GLM pricing studio
Bring your own policy-level data. Fit Poisson frequency, Gamma severity, Tweedie pure-premium, logistic retention or Gaussian models with an exposure offset, then read relativities, standard errors, deviance, Gini and lift β all computed locally in your browser; nothing is uploaded.
Assumptions & limitations
- Treatment coding with the most-exposed level as base; numeric predictors enter linearly (bin them into categories to capture curvature).
- Standard errors use the observed Fisher information scaled by the Pearson dispersion (Ο fixed at 1 for Poisson and binomial).
- AIC is deviance-based and comparable across models of the same family and data, not across families.
- No interactions, splines or regularisation yet β use the notes to record judgemental selections before saving.