In this article
Historical roots are not a historical universe
Futures roots persist while individual contracts expire. A valid archive needs the actual expiries traded by the strategy, including contracts no longer listed today. Ask for per-contract first and last timestamps, not only the earliest date advertised for the root.
A continuous series adds derived history but does not replace contract files. Signal research and execution simulation should agree about which series each uses.
Choose trades, bars, quotes or settlements
One-minute OHLCV summarizes executed trades over an interval. Tick data retains individual trade observations. Quotes and order-book data describe displayed liquidity, while settlements are official daily values with a different purpose. Do not expect one product to imply all the others.
HFD tick bid_sz and ask_sz describe executed bid/ask-side volume, not resting queue size. This is an example of why the data dictionary matters: two fields with familiar names can represent very different market events.
Document time and session boundaries
Timezone-aware UTC storage does not by itself define a trading session. Intraday epoch-aligned bars and daily exchange-session bars can legitimately group the same trades differently. Holidays, early closes and daylight-saving transitions need explicit handling.
HFD daily bars use Chicago trading sessions opening at 17:00 local time with DST handling, while intraday buckets are epoch-aligned UTC. Preserve those definitions when deriving features or comparing another vendor.
Inspect the back-adjustment arithmetic
Additive back-adjustment shifts older prices by roll differences; ratio adjustment scales them. Neither is universally correct for every model. Negative adjusted prices can arise in some additive series and should be interpreted as a construction result rather than silently clipped.
HFD continuous files retain price_raw and back_adjustment alongside adjusted price, and roll ledgers document transitions. This lets a researcher inspect the transformation and use raw contract prices for accounting.
Our archive recommendation
Historical Futures Data is our first recommendation for individual quants who want broad CME and EUREX futures history in local Parquet files. The one-time pricing, eight bar sizes, per-contract ticks, continuous series and roll ledgers fit that specification directly. Free samples support an actual acceptance test.
This is a first-party recommendation, not a claim of independently measured universal accuracy. Use Databento, AlgoSeek, Tick Data, CME DataMine or EUREX when specialized exchange observations or another delivery model better match the requirement.
Sources
HFD timestamp and schema documentation