CryptoDataDownload – Binance BTC/USDT Hourly OHLCV

Rows: 79,429
Columns: 10
Data range: 2017-08-17T04:00:00 to 2026-10-29T23:00:00
Auto-refresh: Re-checks its source daily
Last refreshed with new data: 2026-10-30

Hourly OHLCV for BTC/USDT on Binance from 2017 to present. Includes Open, High, Low, Close, Volume (BTC and USDT). One of the deepest liquidity pools in the crypto market. Free, CC BY-NC-SA 4.0. Widely used for ML model training and backtesting.

AI analysis

Dataset Analysis: Binance BTC/USDT Hourly OHLCV

1. Dataset Overview and Correlation Value

This dataset represents one of the most comprehensive publicly available records of Bitcoin price action, capturing hourly Open, High, Low, Close, and Volume (OHLCV) data for the BTC/USDT trading pair on Binance — the world's largest cryptocurrency exchange by volume. Sourced from [CryptoDataDownload](https://www.cryptodatadownload.com/cdd/BinanceBTCUSDT1h.csv) under a CC BY-NC-SA 4.0 license, it spans nearly nine years from August 17, 2017 through June 25, 2026, making it exceptionally deep for time-series modeling. The hourly granularity strikes a balance between noise and resolution, making it well-suited for machine learning model training, technical indicator backtesting, and macro-level correlation studies examining how Bitcoin price moves in relation to broader financial, sentiment, or on-chain variables. The fact that this dataset is freely available and widely adopted in the ML community means any derived correlations can be reproduced and validated by the broader research community.

2. Data Quality Observations

Data quality here is notably strong. All 76,623 rows contain zero null cells across all 10 columns, which is exceptional for a dataset of this length and vintage — early 2017–2018 crypto data is frequently plagued by gaps from exchange downtime or API inconsistencies. The Symbol column contains exactly one distinct value (BTC/USDT), confirming dataset purity with no cross-pair contamination. However, one structural concern warrants attention: the Date column shows only 76,117 distinct values against 76,623 total rows, implying approximately 506 duplicate timestamps. This is a meaningful discrepancy for a time-series dataset where hourly uniqueness is assumed — downstream analyses should deduplicate on Date or Unix before any rolling-window or join operation. The Unix column, by contrast, shows 76,623 distinct values (perfectly unique), making it a more reliable row identifier, though its extreme skewness (skew = 10.5, 672 outliers) and implausible max value of ~1.74 × 10¹⁵ suggest at least some timestamp encoding anomalies that deserve validation against the Date column before use as a join key.

3. Key Column Distributions

The four price columns (Open, High, Low, Close) are tightly co-distributed as expected, with means clustering around $37,450–$37,617 and medians near $27,300–$27,476, reflecting the right-skewed nature of Bitcoin's price history — long stretches at lower prices punctuated by dramatic bull runs to the $100K+ range (max Close = $126,011). The modest positive skew (~0.88) across all four price columns confirms this shape without extreme distortion, and the wide interquartile ranges (Q1 ≈ $9,286 to Q3 ≈ $60,363) capture the full arc from early bear markets through the 2021 and 2024–2025 bull cycles. Volume columns tell a more dramatic story: Volume BTC carries a skew of 5.71 with 7,893 flagged outliers, and Volume USDT skews at 4.44 with 5,457 outliers — both consistent with episodic liquidity spikes during major market events (liquidation cascades, ETF announcements, macro shocks). The tradecount column (skew = 3.47, 6,886 outliers, max = 3.52 million trades/hour) mirrors this pattern and is particularly valuable as a proxy for market participation intensity, independent of price direction.

4. Recommended Join Key Columns

The Date (DateTime) column is the natural join key for cross-dataset correlation, but given the ~506 duplicate timestamps identified, it should be deduplicated first — retaining the row with the higher tradecount or Volume USDT as the canonical record for that hour is a reasonable heuristic. For programmatic joins, Unix (epoch milliseconds) offers perfect row-level uniqueness, but requires validation of the anomalous max value (~1.74 × 10¹⁵, which would correspond to the year ~57,000 — almost certainly a data entry error in a small number of rows). A composite deduplication strategy of (Date, Unix) cross-validated against each other would yield the cleanest join surface. For lower-frequency correlation studies (daily, weekly), truncating Date to the day level and aggregating via OHLCV conventions (first Open, max High, min Low, last Close, sum Volume) is standard practice and would expand joinable dataset options considerably.

5. Suggested Pairing Datasets for Correlation Discovery

Several dataset categories would pair powerfully with this data. Macro financial markets — specifically S&P 500, NASDAQ, gold (XAU/USD), and DXY (US Dollar Index) at daily or hourly resolution — would test the long-debated question of Bitcoin's evolution from uncorrelated asset to risk-on proxy, a relationship that appeared to shift materially during the 2020 COVID crash and 2022 Fed tightening cycle. On-chain Bitcoin metrics (active addresses, hash rate, miner revenue, exchange net flows from sources like Glassnode or CryptoQuant) would complement price data with fundamental network health signals, enabling studies of whether on-chain accumulation patterns lead or lag price movements. Crypto Fear & Greed Index data (daily, from Alternative.me) offers a sentiment layer that could reveal whether extreme readings in tradecount spikes correspond to greed or fear regimes. Altcoin OHLCV data (ETH/USDT, BNB/USDT from the same CryptoDataDownload source) would enable Bitcoin dominance and correlation-rotation studies across crypto market cycles. Finally, macroeconomic event calendars (FOMC meeting dates, CPI release timestamps, ETF approval dates) could be joined as point-in-time markers to isolate event-driven volume and volatility signatures in the tradecount and Volume USDT outlier clusters already flagged in this dataset.

Columns

Date (datetime)
Unix (integer)
Symbol (string)
Open (decimal)
High (decimal)
Low (decimal)
Close (decimal)
Volume BTC (decimal)
Volume USDT (decimal)
tradecount (integer)