ADAGE: A generic two-layer framework for adaptive agent based modelling

TL;DR

ADAGE addresses the Lucas critique in agent-based modelling by introducing a bi-level, Stackelberg framework where an outer layer optimizes environment characteristics $\boldsymbol{θ}$ and an inner ABM layer learns $\pi_i(a|o_i, \hat{θ}_i)$ conditioned on those characteristics. The framework unifies policy design, calibration, scenario generation, and robust behavioural learning as instantiations of a single optimisation problem, solved approximately via alternating gradient methods and DRL or alternative optimisers. Across TaxAI policy design, Cobweb calibration, Tobin-tax scenario generation, and robust meta-learning for market makers, ADAGE yields improved welfare, tighter calibration to data, reduced volatility, and robust policies that generalise across varied conditions. These results demonstrate a general, algorithm-agnostic approach to adaptive ABMs with practical implications for policy analysis, economic modelling, and risk assessment.

Abstract

Agent-based models (ABMs) are valuable for modelling complex, potentially out-of-equilibria scenarios. However, ABMs have long suffered from the Lucas critique, stating that agent behaviour should adapt to environmental changes. Furthermore, the environment itself often adapts to these behavioural changes, creating a complex bi-level adaptation problem. Recent progress integrating multi-agent reinforcement learning into ABMs introduces adaptive agent behaviour, beginning to address the first part of this critique, however, the approaches are still relatively ad hoc, lacking a general formulation, and furthermore, do not tackle the second aspect of simultaneously adapting environmental level characteristics in addition to the agent behaviours. In this work, we develop a generic two-layer framework for ADaptive AGEnt based modelling (ADAGE) for addressing these problems. This framework formalises the bi-level problem as a Stackelberg game with conditional behavioural policies, providing a consolidated framework for adaptive agent-based modelling based on solving a coupled set of non-linear equations. We demonstrate how this generic approach encapsulates several common (previously viewed as distinct) ABM tasks, such as policy design, calibration, scenario generation, and robust behavioural learning under one unified framework. We provide example simulations on multiple complex economic and financial environments, showing the strength of the novel framework under these canonical settings, addressing long-standing critiques of traditional ABMs.

Figures (10)

• Figure 1: ADAGE: Two-layer framework.
• Figure 2: TaxAI: Economic Simulator
• Figure 3: Social welfare throughout training. The solid line shows the mean value, and the filled region shows the entire range (min $\dots$ max).
• Figure 4: Policy Design: Resulting mean household work, savings, and wage rates throughout training.
• Figure 5: Household inequality from multiple rollouts.