BovDB: Stock Quotes Dataset for Machine Learning on B3 Equities

---

Andrey Vinicius Souza, Fabian Corrêa Cardoso, Juan Malska, Paulo Ramiro, Giancarlo Lucca, Eduardo N. Borges, Viviane de Mattos, Rafael Berri

• Grupo de Pesquisa em Informática na Gestão da Informação e Finanças (GINFO)
• Programa de Pós-Graduação em Modelagem Computacional (PPGMC) — Universidade Federal do Rio Grande (FURG)
• Centro de Ciências Computacionais (C3) — Universidade Federal do Rio Grande (FURG)
• Instituto de Matemática, Estatística e Física (IMEF) — Universidade Federal do Rio Grande (FURG)
• Contato: {andreyvinicius, fabiancorrea, giancarlo.lucca, eduardoborges, vivianemattos, rafaelberri}@furg.br

---

Overview

BovDB is an open-access, relational dataset and Python toolkit of historical stock quotes and high-frequency intraday data covering all companies listed on the Brazilian Stock Exchange (B3), designed for machine learning benchmarks, econometric modeling, time-series forecasting, and algorithmic trading research.

Key Capabilities

• Complete Equity Coverage: Covers all tickers traded on B3 across common (ON), preferred (PN), units (UNT), and Brazilian Depositary Receipts (BDRs).
• Corporate Action Adjustments: Accounts for stock splits, reverse splits (inplits), and bonuses via dedicated factor multipliers.
• Multi-Resolution Data: Includes daily historical OHLCV series (1995–Present) along with high-frequency 5-minute intraday bars (price5).
• Safe Read-Only Python SDK: Modular bovdb Python package opening SQLite connections in strict read-only mode by default to prevent database corruption.
• Zero-Configuration Relational Storage: Stored in a high-performance, single-file SQLite database with straightforward Python integration.

---

Dataset Versions & Scope

Version	Status / Folder	Period	Resolution	Features & Archive Formats
BovDB v1.0	BovdbV1/	1995 – 2020	Daily	All B3 equities, daily OHLCV, adjustment factors. Formats: CSV, JSON, SQLite (.zip). Published at SBBD 2021.
BovDB v2.0	BovdbV2/ — Harvard Dataverse	1995 – 2024	Daily + Intraday	Daily historical quotes + 5-min intraday price5 (Jan–Jun 2024). Official published version (DOI: 10.7910/DVN/TMB4IG).
BovDB v3.0	🚧 In Development — BovdbV3/	1995 – Present	Daily + Intraday	Modular bovdb Python package, CI/CD, complete schema dictionary, multi-mirror links.

---

Quick Start

1. Clone & Environment Setup

git clone https://github.com/Ginfofinance/BovDBrepository.git
cd BovDBrepository

# Create and activate virtual environment
python -m venv .venv

# On Windows:
.venv\Scripts\activate
# On Linux / macOS:
source .venv/bin/activate

# Install base package with visualization tools:
pip install -e ".[viz,analysis,dev]"

2. Download the Database

1. Open BovdbV3/link_database.txt (or BovdbV2/link_database.txt) to access download links.
2. Download and extract Database_define.rar (or .zip).
3. Place the extracted SQLite database file inside the DataBase/ folder:

   DataBase/DataBase.db
   

(Optional) You can also set a custom path using the BOVDB_PATH environment variable.

3. Run Example Scripts

• Daily Candlestick Chart (with and without factor adjustment):

  python Codes/daily_candlesticks_chart.py

• Annual Mean and Standard Deviation Graph:

  python Codes/annual_mean_and_standard_deviation_graph.py

• Interactive 5-Minute Intraday Candlestick Chart:

  python Codes/intraday_5min_candlestick.py

• Top & Bottom Extrema Detection (Jupyter Notebook):

  jupyter notebook Codes/low_and_bottom.ipynb

---

Python API Example

You can query the dataset directly in your research workflows using the bovdb package:

import bovdb

# Automatically connects to DataBase/DataBase.db in safe read-only mode
conn = bovdb.get_connection()

# 1. Fetch split-adjusted daily prices for Petrobras (PETR4)
df_daily = bovdb.get_daily_prices(conn, ticker="PETR4", start_date="2023-01-01", end_date="2024-06-30", adjusted=True)
bovdb.plot_candlestick(df_daily, ticker_name="PETR4 (Adjusted)")

# 2. Fetch 5-minute intraday quotes
df_intra = bovdb.get_intraday_5min(conn, ticker="PETR4", start_date="2024-06-26", end_date="2024-06-26")
bovdb.plot_intraday_5min(df_intra, day="2024-06-26", ticker_name="PETR4")

conn.close()

---

Documentation

• Architecture & Design — Relational ER model and table structures.
• Data Dictionary — Detailed table schemas (company, ticker, price, price5).
• Data Provenance & Lineage — Origin feeds (B3/CVM), data harmonization, and quality gates.
• Data Reproducibility Standards — Protocols for machine learning benchmarks and random seed fixing.
• Repository Governance — Branch protection rules, PR reviews, and semantic commit standards.
• Contributing Guide — How to propose enhancements and submit pull requests.
• Code of Conduct — Community standards and academic integrity.
• Security Policy — Guidelines for reporting vulnerabilities.

---

Citation

If you use BovDB in scientific publications or academic projects, please cite:

BovDB V2 (Harvard Dataverse — Official Published Version)

@data{DVN/TMB4IG_2024,
  author    = {Souza, Andrey S. and Lucca, Giancarlo and Borges, Eduardo N. and Cardoso, Fabian C. and Dalmazo, Bruno L. and Berri, Rafael},
  publisher = {Harvard Dataverse},
  title     = {{Dataset for Intraday Analysis of B3 stock prices}},
  year      = {2024},
  version   = {V1},
  doi       = {10.7910/DVN/TMB4IG},
  url       = {https://doi.org/10.7910/DVN/TMB4IG}
}

BovDB V1 (SBBD 2021)

@inproceedings{cardoso2021bovdb,
  title     = {Bovdb: The data set of stock quotes for machine learning on all companies from B3 between 1995 and 2020},
  author    = {Cardoso, Fabian and Malska, Juan and Ramiro, Paulo and Lucca, Giancarlo and Borges, Eduardo N and Mattos, Viviane de and Berri, Rafael},
  booktitle = {Proceedings of the Brazilian Symposium on Databases - Dataset Showcase},
  year      = {2021},
  organization = {SBC}
}

---

License

This dataset and accompanying software are licensed under the Creative Commons Attribution-NonCommercial 4.0 International License (CC BY-NC 4.0).