Blockchain analytics firm Coin Metrics has recomputed its historical Ethereum (ETH) Standard Flow Metrics from the network's first block, exposing critical data vintage risks for trading algorithms that rely on historical exchange flows. The revision notice, published on Oct. 1 at 17:04 UTC, updated all daily and hourly ETH exchange metrics using newly identified address clusters.
Key Takeaways
- Coin Metrics updated all historical ETH Standard Flow Metrics from block one, backfilling data with newly discovered wallets.
- Restated historical data creates discrepancies between current charts and actual point-in-time (PIT) information available on past trading dates.
- Competitor CryptoQuant warned that its ETH exchange flow endpoints do not support PIT accuracy, revising data weekly every Tuesday at 00:00 UTC.
- A Glassnode test demonstrated that backtests using restated data yielded unrealistically superior returns compared to true historical PIT datasets.
Standard Metrics Versus Point-in-Time Data
The distinction between Standard and Point-in-Time (PIT) metrics centers on wallet attribution. Standard flow metrics assign present-day entity knowledge to historical transactions, starting an address's history at its first non-zero balance. When Coin Metrics identifies a new exchange wallet, it retroactively classifies all past transfers as exchange flows.
In contrast, PIT data reflects only what was known to market participants at a specific historical moment. An address contributes to PIT metrics only after its discovery date. Coin Metrics initially outlined the ETH recomputation on Sept. 28, aiming for a Sept. 30 completion before publishing final notice on Oct. 1 at 17:04 UTC.
Other data providers operate under similar constraints. CryptoQuant explicitly documents that its ETH Exchange Flows endpoint does not offer PIT accuracy, as historical values change when exchange wallets are identified and added during clustering updates scheduled every Tuesday at 00:00 UTC. While these revisions provide a clearer picture of past network activity—even as institutional products shift weekly Ethereum ETF flows—they introduce look-ahead bias if used unadjusted in strategy backtesting.
Quantifying Backtest Distortion and Vintage Risk
The impact of data revisions on trading strategies was illustrated in a March 13, 2026 study by Glassnode. In a hypothetical backtest spanning Jan. 1, 2024 to March 9, 2026, Glassnode simulated a momentum rule using Binance's BTC exchange balance. The strategy entered positions when a 5-day moving average dropped below a 14-day moving average, exiting when the shorter average rose above the longer one. Starting with $1,000 and charging a 0.1% fee per trade, the rule yielded significantly worse performance when tested on true PIT data versus revised historical balances.
Glassnode noted that PIT data tracking expanded across all platform metrics in July 2025, having previously been limited to BTC, ETH, and select assets. Additionally, the platform has recorded computed_at API publication timestamps since September 2024 to account for computation delays.
Why It Matters
Data vintage risk is a hidden trap for quantitative crypto traders and algorithmic funds. When analytics platforms retroactively update exchange address labels, historical charts look far more predictive than the real-time data actually was. Traders who backtest strategies on revised historical datasets risk building algorithms around signals that were impossible to execute in real time, leading to disappointing real-world performance.
