rating variables

When you ask for an insurance quote, you answer a list of questions: how old are you, where do you live, what do you drive, have you claimed before? Each of those questions corresponds to a rating variable — a measurable characteristic of the risk that the insurer uses to predict expected cost and therefore to set the price.

A rating variable is any attribute the price is allowed to depend on. Some are about the insured (a driver's age, a homeowner's claim history), some about the property (a car's model, a building's construction and age), and some about the environment (territory, distance to a fire station). For pricing, each variable is split into levels, and each level carries a relativity. A good rating variable should be related to the loss for a sensible reason, be measurable and verifiable, be hard to manipulate, and be acceptable both legally and to the public. Mileage driven, for instance, is an excellent variable because more driving genuinely means more chance of a crash; it is increasingly measured directly through telematics.

Choosing rating variables is where statistics, business, law, and ethics all collide. A variable can predict loss strongly yet still be barred — using race is illegal everywhere, and many jurisdictions restrict gender, credit, or postcode. Two variables can also be correlated (older drivers tend to own different cars), which is exactly why modern pricing uses generalized linear models to untangle each variable's true effect rather than crediting it twice. The honest caveat: a variable that merely correlates with risk is not the same as a cause, and regulators increasingly ask whether a chosen variable is fair as well as predictive.