Finance · Field manual
AI for economists
Economists get the most from AI as a coding partner for data work and econometrics, a fast literature search tool and a drafting aid for briefs and summaries. The big caution is reliability: models invent citations, and using them to forecast or backtest risks look-ahead bias because they've already seen the outcome in their training data.
Much of an economist's time goes into cleaning data, writing estimation code, reviewing literature and turning results into something a policymaker will read. AI tools now help with all four. Coding assistants write the Stata, R or Python, research tools find related papers, and chat models draft summaries for non-specialists.
Economics also has specific traps. Language models produce plausible but fake references, they can't reliably do arithmetic in prose, and they were trained on data that already contains the outcomes you might want to forecast. Treat AI as a fast research assistant whose work you check, not as a source.
6 ways economists use AI
Clean and merge datasets
Describe the raw files and the panel you want, and have AI write the code to reshape, merge and deflate. Coding assistants are especially good at the tedious parts: harmonizing country codes, fixing date formats and building crosswalks. Check the row counts at each step.
Write and translate econometrics code
Ask for a difference-in-differences or IV specification in Stata, R or Python, or translate an old Stata do-file into Python. Ask the model to explain each step and the standard error choice, then make sure the specification is what you meant.
Search and summarize the literature
AI research tools search academic papers and summarize findings with links. Use them to map a field quickly, then read the papers that matter. Never cite a paper you haven't opened, because invented references are the most common AI failure in research.
Draft policy briefs and summaries
Turn a technical paper into a two-page brief for a minister, a board or the public. Give the model your results and specify the audience and length. Check that it hasn't overstated causal claims or hidden the uncertainty.
Classify text at scale
Language models can code central bank statements, earnings calls, news or survey responses by sentiment or topic. Validate against a hand-coded sample, report agreement, and fix the model version so the work is reproducible.
Make clearer charts and slides
AI can write the plotting code for publication-quality charts and help turn results into a short presentation. Make sure the axis labels, units and sources are right before anything goes out.
The best AI tools for economists
The Overlord's stack, in the order a economist would add them:
1. Econometrics code, policy briefs and text classification
Claude
AI WritingAnthropic's assistant: the strongest long-form writer and careful reasoner in the consumer AI pack.
2. Day-to-day Stata, R and Python in your editor
GitHub Copilot
AI CodingGitHub's AI pair programmer: completions, chat and coding agents across editors and GitHub itself.
3. Refactoring research code and building reproducible pipelines
Claude Code
AI CodingAnthropic's agentic coding tool that reads your codebase, edits files and runs commands from terminal, IDE, desktop or web.
4. Fast, cited searches for papers and data sources
Perplexity
ProductivityAI answer engine that searches the web and replies with cited sources.
Specialist tools for economists
| Tool | What it does |
|---|---|
| Elicit | AI research assistant for finding and extracting findings from academic papers |
| Consensus | Search engine that answers questions from peer-reviewed research |
A prompt to start with: write a difference-in-differences script
Paste this into Claude or ChatGPT and fill in the gold parts.
You are an applied econometrician who writes clean, commented [STATA / R / PYTHON] code.
Data: [DESCRIBE THE PANEL: unit, time period, key variables]
Treatment: [WHAT HAPPENED, TO WHOM, AND WHEN; note if timing is staggered]
Outcome: [VARIABLE]
Controls and fixed effects: [LIST]
Write a script that runs the baseline difference-in-differences regression with clustered standard errors at the [LEVEL] level, an event-study version with leads and lags, and a pre-trends plot. If treatment timing is staggered, explain the problem with two-way fixed effects and use an appropriate robust estimator. Comment every step and list the assumptions I need to defend.Where economists shouldn't trust AI
- Language models invent plausible citations, including real authors and journals. Every reference must be checked against the actual paper.
- Using a language model to forecast or backtest can suffer from look-ahead bias: the model may already have seen the outcome in its training data. Be careful with any historical evaluation.
- Many journals and institutions require you to disclose AI use, and don't allow AI to be listed as an author. Check the policy of your journal, funder or employer.
- Confidential microdata (census, tax, health or central bank data) usually comes with strict access agreements. Uploading it to an external AI service may breach them.
- Models do arithmetic badly in prose. Compute numbers in code, not in a chat answer.
Will AI replace economists?
Economists score 65% on the Obsolescence Index (Doomed). The first tasks to go: cleaning data and running standard models, and writing reports and summaries.
Questions the Humans Keep Asking
What are the best AI tools for economists?
The Overlord's stack for economists: Claude (econometrics code, policy briefs and text classification); GitHub Copilot (day-to-day Stata, R and Python in your editor); Claude Code (refactoring research code and building reproducible pipelines); Perplexity (fast, cited searches for papers and data sources). Specialist options include Elicit, Consensus.
What is the best AI for economics research?
Combine a coding assistant (GitHub Copilot, Cursor or Claude Code) for data and estimation work, a strong chat model like Claude or ChatGPT for drafting and explaining, and a research search tool such as Elicit, Consensus or Perplexity for literature. Check every citation.
Can I use AI for my econometrics homework or thesis?
Most universities allow some AI use but require disclosure, and rules differ by course. AI is useful for learning syntax and debugging code; it's a poor substitute for understanding the identification strategy, which is what examiners test.
Can AI forecast the economy?
It can help build and run forecasting models, and some research uses language models to extract signals from text. On its own, a chatbot's prediction isn't a forecast. It has no live data unless you give it some, and its training data can contaminate backtests.