Elamur Research⌕
☰
home / papers / 2026 / Agentic Economic Modeling

Agentic Economic Modeling

Paper · 2026 · Bohan Zhang, Jiaxuan Li, Ali Hortacsu et al.

Summary

Business experiments such as RCTs and conjoint surveys are expensive. LLM "homo silicus" agents can produce choice data cheaply, but their preferences are systematically skewed and less varied than those of real people. Econometric estimates computed directly on raw LLM output are therefore untrustworthy. The authors, from Amazon and several universities, propose Agentic Economic Modeling (AEM), a three-step Generation–Correction–Inference pipeline:

  1. Persona-conditioned LLMs simulate decisions.
  2. A correction model, fitted on a small human sample, maps task features and LLM choices to human-like choices.
  3. Standard estimators recover demand elasticities or treatment effects from the corrected data.

The design draws on random-coefficient discrete choice. It treats a population as a weighted blend of synthetic personas and learns the weights from aggregate data. Tests on proprietary e-commerce delivery data show the following:

  • Conjoint study: calibrating on a tenth of the human data improves demand-parameter accuracy, whereas raw LLM choices make it worse.
  • Regional field experiment: the persona mixture reproduces the national treatment effect in held-out regions.

AEM could thus cut how much human experimentation an RCT needs.

Key points

  • Conjoint study: the study covers over 4 million observations from tens of thousands of customers. Training the correction model on only 10% of customer choices lowers the error of demand estimates, while uncorrected LLM choices raise it.
  • Regional experiment setup: treatments were assigned by ZIP3 (900 ZIP3s × 60 days, 35k+ region-day observations), and 17k+ synthetic personas were generated per region. The outcome was the share of Same-Day delivery.
  • Out-of-domain regions: a mixture model calibrated on 10% of regions estimated −65 ± 10 bps, against −60 ± 8 bps for the full human experiment. A black-box integrated model reached only −39 ± 9 bps.
  • Time-wise extrapolation: with only first-day human data, the corrected estimate was −24 bps (95% CI [−26, −22], p<1e-5). The human-only estimate for the same day was −17 bps (CI [−43, +9], p=0.2049). Raw LLM output gave −350 bps.
  • Uncertainty estimation: uncertainty comes from a full-pipeline bootstrap with 20 resamples of personas and regions, retraining the correction each time. This reflects ideas from prediction-powered inference (PPI).

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Massachusetts Institute of TechnologyCambridge · United States243 works · 76,506 citations🎓University of MichiganAnn Arbor · United States86 works · 15,493 citations🎓University of ChicagoChicago · United States52 works · 50,893 citations

 Fields, topics and concepts

📚 Shelves 8The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksSocial Sciences · 1,954 worksGeneral Social Sciences · 254 worksEconomics, Econometrics and Finance · 23 worksEconomics and Econometrics · 16 works

Cite

Cite as Bohan Zhang, Jiaxuan Li, Ali Hortacsu, Xiaoyang Ye, Victor Chernozhukov, Anqi Ni and Edward W Huang. 2026. Agentic Economic Modeling. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-industry.80

@inproceedings{zhang2026agentic,
  title = {{Agentic Economic Modeling}},
  author = {Bohan Zhang, Jiaxuan Li, Ali Hortacsu, Xiaoyang Ye, Victor Chernozhukov, Anqi Ni and Edward W Huang},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-industry.80},
  url = {https://doi.org/10.18653/v1/2026.acl-industry.80}
}

References

27 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Anastasios N. Angelopoulos, Stephen Bates, Clara Fannjiang, Michael I. Jordan, and Tijana Zrnic. 2023a. Prediction-powered inference. Science, 382(6671):669–674. find
  2. Anastasios N. Angelopoulos, John C. Duchi, and Tijana Zrnic. 2023b. PPI++: Efficient prediction-powered inference. arXiv preprint arXiv:2311.01453. — [arXiv:2311.01453](https://arxiv.org/abs/2311.01453) find
  3. Dzmitry Bahdanau, Kyunghyun Cho, and Yoshua Bengio. 2014. Neural machine translation by jointly learning to align and translate. arXiv preprint arXiv:1409.0473.
  4. Marcel Binz, Elif Akata, Matthias Bethge, Franziska Brändle, Fred Callaway, Julian Coda-Forno, Peter Dayan, and 1 others. 2025. A foundation model to predict and capture human cognition. Nature, pages 1–8. find
  5. Julien Boelaert, Samuel Coavoux, Etienne Ollion, Ivaylo Petev, and Patrick Präg. 2025. Machine bias. how do generative language models answer opinion polls? Sociological Methods & Research, page 00491241251330582. find
  6. Ngoc Bui, Hieu Trung Nguyen, Shantanu Kumar, Julian Theodore, Weikang Qiu, Viet Anh Nguyen, and Rex Ying. 2025. Mixture-of-personas language models for population simulation. In Findings of the Association for Computational Linguistics: ACL 2025, pages 24761–24778, Vienna, Austria. Association for C… find
  7. Chaoran Chen, Weijun Li, Wenxin Song, Yanfang Ye, Yaxing Yao, and Toby Jia-Jun Li. 2024. An empathy-based sandbox approach to bridge the privacy gap among attitudes, goals, knowledge, and behaviors. In Proceedings of the 2024 CHI Conference on Human Factors in Computing Systems, pages 1–28. find
  8. Yiting Chen, Tracy Xiao Liu, You Shan, and Songfa Zhong. 2023. The emergence of economic rationality of GPT. Proceedings of the National Academy of Sciences, 120(51):e2316205120. find
  9. R. Maria del Rio-Chanona, Marco Pangallo, and Cars Hommes. 2025. Can generative AI agents behave like humans? evidence from laboratory market experiments. Preprint, arXiv:2505.07457. — [arXiv:2505.07457](https://arxiv.org/abs/2505.07457) find
  10. Yann Dubois, Xuechen Li, Rohan Taori, Tianyi Zhang, Ishaan Gulrajani, Jimmy Ba, Carlos Guestrin, Percy Liang, and Tatsunori B. Hashimoto. 2024. Alpacafarm: A simulation framework for methods that learn from human feedback.
  11. Ali Goli and Amandeep Singh. 2024. Frontiers: Can large language models capture human preferences? Marketing Science, 43(4):709–722. find
  12. John J Horton. 2023. Large language models as simulated economic agents: What can we learn from Homo Silicus? http://www.nber.org/papers/w31122, (31122). NBER Working Paper. ↗
  13. Cheng-Ping Hsieh, Simeng Sun, Samuel Kriman, Shantanu Acharya, Dima Rekesh, Fei Jia, Yang Zhang, and Boris Ginsburg. 2024. RULER: What’s the real context size of your long-context language models? arXiv preprint arXiv:2404.06654.
  14. Wenlong Ji, Lihua Lei, and Tijana Zrnic. 2025. Predictions as surrogates: Revisiting surrogate outcomes in the age of AI. arXiv preprint arXiv:2501.09731. — [arXiv:2501.09731](https://arxiv.org/abs/2501.09731) find
  15. Diederik P. Kingma and Jimmy Ba. 2017. Adam: A method for stochastic optimization. Preprint, arXiv:1412.6980.
  16. Anton Korinek. 2023. Generative AI for economic research: Use cases and implications for economists. Journal of Economic Literature. find
  17. Austin C Kozlowski and James Evans. 2025. Simulating subjects: The promise and peril of artificial intelligence stand-ins for social agents and interactions. Sociological Methods & Research, page 00491241251337316. find
  18. Ang Li, Haozhe Chen, Hongseok Namkoong, and Tianyi Peng. 2025. LLM generated persona is a promise with a catch. arXiv preprint arXiv:2503.16527. — [arXiv:2503.16527](https://arxiv.org/abs/2503.16527) find
  19. Jens Ludwig, Sendhil Mullainathan, and Ashesh Rambachan. 2025. Large language models: An applied econometric framework. Working Paper 33344, National Bureau of Economic Research. find
  20. Alex Lyman, Bryce Hepner, Lisa P Argyle, Ethan C Busby, Joshua R Gubler, and David Wingate. 2025. Balancing large language model alignment and algorithmic fidelity in social science research. Sociological Methods & Research, 54(3):1110–1155. find
  21. Benjamin S Manning and John J Horton. 2025. General social agents. arXiv preprint arXiv:2508.17407. — [arXiv:2508.17407](https://arxiv.org/abs/2508.17407) find
  22. Daniel McFadden. 1980. Econometric models for probabilistic choice among products. The Journal of Business, 53(3):S13–S29. find
  23. Nils Reimers and Iryna Gurevych. 2019. Sentence-BERT: Sentence embeddings using siamese BERT-networks. arXiv preprint arXiv:1908.10084.
  24. Xinrui Ruan, Xinwei Ma, Yingfei Wang, Waverly Wei, and Jingshen Wang. 2025. Can language models boost the power of randomized experiments without statistical bias? arXiv preprint arXiv:2510.05545. — [arXiv:2510.05545](https://arxiv.org/abs/2510.05545) find
  25. Matteo Tranchero, Cecil-Francis Brenninkmeijer, Arul Murugan, and Abhishek Nagaraj. 2024. Theorizing with large language models. Working Paper 33033, National Bureau of Economic Research. find
  26. Mengxin Wang, Dennis J Zhang, and Heng Zhang. 2026. Large language models for market research: A data-augmentation approach. Marketing Science. find
  27. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 2025. Qwen3 technical report. arXiv preprint arXiv:2505.09388.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zhang-2026-agentic-economic-modeling/