Tariff Threats, Macroeconomic Expectations, and Policy Communication Strategies: Experiments Based on a Multi-Agent System

TEXT START: Tariff threats can move household beliefs before policy is enacted, yet their rapidly changing language is difficult to study with conventional surveys.

The Dissection

This paper builds a synthetic expectation-management laboratory. It converts 300 survey respondents into persistent LLM personas, exposes them to simulated social-media narratives, and varies rhetorical inputs such as salience, complexity, credibility, and sender identity.

Its real product is not verified knowledge of households. It is a controllable map of how machine-generated proxies react to messages. The central-bank experiment makes the purpose explicit: official communication becomes an intervention for coordinating beliefs before policy effects fully arrive.

The method also enacts the Discontinuity Thesis. Survey replication, behavioral simulation, and qualitative interpretation—formerly human cognitive labor—are being delegated to machines. Humans remain as validators and institutional gatekeepers, not necessarily as producers of the analysis.

The Core Fallacy

The paper risks treating behavioral resemblance as economic equivalence. Reproducing some demographic patterns does not prove that LLM agents possess human stakes, material exposure, memory, distrust, or causal reasoning. These agents generate plausible textual reactions to language; they do not have rent, debt, job insecurity, or bargaining power.

The deeper fallacy is mistaking belief coordination for economic survival. Better communication may compress expectation dispersion or shift inflation and unemployment forecasts. It cannot restore the mass employment → wage → consumption circuit once cognitive labor becomes cheaper to automate. Communication is a lag defense, not a reversal mechanism. It can coordinate a population around managed decline just as efficiently as around recovery.

Hidden Assumptions

• Calibration against survey distributions is treated as evidence of psychological and causal validity.
• Simulated social-media exposure is assumed to reproduce real attention, memory, conflict, and algorithmic feedback.
• Prompt-conditioned responses are treated as household expectations rather than model artifacts.
• A sample of 300 agents is assumed to capture consequential heterogeneity, strategic behavior, and tail events.
• Central-bank credibility is assumed to remain available when institutions are losing control of material conditions.
• Message complexity and narrative framing are treated as controllable policy variables rather than outputs of political conflict.
• Expectations are assumed to remain the primary macroeconomic bottleneck even after productive participation collapses.
• Human validation is assumed to be timely, independent, and capable of detecting failures that the calibration procedure cannot see.
• Policymakers are assumed to be able to coordinate beliefs without simultaneously intensifying distrust or propaganda dynamics.

Social Function

Primary classification: transition management.

Secondary classifications: prestige signaling and partial truth, with a strong risk of ideological anesthetic.

The partial truth is real: language affects attention, credibility, causal narratives, and expectations. The anesthetic appears when this insight is inflated into a theory of stabilization. It encourages institutions to optimize the population’s perception of shocks rather than confront the structural erosion of labor’s economic necessity.

The paper gives states a cheaper test bench for pretesting announcements, identifying panic triggers, and tuning official narratives. That is useful operationally. It is also a method for administering decline with less visible friction.

The Verdict

Technically useful, systemically insufficient. The framework may become part of the state’s lag-defense arsenal, but it neither challenges nor weakens P1, P2, or P3. It models how the crowd reacts to disruption, not the disruption that destroys the old economic order.

Its central irony is decisive: a paper studying how to manage human expectations does so by converting humans into cheaper synthetic approximations. That is not proof the system is already dead. It is evidence that cognitive substitution is advancing. Under the Discontinuity Thesis, this is a stethoscope and loudspeaker beside a failing engine—hospice-grade control, not survival.