Definition

What is a frequency-severity model?

A frequency-severity model is the standard actuarial pricing model that estimates expected loss as the product of two separately fitted models: how often claims occur per unit of exposure (frequency) and how large they are when they do (severity).

Also: frequency severity GLM · Poisson-Gamma model · loss cost model · pure premium model

Updated 27 October 2026 · Written and reviewed by the Soldek team

Frequency is usually a Poisson generalised linear model on claim counts with the logarithm of exposure as an offset; severity a Gamma model on average claim size, weighted by the number of claims. Both produce multiplicative relativities per rating factor: a value of 1.3 for a feature means that segment runs at 130 % of the base. Expected loss per exposure unit is frequency times severity, calibrated so that total expected equals total actual on the portfolio; expenses and margin are loaded on top to reach a premium.

The method is preferred over single black-box models in insurance because its output is a set of factors a regulator, a capacity provider and an underwriter can read, and because those factors have the same shape as a rate table. Gradient-boosted models are often fitted alongside to find interactions, then distilled back into additional GLM terms or bands so the tariff stays explainable.

On Soldek

Soldek does not fit models; it produces the frame they are fitted on and applies the result. The relativities a frequency-severity model yields are written into the product workbook as rate factors and the uncertain segments as referral rules; the deterministic rating engine then prices every quote from that tariff, so each premium traces back to sum insured × base rate × the factors that applied.

Your data

Questions

Why fit frequency and severity separately instead of modelling loss directly?
Because different factors drive them. Age may change how often a pet claims; breed may change how expensive a claim is. Modelling them separately lets each factor act where it matters, gives cleaner diagnostics, and matches how reinsurers and regulators expect a rate review to be presented.
Can the model price quotes live?
It can, but in insurance it usually should not: a live model is a black box between the insurer and its price. The common practice is to let the model write the tariff (factors, bands, referral rules) and have a deterministic engine apply it, which keeps every premium explainable.

See it running.

Book a demo and we show you the complete platform live, from product builder to bordereau.