Multiplicative coefficient applied to a base premium for a given level of a rating criterion, the elementary unit of a rating table and the form in which a statistical model becomes a rate.
A model returns coefficients on a logarithmic scale, unreadable outside the team that produced them. A relativity is their operational translation: the exponential of the coefficient, expressed against a reference level set to one. A relativity of 1.30 means thirty percent more than the reference, all else equal, and it is that last qualifier that most often gets lost along the way. Two pitfalls attend the move from model to table. The first is the choice of reference, which changes nothing in the final rate but everything in its reading: taking a rare level as reference makes every relativity unstable. The second is the gap between raw relativity, read from the data without controlling for other variables, and model relativity, which controls. A powerful vehicle looks twice as risky raw, but part of that gap belongs to its driver's age, and presenting the raw relativity double counts the same effect. Finally, technical relativity is not applied relativity: commercial smoothing, rate increase caps and market constraints slot in between, and the actuary's job is to document the gap rather than quietly absorb it.
Motor rating table, 2026 year, base premium 420 EUR on the reference level. Raw relativity of the young driver level: 2.45. Model relativity, after controlling for license seniority, vehicle and territory: 1.72. The 0.73 gap was carried by the three other criteria, and applying it would have overcharged an already expensive segment by 43%, with the adverse selection that follows.
relativité, rating relativity, coefficient tarifaire, facteur de tarif