burning-cost.github.io

Insurance pricing education and open-source tooling for UK actuaries and pricing teams.

The site at burning-cost.github.io publishes worked examples, methodology explainers, and links to all open-source libraries.

Flagship Libraries

Ten libraries we consider genuinely differentiated — tools for hard problems in UK pricing where no adequate open-source Python solution existed before.

Regulatory compliance

Library	Problem it solves
insurance-fairness	Proxy discrimination auditing aligned to FCA Consumer Duty and Equality Act 2010
insurance-governance	PRA SS1/23 model validation reports: bootstrap Gini CI, A/E CI, double-lift, renewal cohort test

Causal inference

Library	Problem it solves
insurance-causal	Double machine learning to deconfound rating factors where channel and behaviour bias standard GLM coefficients

Uncertainty quantification

Library	Problem it solves
insurance-conformal	Distribution-free prediction intervals with finite-sample coverage guarantees; Solvency II SCR bounds
insurance-distributional-glm	GAMLSS for Python: model mean, dispersion, shape, and zero-inflation as functions of covariates

Smoothing and experience rating

Library	Problem it solves
insurance-whittaker	Whittaker-Henderson 1D/2D smoothing with REML lambda selection and Bayesian credible intervals
insurance-credibility	Bühlmann-Straub credibility in Python with mixed-model equivalence — caps thin segments, stabilises NCD factors

Telematics

Library	Problem it solves
insurance-telematics	HMM driving state classification from raw 1Hz GPS/accelerometer data to GLM-compatible risk scores

Advanced modelling

Library	Problem it solves
insurance-frequency-severity	Sarmanov copula joint frequency-severity with analytical premium correction — tests the independence assumption every model makes
insurance-gam	EBM and Neural Additive Model: transparency of a GLM, predictive power of a neural network

Full Library Portfolio

34 libraries in total, covering the full pricing workflow from data validation through to deployment and monitoring. See burning-cost.github.io/tools for the complete list.

Site structure