PolicyEngine — UK open-source tax-benefit microsimulation for policy analysis

PolicyEngine is a Python-based, open-source microsimulation model of the UK tax and benefit system, developed by the US-registered nonprofit PolicyEngine, applying machine-learning calibration and gradient-descent imputation to five UK household datasets (chiefly the Family Resources Survey) to simulate income tax, National Insurance, Universal Credit, Child Benefit, capital gains tax, council tax, and over 50 other provisions.

Enable population-level simulation of tax and benefit reforms, producing household-level net income/marginal tax rate estimates and population-level poverty and inequality metrics.

Since summer 2025, PolicyEngine co-founder and CTO Nikhil Woodruff has served as an Innovation Fellow with 10DS (10 Downing Street's data science team), which built a bespoke '10ds-microsim' package atop PolicyEngine to estimate distributional and local-authority-level impacts of proposed reforms; HM Treasury separately published a Beta/Pilot Algorithmic Transparency Record documenting its exploration of PolicyEngine alongside its existing IGOTM model; £251,296 in Nuffield Foundation funding (2024-2025) extended geographic analysis to constituency and local-authority level.

The model reduces deviation from official administrative totals by approximately 97% through reweighting; results were intensively validated against OBR forecasts before deployment; the HMT transparency record notes that approximately 60% of National Insurance calculations fall within 0.5% of existing model outputs, while income-tax comparisons require further evaluation.

Used as supplementary, experimental analysis when the policy process moves faster than traditional analytical pipelines allow; the fully open-source codebase (Apache 2.0, GitHub) has also been used by UK political parties for manifesto costing and by academic researchers.