creditors of the consumer, and generate credit scores that are used in insurance to predict likelihood of claims, not premium-paying ability. (Scores are also used for employment, consumer finance, and rental housing.) Scores are based on the proprietary algorithm invented by Fair Isaac & Co in 1950,6 (thus the name FICO scores), revised many times over the years as data feeds improved and experience was gained. “A positive credit record may also indicate a high level of responsibility toward possessions or property. Adverse consumer credit history may lead to a lack of property or vehicle maintenance, a propensity to file fictitious claims, and a motivation for the inflation of legitimate claims,” according to an article in CPCU Journal by Lamont Boyd and Dale Halon, both with Fair Isaac.7
The Federal Trade Commission found the use of credit scoring turned out to be highly valuable in insurance pricing, but it also came at some social cost.8 Insurers’ use of credit scores (more specifically insurance scores, which are a bit different9) can have adverse impacts on rates, and on minorities. The U.S. Supreme Court ruled in 2007 that when an insurer charges a higher rate (above some “neutral rate”) to a consumer when the insurer uses a credit score than if it did not use the score, that constitutes adverse action under the Fair Credit Reporting Act.10 The National Conference of Insurance Legislators has a Model Act Regarding Use of Credit Information in Personal Insurance, adopted in most states, that prohibits the use of an insurance (credit) score calculated using income, gender, address or zip code, ethnicity, religion, marital status or nationality as a factor.11 These categories are listed in many state statutes as factors that constitute unfair discrimination. Controversies remain as consumer advocates continue to push prohibitions on the use of insurance scores, and the Federal Housing Administration has issued new regulations.
Beyond Credit Scores
Credit scores are the tip of the iceberg of data analysis, and serve as an example of the challenges insurers will face in using analytics.
State laws typically require insurance rates to be “adequate, not excessive, and not unfairly discriminatory.” Adequate rates are thought to be important to protect against insurers’ destructive competition that could lead to insolvencies. “Not excessive” rate regulation is to protect consumers.12 Fair discrimination in insurance means persons with similar exposures are treated similarly, so that high risk exposures are priced at a higher rate than low risk exposures. The goal is to prevent differences in rates that do not reflect differences in risks underwritten,13 and to avoid some cross-subsidization. Unfair discrimination has been the biggest concern of late, mostly stemming from insurance credit scores. For all insurers, unfair discrimination is prohibited, whether by insurance states and regulations, or by civil rights statutes.14 Thus the Fair Housing Act applies to insurers issuing homeowners insurance.15
Unfair discrimination is where predictive analytics gets tricky, resulting in possible “digital redlining.” According to a White House report, “…big data could enable new forms of discrimination and predatory practices.”16 Price precision or digital redlining can go different ways. One is to rate an insured higher because of new factors that show higher risk of loss. This seems legitimate until an insurer crosses the unmarked border to unfair discrimination that disproportionately and inadvertently impact protected classes. Disparate impact on protected classes of insureds, resulting from otherwise legitimate factors, will be a bar to the use of such data and analysis. A second way analytics can affect rate is to charge a higher premium because some consumers might well tolerate it, even if there is no difference in risk, called “price optimization,” (like Orbitz offering higher hotel rates to Mac users). California, Ohio and Maryland have banned price optimization as unfair discrimination in insurance pricing, and the NAIC is looking at this.17 Unfair business practice statutes can also bar such pricing.
Predictive analytics might allow the risk to be more precisely marked, allow actuaries to tighten the bands of homogeneity for a class, and test whether underwriters’ intuitions are accurate. At the far end, underwriters might be able to provide a unique premium for the individual risk – if the regulators allow any of this. The do-not-cross line between fair discrimination and unfair discrimination due to policy concerns is going to shift, especially when the risk price leads to policy concerns.
For standard lines insurers, rating consumers is subject to state rate and form filing laws, which restrict what factors can be used. Even if regulators bar the use of data, it may be hard to escape the impact as data becomes ever easier to access and analyze. If the standard market faces a regulatory handicap in using information, it follows that more and more of this business will migrate to the unregulated market where surplus lines insurers can use information that the standard underwriters can’t. This is an opportunity for surplus lines insurers, free from rate regulation and which operate more freely than standard carriers.
Yet surplus lines insurers too will run into federal laws that prohibit unfair price discrimination. As a federal court said in one of the early cases on insurance pricing: “Risk discrimination is not race discrimination. Yet efforts to differentiate more fully among risks may produce classifications that could be generated by discrimination.”18 Surplus lines insurers might even find themselves under attack because they lack the protection of state regulator having approved rates in advance.
Before the regulators mark the course, the underwriter even in surplus lines markets will need to exercise judgment rather than be merely a data jockey on a runaway horse of analytics.
- Lawrence Maisel and Gary Cokins, Gary, Predictive Business Analytics: Forward Looking Capabilities to Improve Business Performance, (John Wiley & Sons, 2013), p. 4. ↩
- Eric Siegal, Predictive Analytics (John Wiley & Sons, 2013), pp. 45-51. ↩
- “Meet the Market Shapers,” The Economist, (Jan. 10, 2015), p. 67. ↩
- Dana Mattioli, “On Orbitz, Mac Users Steered to Pricier Hotels,” Wall Street Journal (Aug. 23, 2012). Adrian Kingsley-Hughes, “Mac Users Have Money to Spare, Says Orbitz,” Forbes (June 26, 2012) ↩
- Charles Duhigg, “Psst, You in Aisle 5,” New York Times (Feb. 16, 2012), p. MM30. ↩
- Garry Boulard, “The Roar Over Credit Scores.” State Legislatures, vol. 30, no. 9: 19-21 (October 2004). ↩
- Dale M. Halon and Lamont D. Boyd, “The Role of Credit History Scoring in Peron Personal Lines Insurance Underwriting,” CPCU Journal, vol. 49 (1), p. 40 (Spring 1996). A more detailed and nuanced account of why insurance scoring is reliable is in Patrick L. Brockett and Linda L. Golden, Biological and Psychobehavioral Correlates of Credit Scores and Automobile Insurance Losses: Toward an Explication of Why Credit Scoring Works,” Journal of Risk and Insurance vol. 74(1), p. 23 (2007). ↩
- “Credit-Based Insurance Scores: Impacts on Consumers of Automobile Insurance: A Report to Congress By the Federal Trade Commission,” Federal Trade Commission (July 2007). ↩
- Diana Lee, et al., “Give Us Some Credit: The Use of Credit Information in Insurance Underwriting and Rating,” Risk Management and Insurance Review, vol. 8 (1): 31 (2005). ↩
- Safeco Ins. Co. of Am. v. Burr, 551 U.S. 47, 49 (2007). ↩
- Model Act Regarding Use of Credit Information in Personal Insurance, National Conference of Insurance Legislators. And see http://www.insurancecompliancecorner.com/credit-scoring-use-extraordinary-life-circumstances-exception-creates-added-consumer-protection/; http://www.namic.org/issues/InsuranceScoring.asp. ↩
- Scott E. Harrington, “Insurance Rate Regulation in the 20th Century,” Journal of Insurance Regulation, vol. 19 (2), p. 204 (2000). ↩
- Angelo Borselli, “Insurance Rates Regulation in Comparison with Open Competition, Connecticut Insurance Law Journal, vol. 18, p. 109, 130. ↩
- Dehoyos v. Allstate Corp, 345 F.3d 290 (5th Cir (Tex.) 2003). ↩
- Ojo v. Farmers Group, Inc., 565 F.3d 1175 (9th Cir. (Cal.) 2009). ↩
- “Big Data: Seizing Opportunities and Preserving Values,” Executive Office of the President (May 2014, p. 53.) ↩
- “Notice Regarding Unfair Discrimination in Rating: Price Optimization,” California Department of Insurance (Feb. 18, 2015). Bulletin 14-23, Maryland Insurance Administration (Oct. 31, 2014). Bulletin 2015-01, Ohio Department of Insurance (Jan. 29, 2015). ↩
- N.A.A.C.P. v. Am. Family Mut. Ins. Co., 978 F.2d 287, 290 (7th Cir. 1992). ↩