Find papers for 2025 January-March on using LLM for macroeconomic modelling
- User
Find papers for 2025 January-March on using LLM for macroeconomic modelling
Assistant
Based on the search results, there are several relevant papers from early 2025 that discuss using large language models (LLMs) for macroeconomic modeling:
- "Macroeconomic Forecasting with Large Language Models" by Andrea Carriero, Davide Pettenuzzo, and Shubhranshu Shekhar (March 2025)[1][3]. This paper compares the accuracy of LLMs against traditional macro time series forecasting approaches using the FRED-MD database. Key findings include:
- LLMs are competitive with simple AR benchmarks and broadly comparable to Bayesian VARs and Factor Models.
- LLMs perform relatively better in the post-COVID era.
- LLMs can generate very good forecasts for some series but are less reliable overall compared to econometric models.
- "Large Language Models: An Applied Econometric Framework" by Ludwig et al. (January 2025)[2][5]. This paper develops an econometric framework for using LLMs in empirical research, distinguishing between prediction and estimation tasks. Key points:
- Using LLMs for prediction is valid if there's no "leakage" between training data and the researcher's sample.
- For estimation tasks, researchers must collect validation data to assess LLM errors.
- The paper demonstrates applications in finance and political economy.
- "Leveraging large language models for large-scale information retrieval in economics" (February 2025)[7]. This study introduces a methodology using LLMs to process and synthesize tens of thousands of economics papers, focusing on:
- Constructing a "causal knowledge graph" from economics literature.
- Identifying causal relationships between economic concepts supported by rigorous inference methods.
- Extending the approach to other domains like patents, financial filings, and policy reports.
These papers highlight the growing interest in applying LLMs to macroeconomic modeling and research, while also noting the challenges and limitations of these approaches[6][8]. Researchers are exploring how LLMs can complement traditional econometric methods, improve information retrieval and synthesis, and potentially enhance forecasting accuracy in certain contexts.
Other models, same or similar prompt
Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.