Generative AI is no longer confined to marketing, creative work, or coding. A recent study by the Bank of Japan (BOJ) shows its potential as a tool for economic analysis. The research explores how large language models (LLMs) can act as “economic agents” in simulations, testing whether their decision patterns align with real-world economic behavior.

Experiment Design: AI as Economic Agents

The study uses an Agent-Based Model (ABM) framework, placing LLMs in the roles of consumers and firms:

Consumer Simulation

Scenario: AI adjusts spending decisions under different nominal wages and price levels.

Observation: Spending generally rises when real wages increase and contracts when prices rise.

Insight: The AI's behavior aligns with classical Keynesian consumption theory, demonstrating that LLMs can replicate realistic consumer responses.

Firm Simulation

Scenario: Two market structures—monopoly and duopoly.

Observation: In monopoly-like conditions, AI tends to pass rising costs onto consumers; in duopoly competition, it moderates price changes to maintain market share.

Insight: These outputs reflect fundamental microeconomic principles, suggesting that AI can mimic strategic pricing behavior.

Why It Matters

Traditional ABM relies on extensive parameterization and behavioral assumptions, which can be limiting in complex or data-sparse markets. LLMs as agents offer key advantages:

Behavioral heterogeneity: AI can generate diverse, non-linear decision patterns across simulated agents.

Data supplementation: LLMs can produce “synthetic microdata” where survey or transaction data is insufficient.

Scenario analysis: Multiple policy or market shock scenarios can be tested rapidly with AI-generated behaviors.

For example, changes in wages or prices simulated by LLMs could inform stress-testing of policy measures or market demand projections.

Limitations and Risks

The BOJ study also highlights important constraints:

Training-data dependency: AI behavior reflects patterns in its training corpus, not economic theory.

Limited interpretability: Generated decisions lack fully traceable mechanisms, complicating direct policy use.

Context sensitivity: AI may deviate from rational behavior in unfamiliar or extreme scenarios.

Consequently, AI currently serves as a supplementary tool rather than a replacement for traditional economic models.

Implications for Investors and Policymakers

For cross-border investment, infrastructure, and policy analysis:

Augment economic modeling: In markets or sectors lacking microdata, AI can provide actionable behavioral insights.

Enhance scenario analysis: Rapidly evaluate potential responses to policy, cost, or market shocks.

Maintain risk awareness: AI-generated outputs should be interpreted alongside traditional models and expert judgment to ensure robustness.

BOJ’s research illustrates the emerging potential of integrating generative AI into economic simulations. It is not a crystal ball for predicting the future, but a new instrument for enriching analysis, informing policy decisions, and supporting investment research in data-constrained environments.