arXiv subjects

Jianhao Lin

Publications and source records attributed to Jianhao Lin.

3 recordsLinked to original sources

arXiv2026

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

Tariff threats can move household beliefs before policy is enacted, yet their rapidly changing language is difficult to study with conventional surveys. We build a multi-agent system that turns 300 households from the Michigan Surveys of Consumers into persistent large-language-model agents exposed to social-media information over several simulated months. Calibrated agents reproduce some distributional and demographic patterns in human survey data collected after the announcement of Liberation Day tariffs. Simulated experiments indicate that immediacy, rate salience, semantic progression, message complexity, narrative, and sender identity jointly shape inflation and unemployment expectations and their dispersion. Open-ended responses trace these effects to attention, ambiguity, credibility, and causal narratives. A second experiment finds that central-bank explanations can coordinate beliefs, although their effects on average expectations depend on message content. The framework supports disciplined exploration of policy communication, subject to human validation rather than as a substitute for it.

econ.GN↗

arXiv2026

DataClawBench: An Agent Benchmark for Exploratory Real-World Financial Data Analysis

Autonomous data analysis agents are increasingly expected to conduct exploratory analysis with limited human guidance about data. However, existing benchmarks typically evaluate such agents in prior-guided settings, providing selected data sources, explicit data schemas, or cleaned data, thereby understating the exploratory burden. To evaluate this realistic exploratory data analysis task, we introduce DataClawBench, a benchmark built from financial think-tank consulting scenarios where agents must independently explore unfamiliar, noisy, cross-domain data and produce verifiable conclusions. DataClawBench provides a unified real-world data environment with approximately 2.06 million records across enterprise, industry, and policy domains, with native data noise preserved. On top of this data environment, it defines 492 multi-step cross-domain tasks, each annotated with intermediate milestones that diagnose exploration and reasoning failures beyond outcome accuracy. A systematic evaluation of eight advanced LLMs under the OpenClaw agent reveals that exploratory data analysis breaks agent reliability: more exploration does not reliably translate into task-relevant progress or correct final answers.

cs.AI↗

arXiv2025

Simulating Macroeconomic Expectations in Survey Experiments with LLM-based Economic Agents

We introduce a framework for simulating macroeconomic expectations in survey experiments using LLM-based economic agents (LLM Agents). We construct LLM Agents equipped with several functional modules that retrieve personal characteristics, prior expectations, and dynamic external information. We validate our framework by recapitulating three representative survey designs covering various expectations across different types of respondents. Our results show that LLM Agents generate expectation distributions highly similar to human data and capture human-aligned qualitative patterns in open-ended responses. Evaluation reveals that priors are crucial for matching distributions, whereas personal and external information drive human-like thought processes. Our findings offer guidance for narrowing the belief gap between generative AI and humans at the aggregate level while delineating the boundaries of the framework.

econ.GN↗