Heterogeneous AI News and Magnificent Seven Hubs

Event-Day Causal Network Evidence. This page collects the rendered replication notebooks for the empirical analysis, robustness checks, simulations, and placebo exercises accompanying the paper.

Milán Csaba Badics and Márton Espán

Overview

The paper studies how major AI-related news events affect short-run causal network structures among large technology stocks, with special attention to the role of the Magnificent Seven firms. The empirical analysis uses high-frequency intraday financial data and directed network methods based on one-minute log returns around selected event windows.

The replication workflow relies on the R2DAG package, which implements directed R²-based connectedness measures using contemporaneous causal identification. The pages below are rendered HTML versions of the replication notebooks, intended to make the full empirical pipeline transparent from raw intraday data preparation to event-window estimation, statistical testing, LiNGAM accuracy checks, and placebo comparisons.

Replication pages

The links below assume that this page is placed inside the research/ folder and that the rendered notebook HTML files are stored in research/heterogeneous_ai_news_replication/.

Page	Role in the replication workflow
01_read_data.html	Reads raw one-minute stock price files, constructs synchronized intraday log returns, and saves the cleaned dataset used in the empirical analysis.
02_an_illustrative_example.html	Provides a deterministic illustrative example showing how the proposed directed R² network differs from alternative network representations.
03_network_of_each_event.html	Computes and plots event-specific directed networks for the selected AI news events.
04_statistical_tests.html	Performs statistical tests comparing network structures and group-level effects across event windows.
05_accuracy_of_the_lingam.html	Evaluates the accuracy of the LiNGAM-based contemporaneous causal identification procedure in Monte Carlo experiments.
06_placebo_distribution.html	Constructs placebo distributions and compares event-day network statistics to non-event benchmark distributions.

Data requirements

The raw high-frequency data are not included in the replication folder. To run the notebooks, users need to provide the underlying one-minute price data locally. The input folder should contain one .txt file for each ticker, named only by the ticker symbol.

AAPL.txt MSFT.txt NVDA.txt AMZN.txt META.txt GOOG.txt TSLA.txt SP500.txt

The file SP500.txt is required because the empirical workflow uses the market index for demarketing. The first notebook creates the cleaned one-minute log return dataset and saves it as full_dataset.csv, which is then used by the later notebooks.

Installation

The notebooks require R and several standard R packages. The main package used for the directed R² network calculations is R2DAG, which can be installed from GitHub:

devtools::install_github(
  "espanm/R2DAG",
  dependencies = FALSE,
  upgrade = "never"
)

Depending on the notebook, additional packages such as readr, dplyr, purrr, zoo, xts, openxlsx, igraph, ggplot2, and devtools may also be required.

Output and citation

The notebooks produce cleaned datasets, network tables, statistical summaries, and figures used in the paper and appendix. Some figures in the final paper may differ slightly from the notebook output because of formatting adjustments for publication.

If you use these replication files, please cite the accompanying paper:

Badics, Milán Csaba and Espán, Márton. "Heterogeneous AI News and Magnificent Seven Hubs: Event-Day Causal Network Evidence."

These files are intended for academic replication purposes. Please contact the authors before reusing the code or results in other projects.