Open, tested algorithms for actuaries and risk managers, as an alternative to closed commercial software.
Everything here is free to use and fork under the MIT licence (the asset-liability model uses MPL-2.0). Each algorithm comes with its source paper or regulatory document, a worked example and tests, so you can check the result instead of trusting it.
Open-Source Modelling started in Milan in 2021. It is funded and maintained by Qnity Consultants.
| Repository | What it is |
|---|---|
| insurance_python | Every Python algorithm in one place: yield curves, short-rate models, bootstrapping, time series |
| insurance_matlab | The same algorithms for Matlab users |
| insurance_jupyter | Notebooks that walk through the methods step by step |
| insurance_skills | AI assistant skills for actuarial work, where the calculation runs in Python and the assistant is just the interface |
| Open_Source_Economic_Model | A full asset-liability model (OSEM) for insurers and pension funds, with a methodology document |
| Light_Economic_Generator | Simple economic stochastic generator with 3 different stochastic models |
Yield curves and Solvency II
- Smith-Wilson interpolation and extrapolation, following EIOPA's technical documentation, with calibration of alpha
- Nelson-Siegel-Svensson curve fitting
- Checks that recalculate EIOPA's monthly risk-free rate curves
- Every historical EIOPA curve since December 2014, in three CSV files
Interest rates and economic scenarios
- Vasicek (one- and two-factor), Hull-White one-factor and Dothan short-rate models
- Black-Scholes, correlated Brownian motion and binomial option pricing
- A lightweight economic scenario generator
- A validation framework for a monthly scenario generator, a reduced version of a framework built for an Italian bancassurance group
Statistics and time series
- Stationary bootstrap, with automatic block-length calibration
- Singular spectrum analysis
Open data
- SFCR tables for 22 Italian life insurers (S.02.01, S.23.01, S.25.01), extracted with integrity checks
- Pipelines for EDGAR sovereign emissions and IMF PPP GDP, for carbon-intensity reporting
Italian-language versions of several algorithms are in assicurazione_python.
Teams are starting to use AI assistants such as Claude, Copilot and ChatGPT in their daily work, but there's been no shared place for insurance-specific tools. insurance_skills is that place: an open collection of reusable assistant skills.
Every skill follows one rule. The checks and calculations run in tested Python; the assistant is only the interface. You ask in plain English, and the answer comes from code you can read, so the model can't make up a number.
| Skill | What it does |
|---|---|
| historic-eiopa-yield-curve | Returns any historic EIOPA risk-free rate curve, for any country, maturity and date since December 2014, including forwards, discount factors and stressed curves |
| spontaneous-testing-for-excel | You list tests for a workbook in plain English on a TEST sheet. The skill runs each one and writes back what it checked and whether it passed, without touching anything else |
| validate-ul-fund-data | Runs 22 checks on the shape, types and internal consistency of a periodic fund data extract. It returns a pass/fail report and a SHA-256 fingerprint of the file, so bad data is caught before a re-run |
| ranking_life_script | Gives several AI models the same customer scenario and records how each one ranks a list of life insurers, and why. It measures how AI assistants perceive your company next to its peers |
AI also does the heavy lifting in our open data. The SFCR tables are extracted from PDF reports with Mistral OCR, then validated with accounting integrity checks. Those checks have already caught a likely transposition error in a published report.
Each team's data is different, so fork a skill and adapt its rules. Does your team have a skill that works well? Share it. Contributions are very welcome.
Suggestions, bug reports and new algorithms are welcome.
- Found a bug or a question? Open an issue on the relevant repository.
- Have an implementation or an AI skill your team uses? Get in touch and we'll help publish it in
insurance_skills.
We support the Open-Source Modelling development by taking paid project work. Open code is where a model starts. Making it hold up in reporting takes more work: validation, integration and training. Through Qnity Consultants we do exactly that for life insurers, pension funds and ALM teams.