AI Workflows and Publishing in Finance

Download Slides (PDF)

Three-block lecture covering AI-assisted research workflows, the publishing process in finance, and social-media signal construction.

Demo 1 — Social Signal: Stock-Level Analysis

Starting from 821,534 firm-day observations from The Social Signal (Cookson, Lu, Mullins & Niessner, JFE 2024), a single Claude Code session produces a publication-style stock-level analysis in under 15 minutes. The researcher supplied the economic questions; Claude supplied the mechanics.

demo_script.py signal_index.do Paper (PDF) Data (.dta, 11 MB) Data on Mendeley ↗

Sentiment and attention time series — Fig 1 — Sentiment & attention time series

Return predictability — sentiment — Fig 3 — Return predictability (sentiment)

Demo 2 — Market-Level Analysis

Four plain-English prompts build a complete market-level analysis from the same data: time-series properties, predictive regressions, a leads-and-lags event study, an FOMC event study, and a cap-weighted robustness check. See the demo guide for the full prompt sequence and methodology notes.

Demo Guide market_analysis.py market_analysis_capweighted.py

Leads and lags cumulative return — Fig 3 — Leads & lags cumulative return

AI in Econ Wiki — curated by Mihail Velikov (Penn State Smeal): velikov-mihail.github.io/ai-econ-wiki

AI Workflows and Publishing in Finance

Slides

In-Class Demos

Demo 1 — Social Signal: Stock-Level Analysis

Demo 2 — Market-Level Analysis

Resources