Finance & Econometrics Applications¶

AI applied to finance research, policy evaluation, and econometric analysis.

Summaries¶

Day 1: Foundations, Applications, and Agentic Tools — LLM as the newest economic instrument; provider vs. researcher choices; five levels of agentic research; "use agents for labor, use economics for judgment"
Day 2: Natural Language Understanding and Measurement — Fedspeak benchmark (500 FOMC sentences, 3 RAs), GPT-4 vs. BERT/dictionaries, "digital Romers," quantized local models on earnings calls, LLM→embedding→K-means topic pipeline
Day 3: Forecasting, Prediction, and Survey Augmentation — Synthetic SPF forecasters beat humans at longer horizons, but only with human input in the prompt; the prediction-vs-estimation "econometric contract"; leakage tests and debiased estimation
Day 4: Synthetic Agents, Simulations, and Counterfactuals — Homo silicus; a 200-cell bank-run simulation benchmarked against 1,158 humans across 8 LLMs; cell-level bias correction; agent architectures; LLM survey augmentation
Day 5: LLMs as Economic Subjects — Revealed preference applied to models: race vectors in loan decisions, "aligned models act too fair," and 13 models tracking biased human self-reports on sensitive questions

Project APE — Fully automated economics research pipeline with multi-model tournament
Automated Research in Finance — Lopez-Lira's pipeline seeking collaborators for journal submissions
ZeroPaper Gallery — Public catalog of 22 finance papers produced end-to-end by Lopez-Lira's ZeroPaper pipeline
ZeroPaper Template — Public release of Lopez-Lira's autonomous pipeline as a clonable template (Claude Code / Codex / Gemini)
ZeroPaper Paper — Lopez-Lira's design discipline: 10 premises + 6 principles behind a long-running autonomous research pipeline
AI-Powered (Finance) Scholarship — Novy-Marx & Velikov: 30K signals → 95 survivors → 380 LLM-generated papers
Applications (Korinek) — Comprehensive guide to 7 LLM application domains for economists
ai-asset-pricing (Dickerson) — Multi-agent empirical asset pricing repo with WRDS, PyBondLab, LaTeX
ralph-wiggum-asset-pricing (Chen) — Open-source Ralph loop framework for generating asset pricing papers
Claude WRDS Toolkit (Liu/Orłowski) — Subagents + skills for autonomous CRSP/Compustat/IBES/OptionMetrics/TAQ queries

Thread: Arin Dube — LLMs haven't raised NBER output; ideas, not RA labor, are the constraint
Thread: Jesús Fernández-Villaverde — 12 arguments for traditional higher ed survival in AI era
Thread: Alex Olegi — Licklider's "Man-Computer Symbiosis" as frame for AI-research moment
Thread: Chris Blattman — Conversion from Claude Code skeptic to power user

What Will Be Scarce? — Imas: structural change toward relational sector with mimetic preferences