Econometrics-Agent

Report Abuse

Basic Information

This repository contains the official implementation of the research project "Can AI Master Econometrics?" and provides a ready-to-run, LLM-driven AI agent specialized for automating complex econometric analysis. The agent is built on the MetaGPT framework and uses a zero-shot learning approach to integrate econometric domain knowledge without costly model fine-tuning. The codebase includes the backend agent, a web frontend, configuration files, evaluation datasets from coursework and published papers, a demo video, and scripts to install dependencies and start the application. The README documents installation steps, environment configuration, and how to point the agent to an LLM provider via config/config2.yaml and .env. The project aims to provide a reproducible research artifact and a practical tool for running expert-level econometric tasks locally with configured LLM access.

Links

App Details

Features
The agent provides strategic task planning and decomposition, automated Python code generation and execution for econometric analyses, and an error-based reflection mechanism to evaluate and improve action outcomes. It supports iterative multi-round conversations so users can refine tasks and inherit task state across rounds. The project ships a domain-specific tool library covering common econometric methods such as IV-2SLS, difference-in-differences, regression discontinuity design, and propensity score methods. The system includes visualizations of workflows, generated code, and execution results and adds user quota management and enhanced file upload support in the frontend. Configuration is centralized in config/config2.yaml and the service runs via provided start scripts and a web UI on the documented local port.
Use Cases
The Econometrics AI Agent lowers the barrier to applying advanced econometric methods by automating planning, coding, execution, and iterative refinement, making these techniques accessible to students, researchers, and practitioners with limited coding expertise. It can accelerate empirical research workflows, improve reproducibility by bundling datasets, code, and prompts, and serve as an educational aid for teaching econometrics through hands-on examples. The domain-specialized toolset and zero-shot prompt design offer a cost-effective alternative to expensive LLM fine-tuning while delivering higher performance than generic agents on expert-level econometric tasks. The included demo, datasets, and web interface facilitate evaluation, experimentation, and adoption in research and applied settings.

Please fill the required fields*