Skip to content

Narrative-Digital-Finance-Block-3

Narratives for Structural Breaks in Financial Markets - SNSF Grant IZCOZ0_213370

View on GitHub

Information

Property Value
Language Python
Stars 0
Forks 0
Watchers 0
Open Issues 0
License No License
Created 2026-01-05
Last Updated 2026-02-19
Last Push 2026-01-05
Contributors 1
Default Branch main
Visibility private

Datasets

This repository includes 5 dataset(s):

Dataset Format Size

| computed_values.json | .json | 2.39 KB |

| data | | 0.0 KB |

| init.py | .py | 0.65 KB |

| generators | | 0.0 KB |

| loaders | | 0.0 KB |

Reproducibility

This repository includes reproducibility tools:

  • Dockerfile for containerization

  • Makefile for automation

Status

  • Issues: Enabled
  • Wiki: Enabled
  • Pages: Disabled

README

Narratives for Structural Breaks in Financial Markets

Reproducibility License: MIT

Prediction, Detection, and Persuasion

This repository contains the reproducible code and manuscript for Block 3 (WP3) of the SNSF-funded Narrative Digital Finance project (Grant IZCOZ0_213370).

Target Journal: Journal of Financial Economics

Research Questions

RQ Question Method
RQ1 Can market narratives predict financial market bubbles? Granger causality, VAR, forecasting
RQ2 Can market narratives detect financial market bubbles? TOPol polarity fields, regime classification
RQ3 Can narratives sway investment opinions? 2x2 experimental design, causal embeddings

Quick Start

# Clone repository
git clone https://github.com/Digital-AI-Finance/Narrative-Digital-Finance-Block-3.git
cd Narrative-Digital-Finance-Block-3

# Install dependencies
poetry install

# Run full pipeline (data -> analysis -> manuscript)
make all

# Or run individual steps
make data       # Fetch/generate data
make model      # Train models
make analysis   # Run analyses
make figures    # Generate figures
make manuscript # Compile LaTeX

Repository Structure

src/narrative_breaks/     # Main Python package
    data/                 # Data loaders and generators
    modeling/             # Embeddings, TOPol, causal inference
    detection/            # Structural breaks, bubble tests
    analysis/             # Granger causality, VAR
scripts/                  # Pipeline scripts (01-10)
data/                     # Raw, processed, synthetic data
figures/                  # Chart folders (one per figure)
manuscript/               # LaTeX paper
tests/                    # Unit, integration, validation tests

Key Methodologies

TOPol Framework (Taibi et al. 2025)

Transformer-based polarity vector fields for detecting narrative regime shifts: 

v = centroid(post) - centroid(pre)

Causally Sufficient Embeddings (Veitch et al. 2020)

Text embeddings that preserve causal adjustment information for unbiased treatment effect estimation.

2x2 Experimental Design

  • Factor A: Narrative Presence (present vs. absent)
  • Factor B: Emotionality (high vs. low)

Data

All analyses use public or synthetic data only for full reproducibility:

Source Data Type
FRED VIX, S&P 500, macro indicators
SEC EDGAR 10-K/10-Q MD&A sections
BIS Central bank speeches
Synthetic Regime-switching corpus, GARCH returns

Citation

@article{osterrieder2026narratives,
  title={Narratives for Structural Breaks in Financial Markets: Prediction, Detection, and Persuasion},
  author={Osterrieder, Joerg and Taibi, Gabin},
  journal={Working Paper},
  year={2026},
  note={SNSF Grant IZCOZ0\_213370}
}

Funding

  • Swiss National Science Foundation (SNSF) Grant IZCOZ0_213370
  • EU Horizon Europe MSCA Grant 101119635

License

MIT License - see LICENSE for details.

Contact

Prof. Dr. Joerg Osterrieder Bern University of Applied Sciences / University of Twente joerg.osterrieder@bfh.ch