The National Association of Insurance Commissioners’ Casualty Actuarial and Statistical (C) Task Force’s (CASTF) draft Regulatory Review of Predictive Models White Paper fails to demonstrate the need for a transformative regulatory response to modern underwriting. Yet it directs regulators to require a showing of causation/intuitiveness for, and to apply a disparate impact/proxy variable standard to, all rating factors — major changes to established norms, with no basis in controlling law, which cannot be instituted by white paper.

The White Paper explains that “predictive models … use statistics to predict outcomes” and “estimate … expected value of … loss” — but this is not a new way of classifying risk. Its caricature of “data mining” practices as “cherry-picking … spurious … results” that “churn up … simply random non-meaningful correlations” misrepresents typical insurer vetting and risk evaluation methods.

The White Paper grafts causation onto state rating statutes: “[T]hroughout this white paper, the regulator asks the modeler to go beyond correlation and document their basic, causal understanding of how variables … are related to risk” by requiring an “explanation … beyond demonstrating correlation.”

This isn’t the law. Courts applying statutory prohibitions on unfair discrimination judge “the equitableness of a rating factor by its predictive accuracy”; reject the standard of “a causal connection to expected losses”; and further instruct that “assessing risks of future … costs on an actuarial basis cannot be prohibited.” This is “fair discrimination”: Safer risks pay more accurate, lower premiums based on their cost to insure, and pairing price with risk promotes insurer solvency – both essential consumer protection concerns.

Correlation, not causation, is the cornerstone of actuarial justification. Actuarial authorities turn on whether “the variation in … anticipated experience correlates to the risk characteristic,” and explain that, because “the categorization of variables as ‘causal’ or ‘non-causal’ is ambiguous[,] … correlated variables provide more accurate premiums and are thus more desirable.”

The White Paper directs reviewers to “obtain a complete data dictionary including … each … proxy variable” and to identify “concerns related to … proxies for prohibited variables.” This implements a disparate impact standard for unintentional discrimination, in conflict with the NAIC’s stated legal position that “‘disparate impact theory’ … overthrows state laws … that allow … neutral underwriting guidelines.”

The White Paper likely constitutes an unpromulgated regulation under state Administrative Procedure Acts, which require administrative rulemaking for each “standard, directive or statement of general applicability which effectuates or interprets law or policy.”

Prevailing law establishes an objective correlation standard, supplemented by legislative bans or limits on specific rating factors. Although a previous NAIC committee studying causation stated that, “as public policy issues, questions regarding discriminatory practices should be determined in the public forum of the legislature,” the White Paper directs regulators to claim these subjective political judgments for themselves.

Resource Details

Publish Date

March 6, 2020

Topics

  • Disparate Impact
  • NAIC