From Model to Scenario Engine: A Reproducible ABM for UK Net Zero Analysis

Abstract

Agent-based modelling (ABMs) has increasingly been used to study socio-economic transitions. These computational models feature heterogeneous agents, decision-making, and feedback effects that are difficult to capture in aggregate models. Yet, these features that make ABMs valuable also make them hard to reproduce and evaluate as research software, especially when stochastic behaviour is involved. In this talk, we share the software engineering effort to transform a baseline ABM for macroeconomic analysis into a tool for comparing UK Net Zero policy intervention scenarios. The methodology combines design choices, including: Coding economic shocks targeting different economic markets A UK-specific data pipeline Monte Carlo simulation (MC) Reproducibility controls Visualisation dashboards Simulation-based inference (SBI) We show that a more straightforward separation of the main logic from scenario interventions in the code makes it easier to test, maintain, and extend with new economic behaviours. We also explain scenario uncertainty with MC and SBI. Finally, we discuss how dashboards make outputs easier to inspect and compare for both technical and non-technical audiences. Although the work is motivated by UK Net Zero policy analysis, the broader contribution is a dialogue of software engineering practices for making large-scale stochastic ABMs more credible and reusable as research software. The talk will be relevant to anyone working on computational models where scientific value depends as much on implementation rigour as on the underlying theory. In addition to making the ABM applicable to policy analysis, we argue that the design choices improve the model's implementation, enabling better software quality assessments.