klines-binance

Binance USDT OHLCV Klines (spot + perpetual futures)

Historical OHLCV candlestick (kline) data for Binance USDT markets — both spot and USDT-margined perpetual futures — stored as zstd-compressed Parquet.

Coverage

• Spot seeded from the DigiTuccar HistoricalDataForTradeBacktest snapshot (static, ~2025-05).
• Futures is fresh to 2026-07 and retains delisted symbols, so it can be used for point-in-time / survivorship-bias-free backtests.

File layout

{market}/{timeframe}/{PAIR}.parquet

Examples: - spot/1d/BTC_USDT.parquet - futures/1m/BTC_USDT_USDT.parquet - futures/1h/BTC_USDT_USDT-mark.parquet (mark price, where available) - futures/8h/BTC_USDT_USDT-funding_rate.parquet (funding rate, where available)

Spot pairs are BASE_USDT; futures pairs are BASE_USDT_USDT (USDT-margined perp).

Each OHLCV Parquet has columns:

Usage

from huggingface_hub import hf_hub_download
import pandas as pd

path = hf_hub_download("rogerdehe/klines-binance",
                       "futures/1d/BTC_USDT_USDT.parquet", repo_type="dataset")
df = pd.read_parquet(path)

Subset / whole dataset:

from huggingface_hub import snapshot_download
snapshot_download("rogerdehe/klines-binance", repo_type="dataset",
                  allow_patterns=["futures/1d/*"])   # e.g. daily futures only

Notes

• Timestamps are UTC. Timeframe coverage varies slightly per pair (some newer listings have shorter history; 1h futures is a partial subset).
• mark / funding_rate files exist only for a subset of futures pairs.
• The HF dataset viewer is disabled (custom per-pair layout); load files directly as above.
• Data provided as-is for research; no warranty of accuracy or completeness.

Coverage (per-file verified)

Every file was inspected directly — parquet footer for the date range and row count, and a full timestamp scan wherever a file was not provably complete. Dates are UTC candle-open. A last date far in the past means the pair was delisted / went inactive.

Summary

The full per-pair coverage table (every pair × timeframe, with first→last ranges) and the complete list of 329 internal gaps are in COVERAGE.md; the raw scan is in coverage.csv. The largest gaps are shown below.

Large internal gaps (largest contiguous hole > 3 days)

329 pair·timeframe series contain a contiguous hole longer than 3 days (usually a delisting/relisting or a trading halt).

… and 249 more — see the complete list in COVERAGE.md.