Definition:Regression discontinuity
📉 Regression discontinuity is a quasi-experimental research design that exploits a known cutoff or threshold in a continuous variable to estimate causal effects, treating observations just above and just below the threshold as near-randomly assigned to different conditions. In insurance, thresholds abound — policy eligibility age limits, premium pricing tiers triggered by credit score bands, deductible breakpoints in commercial programs, regulatory capital trigger levels, and risk classification boundaries — making regression discontinuity a naturally suited tool for the industry's analytical challenges.
⚙️ The design works by comparing outcomes for units narrowly on either side of a cutoff, under the logic that individuals or policies just above and just below the threshold are essentially similar in all respects except their treatment assignment. For example, if an insurer applies a different underwriting protocol to commercial properties above and below a specific total insured value, a regression discontinuity analysis can estimate the causal impact of that protocol on claims frequency or loss ratios by focusing on properties near the threshold. Similarly, regulators studying the effect of a mandatory coverage requirement that applies only to firms above a certain employee count can use the design to measure the policy's impact on claim outcomes. The approach requires careful validation: the assignment variable must not be manipulable by the subjects (a condition called "no sorting around the cutoff"), and the analyst must verify that other relevant characteristics do not jump discontinuously at the threshold. Both "sharp" designs — where the cutoff perfectly determines treatment — and "fuzzy" designs — where the cutoff strongly influences but does not guarantee treatment — are used, with the fuzzy variant being more common in insurance settings where compliance with eligibility rules is imperfect.
💡 Regression discontinuity occupies a valuable niche between the interpretive power of randomized experiments and the practical accessibility of pure observational analysis. Because it requires only data near a threshold rather than a full experimental apparatus, it can be applied retrospectively to existing books of business and historical claims data. Insurers and insurtech firms use it to evaluate whether specific rating factor cutoffs genuinely separate meaningfully different risk profiles or simply impose arbitrary boundaries. Regulators in markets as diverse as the United States, the EU, and parts of Asia have drawn on regression discontinuity evidence when assessing the efficacy of mandatory insurance programs or coverage thresholds. As the industry's appetite for causal — rather than merely predictive — insight grows, regression discontinuity offers a rigorous, transparent methodology that decision-makers and supervisors can readily understand and trust.
Related concepts: