Data Management Plan (DMP)

1 Data collection and documentation

1.1 What data will you collect, observe, generate or reuse?

We use publicly available data from financial markets (accessed via original source URLs) and commercial databases (accessed via provider APIs). Public data is not re-stored by us. Commercial data remains on provider infrastructure per license agreements. Project-generated code, datasets, and research outputs are archived on Zenodo with DOIs.

Commercial Data Sources (Licensed):

Public Data Sources:

Generated Research Outputs:

1.2 How will the data be collected, observed or generated?

The quality of the collected data is checked during the first working package and is an integral part of the research. We apply statistical methods to address data shortcomings. Various documents describe the dataset, its quality, and the methods used to check its consistency.

Data Collection Methods:

• API Access: FRED API (Python fredapi), BIS API for central bank speeches
• Commercial Data: RavenPack SQL access, Deutsche Borse secure file transfer
• NLP Processing: HuggingFace transformers, BERT/FinBERT embeddings, BERTopic modeling, Named Entity Recognition (NER) for institution and region tagging
• LLM-based Analysis: OpenAI gpt-4o-mini for daily narrative sentiment scoring (Evergreen narratives)

Quality Assurance:

• Data checked for consistency and completeness using statistical methods
• Change point detection to identify structural economic shifts

Versioning: Code and databases are versioned with Git. Data releases receive DOIs via Zenodo integration.

1.3 What documentation and metadata will you provide with the data?

The information on the data as well as data sources and survey processes are documented in detail.

Metadata Standards:

• Zenodo metadata (DataCite schema) for all archived outputs
• Standard README documentation with variable descriptions and usage instructions
• Data dictionaries with variable names, types, and descriptions

Documentation Provided:

• README.md files in all repositories with variable descriptions
• Inline code comments where helpful
• Dependency files (requirements.txt) for reproducibility
• Jupyter notebooks with step-by-step methodology
• CC-BY 4.0 licensing for all project outputs

2 Ethics, legal and security issues

2.1 How will ethical issues be addressed and handled?

No personal data or other sensitive data is used in the project. The conditions have been discussed with the data providers. No special security standards are required by the data providers for this data.

Ethical Considerations:

• All financial data is aggregated market data with no individual identification
• Central bank speeches are official public communications
• Research outputs are intended for academic purposes

NLP Bias Assessment Framework:

• Sentiment distribution analysis by sector and geographic coverage
• Temporal drift monitoring for NLP model outputs
• Limitations section included in all publications
• English-language dominance acknowledged in central bank speech analysis

Data Provider Agreements:

Ethics Approval: Not required (non-human-subjects research per Swiss SNSF guidelines)

2.2 How will data access and security be managed?

We use private cloud solutions. No sensitive data or personal data is collected in the project.

Access Control:

Security Measures: Commercial data remains on provider infrastructure per license agreements. Relevant project code is public on GitHub and Zenodo.

2.3 How will you handle copyright and Intellectual Property Rights issues?

The project is based on data that are largely publicly available. The raw data records may not be published without restriction.

Copyright Framework:

Publication Rights:

• Yes Aggregated statistics and derived features
• Yes Code, models, and methodology
• Yes Project-generated datasets (Zenodo)
• No Raw commercial data redistribution

3 Data storage and preservation

3.1 How will your data be stored and backed-up during the research?

The storage capacities are large; the amount of data remains limited in the project. Relevant data is managed in the form of databases on GitHub and Zenodo and private cloud solutions. Backups and versions of the data are continuously created.

Primary Storage:

3.2 What is your data preservation plan?

Most relevant data is stored on Zenodo for long-term preservation. There is no obligation to destroy the data.

10-Year Retention Strategy:

• All project outputs archived on Zenodo with persistent DOIs
• Zenodo hosted at CERN Data Centre with guaranteed long-term preservation
• Public data remains at original sources (FRED, BIS)
• Commercial data is provider responsibility (not archived by us)

DOI Versioning Policy:

• Zenodo provides Concept DOI (resolves to latest version) and Version DOI (resolves to specific snapshot)
• New version created for any data change; README updated for documentation changes
• CHANGELOG.md maintained in each Zenodo deposit
• Citation format included in all README files

4 Data sharing and reuse

4.1 How and where will the data be shared?

The relevant code developed during the project together with all necessary accompanying documentation will be stored on Zenodo. Zenodo offers safe storage for all data and research outputs in CERN's Data Centre.

Published Research Outputs:

Repository URLs:

• Website: digital-ai-finance.github.io/Narrative-Digital-Finance/
• Code and Datasets: zenodo.org

Sharing by Data Type:

4.2 Are there any necessary limitations to protect sensitive data?

We do not use sensitive data in the project. The data come from conventional data providers and are originally collected from public sources.

Commercial Data Post-Project Access:

• Commercial data agreements are project-duration specific
• Analyses using commercial data are reproducible via methodology documentation
• Derived features (shareable) are archived with sufficient detail

4.3 All digital repositories I will choose are conform to the FAIR Data Principles.

Yes

FAIR Data Principles Implementation:

4.4 I will choose digital repositories maintained by a non-profit organisation.

Yes

Repository Operators:

• Zenodo: Operated by CERN (European Organization for Nuclear Research) - non-profit
• arXiv: Operated by Cornell University - non-profit
• GitHub: Operated by Microsoft - commercial, but free for public repositories