🗃️ TroveLedger — Financial Time Series Dataset

A growing ledger of accumulated market history.

⚠️ Temporary Notice: Intraday Data Adjustments (January 2026)

What happened:
A discrepancy has been identified in the minute- and hourly-resolution data: these series are currently not fully adjusted for stock splits and dividends. Daily-resolution data remains correctly adjusted (as provided by the source).

Why this matters:
For accurate backtesting and model training – especially in intraday strategies – proper adjustments are essential to avoid distorted price histories.

What I'm doing:
I am actively working on a comprehensive fix to bring all resolutions to full adjustment consistency. This will ensure the highest possible data quality moving forward.

Short-term plan:
Until the fix is complete, no new indices or symbols will be added to avoid introducing further unadjusted data.

If the full correction takes longer than expected, an intermediate step may be applied: removing affected (unadjusted) intraday segments to keep the published dataset accurate, even if temporarily reduced in scope. (update January 13.: Fix is in place, there seems no need for this)

**Current Status (updated January 19, 2026):**

The adjustment logic for splits and dividends in minute- and hourly data is showing promising results in testing. > It detects and applies known splits correctly and handles most dividend events sensibly. However, the details are taking more time than expected.

Challenges & next steps:

• Some noise/false positives appear in dividend detection (adjustments applied where none occurred).
• Fine-tuning thresholds and filters is ongoing to reduce this noise while preserving real events.
• This requires additional experimentation and manual verification over the weekend.

Progress summary:

• Core fix works in principle → clear improvement visible on tested samples
• Full-dataset re-processing continues in background
• Still in validation phase – no re-publish until confident in the quality

Important reminder:
Until the fix is validated and re-published, do not use the current intraday data for strategies/backtests that depend on accurate split/dividend adjustments.

Ongoing operations:
Daily OHLCV accumulation for existing symbols continues normally (no gaps).
No new indices/symbols are being added during this phase.

Updates will be posted here and on X (@TroveLedger) as soon as next milestones are reached (threshold tuning done, validation samples passed, reduced/clean version live, full version live).

Thank you for your continued patience — this extra refinement will make the dataset noticeably more reliable for serious AI trading bot training and quant work.

🔔 **Latest Dataset Update**

Date: 2026-01-06
New addition: 🇧🇪 BEL 20 (Belgium)

Today, the dataset expands with the BEL 20 — Belgium’s primary equity index and the central barometer of its domestic capital market.
Composed of the 20 most liquid and heavily traded companies listed in Brussels, the BEL 20 reflects an economy shaped by finance, consumer goods, industry, and cross-border European integration.

📊 Market context:
Region: Europe — Belgium
Scope: Large-cap benchmark (BEL 20 constituents)
Sector exposure: Financials, Consumer Staples, Industrials, Utilities
Data coverage: Minute, hourly, and daily OHLC data

📈 What this means:
A compact, liquid market segment that adds depth to Western European coverage with consistent, high-resolution time series.

🔜 What’s next:
Continued expansion across international markets, preserving uniform structure and historical depth.

Click to expand

• Download symbol list (BEL20.txt)

• More information about previous additions can be found further below in The Growing Treasury

Recent Index Additions

Date	Index	Region	Symbols
2026-01-06	BEL 20	Belgium 🇧🇪	20
2026-01-05	DAX	Germany 🇩🇪	40
2026-01-02	ASX 200	Australia 🇦🇺	200
2025-12-30	OMX Stockholm 30	Sweden 🇸🇪	30
2025-12-29	TSX (S&P/TSX Composite)	Canada 🇨🇦	222
2025-12-24	SMI	🇨🇭 Switzerland	20
2025-12-23	NIFTY 50	🇮🇳 India	50
2025-12-22	FTSE 100	🇬🇧 United Kingdom	100
2025-12-19	S&P 500	🇺🇸 US	503
2025-12-18	Hang Seng Index	🇭🇰 Asia	82
2025-12-17	EURO STOXX 50	🇪🇺 Europe	50

📌 Overview

TroveLedger is a public financial time series dataset focused on long-term accumulation of high-quality intraday data.

The dataset provides OHLC and volume data at multiple time resolutions and is designed primarily for machine learning, quantitative research, and systematic trading experiments.

Unlike many freely available data sources, TroveLedger emphasizes continuity over time, especially for minute-level data.

Scale & Granularity

• Total: Over 40 million rows across all symbols and resolutions (growing rapidly)
• Per symbol: Varies significantly – from <1,000 rows (young stocks, daily) to >500,000 rows (established stocks, minute-resolution)
• Ideal for both focused single-symbol training and large-scale multi-market models

🔑 What makes TroveLedger different

High-resolution intraday data is difficult to obtain from free sources over extended periods.

Typical public data access (e.g. via yfinance) provides:

• Daily candles: often spanning decades
• Hourly candles: roughly one year into the past
• Minute candles: usually limited to the most recent 7 days

Repeatedly downloading rolling 7-day windows results in short, fragmented histories that are poorly suited for training models on intraday behavior.

TroveLedger takes a different approach:

• Minute-level data is accumulated continuously
• Time series are extended, not replaced
• Over time, this results in months of gap-free minute data per instrument

This accumulated depth forms a substantially more reliable foundation for intraday research and model training.

🧱 Data Integrity Philosophy

TroveLedger prioritizes continuity over frequency.
The primary goal is not to fetch data as often as possible, but to ensure that once a time series starts, it remains gap-free.

Minute-level data is accumulated incrementally over time, creating long, uninterrupted histories that are not obtainable from fresh API queries alone.

This makes the dataset particularly suitable for model training, backtesting, and regime analysis.

📦 Dataset Structure

The dataset is organized as follows:

• /data/{category}/{symbol}/{symbol}.{interval}.valid.parquet

Where:

• {category}: e.g., equities/us, indices/sp500, indices/eurostoxx50 (growing with new indices)
• {symbol}: Stock ticker (e.g., AAPL, BMW.DE)
• {interval}: One of days (daily), hours (hourly), or minutes (1-minute)

The .valid suffix indicates that these files have passed quality checks and are ready for use. Only these cleaned, validated files are included in the dataset – temporary or intermediate files from the pipeline are excluded.

Tip for users: The .valid part is intentionally kept as a flexible "state" marker. You can easily rename or copy files to add your own states (e.g., .train.parquet or .test.parquet) for train/validation/test splits in your ML workflows. This pattern makes it simple to organize experiments without changing the core data.

Data Instances

Here's an example row from a typical daily Parquet file (e.g., for AAPL.days.valid.parquet):

• time is a Unix timestamp (e.g., 1704067200 = January 1, 2024, 00:00 UTC).
• All prices are in the symbol's native currency (e.g., USD for US equities).

Dataset Creation

Curation Rationale

TroveLedger was created to provide a reliable, expanding source of historical OHLCV data for AI-driven trading research, addressing gaps in continuity and international coverage.

Source Data

All data is sourced from Yahoo Finance via the yfinance Python library. Index components are automatically extracted from Wikipedia pages using a custom API-based pipeline for sustainability.

Data Collection and Processing

• Symbols are selected from major indices (e.g., S&P 500, EURO STOXX 50) and equities.
• Data is fetched at daily, hourly, and 1-minute resolutions, validated for completeness, and stored in Parquet format for efficiency.
• Quality checks remove gaps or anomalies; only ".valid" files are included.
• Updates occur periodically to extend histories and add new indices based on community input.

Who are the source data producers?

Yahoo Finance (public market data). No personal data is included.

🔄 Update Philosophy

The primary objective is data continuity, not guaranteed daily updates.

In particular:

• Daily updates are not guaranteed
• Preventing gaps in accumulated minute data has priority
• Updates are performed on trading days whenever possible

Minute data is updated most frequently to ensure continuity.

Hourly and daily data are updated on a rotation basis to reduce unnecessary repeated downloads and to remain considerate of public data sources. These datasets are guaranteed to be no older than one week.

For most training scenarios, this is fully sufficient. When models are deployed in real-world environments, current market data is typically provided directly by the target trading platform.

📈 Scope & Growth

TroveLedger started with a curated universe of approximately 500 equities inherited from earlier Preliminary datasets.

Going forward:

• Entire indices are added step by step
• The covered universe will grow continuously
• Expansion is performed incrementally to ensure data integrity and operational stability

This gradual approach allows issues to be detected early and handled without disrupting existing data.

🎯 Intended Uses

• Primary Use: Training and evaluating machine learning models for trading strategies and autonomous AI bots.
• Other Uses: Time series analysis, financial research, educational projects, and community-driven extensions.

TroveLedger is suitable for:

• machine learning on financial time series
• intraday and swing trading research
• feature engineering on OHLC data
• backtesting strategies requiring dense intraday history
• exploratory quantitative analysis

⚠️ Limitations & Notes on Data Sources

• Data Freshness: Data is typically a few days old, not real-time.
• Coverage: Not all symbols may have complete historical data, especially for minute-resolution or newly added indices.
• Growth Phase: The dataset is actively expanding; check for updates on new indices and symbols.
• Not financial advice: This dataset is for research and educational purposes only. Past performance is no guarantee of future results.

Data is derived from publicly accessible market data sources (e.g. via yfinance).

While care is taken to ensure consistency and continuity, this dataset is provided as-is and without guarantees regarding completeness or correctness.

Users are responsible for verifying suitability for their specific use cases and for complying with the terms of the original data providers.

📜 License & Usage

This dataset is provided solely for non-commercial research and educational purposes.

The data is retrieved from public sources via the yfinance library (Yahoo Finance). All rights remain with the original data providers.

Redistribution of this dataset is not permitted without explicit permission from the original sources.

See the LICENSE file for full details.

NO WARRANTY IS PROVIDED. Use at your own risk.

💬 Feedback, Suggestions & Community Support

TroveLedger is a growing, community-driven project providing high-quality OHLCV data for training AI models on financial markets and trading strategies. Your input makes it better!

• What are you building? I'd love to hear how you're using TroveLedger! Share your projects, trading bot ideas, ML models, or research directions – it motivates me to keep expanding and might inspire others.
• Desired indices: Which major indices are you waiting for most? I'll prioritize based on demand and feasibility.
• Helping expand indices: The pipeline uses the Wikipedia API to automatically extract components. It works best with a structured table containing both company names and clean, yfinance-compatible ticker symbols.
  • Simply share the Wikipedia page URL (any language) for your desired index.
  • If the table needs tweaks (e.g., missing or unclear ticker column, prefixes in symbols), improving it on Wikipedia is the most sustainable way – the global community then keeps it updated long-term!
  • Once ready, post the link here, and I'll integrate it quickly.

Interested in a deeper dive into the exact table format and config options my pipeline supports (with examples like zero-padding, suffixes, or language overrides)? Let me know – if there's demand, I'll create a dedicated guide soon!

Join the discussion in Hugging Face Discussions.

🏛️ The Growing Treasury

text Copy

Watch TroveLedger expand across global markets – a visual chronicle of added indices:  

🇧🇪 BEL 20 (January 6, 2026) – Measured Wealth in a Small Vault

Behind heavy stone walls and careful accounting, the BEL 20 represents a market where scale is limited but liquidity is deliberate. Belgium’s leading financial, industrial, and consumer firms form a tightly curated index, reflecting a capital market defined by restraint, stability, and European connectivity.

This addition strengthens TroveLedger’s continental European archive, capturing a compact exchange whose value lies not in breadth, but in precision and consistency across timeframes.

• Download symbol list (BEL20.txt)

🇩🇪 DAX (January 5, 2026) – Engineered Capital at Industrial Scale

At the intersection of precision engineering and global trade, the DAX captures Germany’s most influential publicly listed corporations. Industrial manufacturers, automotive leaders, chemical groups, and financial institutions dominate the index, forming a market shaped by export depth and operational discipline.

This entry anchors TroveLedger’s core European coverage, adding a market where long-cycle industry, efficiency, and global exposure define capital behavior across timeframes.

• Download symbol list (DAX.txt)

🇦🇺 ASX 200 (January 2, 2026) – Weighed by Earth and Capital

Across vast distances and resource-rich ground, the ASX 200 captures an equity market anchored in tangible assets and institutional capital. Mining conglomerates, major banks, energy firms, and healthcare leaders dominate the index, forming a market profile distinct from technology-heavy regions.

This addition extends TroveLedger’s reach into Oceania, preserving a market where commodities, yield, and global demand cycles leave clear historical traces across intraday and long-term data.

• Download symbol list (ASX200.txt)

🇸🇪 OMX (December 30, 2025) – Order in the Nordic Ledger

In the measured calm of Northern Europe, the Stockholm Stock Exchange (OMX) records value through discipline, transparency, and long-term orientation. Industrial groups, financial institutions, and globally oriented consumer firms dominate the OMX Stockholm 30, forming a compact yet internationally relevant market profile.

This entry adds a distinctly Nordic balance to TroveLedger — one shaped by export strength, institutional stability, and methodical capital allocation across intraday and long-horizon views.

• Download symbol list (OMX.txt)

🇨🇦 TSX (December 29, 2025) – Beneath the Surface of Canadian Capital

Deep underground, where resources are extracted and value is carefully recorded, the Toronto Stock Exchange (TSX) reflects the structural foundations of the Canadian economy. Banks, miners, energy producers, and industrial firms form the backbone of the S&P/TSX Composite, making it a distinctive counterweight to more tech-heavy global indices.

This entry extends TroveLedger’s North American coverage beyond the United States, adding a market shaped by commodities, capital discipline, and long-cycle industries — all captured across consistent intraday and long-horizon timeframes.

• Download symbol list (TSX.txt)

🇨🇭 SMI (December 24, 2025) – Alpine quality meets market stability

The Swiss Market Index (SMI) has been added to TroveLedger, bringing the premier blue-chip index of Switzerland into our global dataset.
Representing 20 of the largest and most liquid companies listed on the SIX Swiss Exchange — including giants like Nestlé, Roche, and Novartis — the SMI offers a unique exposure to one of the world’s most stable and innovation-driven economies.

The SMI reflects Switzerland’s enduring role as a benchmark for quality, resilience, and long-term value.

• Download symbol list (SMI.txt)

🇮🇳 NIFTY 50 (December 23, 2025) – India takes center stage

The NIFTY 50 Index from India has been incorporated into TroveLedger, enriching the dataset with one of South Asia’s most referenced equity benchmarks. It represents 50 of the largest and most liquid Indian stocks listed on the National Stock Exchange.

• Download symbol list (NIFTY.txt)

🇬🇧 FTSE 100 (December 22, 2025) – Britain weathers the storm

The FTSE 100 represents 100 of the most capitalized and liquid firms on the London Stock Exchange, spanning finance, energy, consumer goods, healthcare, and industrial sectors.
As the UK is no longer part of the European Union, this addition extends TroveLedger’s European coverage beyond the Eurozone without overlap with previously added indices.

• Download symbol list (FTSE.txt)

🇺🇸 S&P 500 (December 19, 2025) – America answers the call

The complete S&P 500 Index (503 constituents) has been fully integrated, adding 173 new symbols.
This provides the premier US large-cap benchmark with extended intraday histories – ideal for multi-sector trading bot training.

• Download symbol list (SPX.txt)

🇭🇰 Hang Seng Index (December 18, 2025) – Asia opens its doors

The Hang Seng Index (HSI) adds 82 entirely new symbols – major Hong Kong-listed companies with strong China exposure across finance, tech, energy, and consumer sectors.

• Download symbol list (HSI.txt)

🇪🇺 EURO STOXX 50 (December 17, 2025) – Europe uncovers its treasures

The EURO STOXX 50 introduces 50 blue-chip companies from the Eurozone, spanning multiple countries and sectors – a cornerstone for European market exposure.

• Download symbol list (STOXX50.txt)

🔖 Citation

If you use TroveLedger in your work, please cite it as:

text Copy

@dataset{Traders-Lab_TroveLedger_2025,
  author = {Traders-Lab},
  title = {TroveLedger Financial Time Series Dataset},
  year = {2025},
  url = {https://huggingface.co/datasets/Traders-Lab/TroveLedger}
}

💰 Support the Treasury Expansion

If you're training AI models or building quant pipelines with the TroveLedger dataset, consider supporting further global indices:

Thank you for helping grow the trove!

🔚 Final note

Markets are not measured by size alone —
but by how faithfully their records endure.