Some States Jumping the Gun on Predictive Modeling Review
The latest state to up the ante is Washington, where Department of Insurance inspectors reportedly have begun to respond to filed rating plans that include credit-based tiers with requests that insurers provide extensive details on the models they used, including all formulas and rating factors. Responding to such inquiries could prove difficult to impossible, as the “special sauce” each third-party modeling firm uses historically has been held to be protected trade secrets.
This shift comes not as a result of new legislation or any formal rulemaking, but simply as an ad hoc interpretation of Washington State’s 17-year-old law on multivariate analysis in insurance filings. In essence, the department is applying “desk drawer rules,” in which regulators’ interpretation of ambiguities in the statutory code or inconsistent application of legal provisions create a lack of clarity. In the blink of an eye, the norms that for more than two decades have allowed insurers to file rating tiers that use credit-based insurance scores have evaporated.
And Washington isn’t the only state where insurers have been reporting changes in how rate filings are being processed. The Maine Bureau of Insurance, headed by NAIC President Eric Cioppa, likewise has begun demanding previously undisclosed details about predictive models that make use of credit-based insurance scores.
Given the role the Maine bureau’s staff played in helping the NAIC’s Casualty Actuarial and Statistical Task Force to develop a draft white paper—first exposed during the fall 2018 national meeting—on regulatory best practices for predictive models, it’s hard to avoid the conclusion that some states are jumping the gun to put the task force’s draft recommendations into practice. Moreover, as regulators prepare for a protracted debate over emerging “big data” models, it’s clear that longstanding industry practices with respect to consumer credit data are being sucked into the fray.
Among the most pressing issues at this stage is that regulatory requests for supporting materials that insurers cannot provide could begin to disrupt the underwriting and ratemaking process. The task force’s white paper—whose latest version was unveiled Oct. 15, with comments due Nov. 22—nods toward the importance of confidentiality. However, it also notes that “insurers should be aware that a rate filing might become part of the public record.” That is going to be cold comfort for third-party vendors who regard their formulas and processes as protected intellectual property.
As I’ve written in this space before, big data is almost certainly going to be a fraught subject for both industry and regulators in the coming years. We are just beginning to sketch the contours of the sort of regulatory review process it will require. Getting from here to there will be an arduous journey.
But that’s also why it is important that we not get ahead of ourselves or prematurely throw away what’s already proven to work well. Multiple reports confirm the use of credit-based insurance scores is not discriminatory. Moreover, the evolution of actuarially credible variables tied to credit information has been the single most important factor over the past 30 years in allowing insurers to craft rates for virtually every auto insurance policyholder. Where nearly half of all consumers in some states once were shunted into expensive high-risk residual markets, there are only four states today—Maryland, Massachusetts, Rhode Island and North Carolina—where residual markets account for even 1 percent of auto insurance policies.
The most important job of insurance regulators is to apply the rules consistently and transparently in a way that fosters competitive markets among solvent insurers. By and large, that’s what we have today. It is only with extreme caution that we should move to disturb that potentially fragile equilibrium.