A Multi-LLM-Agent-Based Framework for Economic and Public Policy Analysis

TL;DR

This work introduces a Multi-LLM-Agent-Based (MLAB) framework that uses heterogeneous Large Language Models (LLMs) to simulate economic decision-making across population segments. It first benchmarks five LLMs on a canonical two-period consumption-savings problem with explicit CRRA utility, deriving analytical solutions and calibrating parameters from CFPS data. It then extends to a two-dimensional heterogeneity approach by mapping LLMs to educational groups and applying this to an interest income taxation case study, revealing distinct, policy-relevant behavioral patterns that differ from standard homogeneous-agent models. The findings demonstrate that combining economic circumstances with intrinsic LLM reasoning creates richer simulations for public policy analysis and highlights potential for customized, segment-aware policy design, while acknowledging the need for further refinement in agent mapping and interaction dynamics.

Abstract

This paper pioneers a novel approach to economic and public policy analysis by leveraging multiple Large Language Models (LLMs) as heterogeneous artificial economic agents. We first evaluate five LLMs' economic decision-making capabilities in solving two-period consumption allocation problems under two distinct scenarios: with explicit utility functions and based on intuitive reasoning. While previous research has often simulated heterogeneity by solely varying prompts, our approach harnesses the inherent variations in analytical capabilities across different LLMs to model agents with diverse cognitive traits. Building on these findings, we construct a Multi-LLM-Agent-Based (MLAB) framework by mapping these LLMs to specific educational groups and corresponding income brackets. Using interest-income taxation as a case study, we demonstrate how the MLAB framework can simulate policy impacts across heterogeneous agents, offering a promising new direction for economic and public policy analysis by leveraging LLMs' human-like reasoning capabilities and computational power.

Figures (6)

• Figure 1: Consumption choices by LLMs. The diagonal line represents the budget constraint, and the dashed lines intersect at the theoretical optimal point. Each point represents one trial response.
• Figure 2: Distribution of consumption choices across periods.
• Figure 3: Consumption choices by LLMs without utility function specification. The diagonal line represents the budget constraint.
• Figure 4: Distribution of consumption choices without utility function specification.
• Figure 5: Saving Rate Analysis (1-c1/(w0+y1))