New Causal Framework Exposes Gaps in Motor Insurance Pricing
A multiscale DAG model shows that annual rating factors often mask the actual mechanisms that cause road crashes.
Annual insurance rating coefficients can predict claim frequencies with high accuracy while failing to explain why crashes happen. This gap exists because road safety operates on the scale of seconds and trips, but insurance data aggregates liability claims over entire policy years. A new multiscale causal Directed Acyclic Graph (DAG) framework bridges this divide by linking a crash-occurrence graph to an observation layer and a downstream claim-administration process.
The framework shifts the goal from estimating the causal effect of a single rating factor to identifying which mechanism laws are compatible with observed insurance contrasts. This approach reveals that traditional rating variables often act as proxies for latent behaviors or exposures rather than direct causes. For example, in a French portfolio, the claim-frequency relativity for drivers aged 18–20 versus 40–49 dropped from 3.388 to 1.235 when conditioning on bonus-malus categories, as those categories summarize endogenous insurance history.
This disconnect creates a structural risk for insurers relying on static rating factors. If a pricing model identifies a correlation between a demographic trait and claim frequency without understanding the underlying crash mechanism, the model may fail to adapt when external conditions—such as vehicle safety tech or urban infrastructure—change. The research shows that a sublinear mileage relation can constrain aggregate exposure without identifying the actual composition of that exposure.
Companies providing telematics and trip-level data are the primary beneficiaries of this shift. The framework demonstrates that stronger mechanistic claims require trip-level intermediate states and linked crash-claim observations to move beyond coarse bookkeeping. This increases the value of high-resolution behavioral data over traditional demographic proxies.
Underwriters and actuaries face a limit on how much they can infer from annual portfolios. In a Spanish age-mediation estimate, the framework narrowed only one coarse bookkeeping block under explicit transport-sensitivity assumptions, leaving the resulting region wide. This suggests that current insurance datasets are mathematically insufficient to isolate specific safety drivers.
Future pricing accuracy depends on integrating real-time road safety analytics into the actuarial pipeline. The transition from predictive correlation to causal mechanism will require a fundamental change in how insurers collect and link data between the moment of impact and the final claim administration.