Market data looks simple — a price and a time — but the way it is recorded, aggregated and delivered shapes every analysis built on top of it. Understanding the main data types helps traders choose sources and avoid subtle errors.
Tick data
Tick data records individual trades and, often, quote updates as they occur. It is the most granular form widely available and is essential for studying execution, short-horizon strategies and microstructure. It is also large, noisy and requires careful cleaning: erroneous prints, out-of-sequence records and trades reported late are common.
Bars
Bars aggregate ticks into open, high, low, close and volume over a fixed interval — one minute, one hour, one day. Most retail tools and many research workflows use bars. Two details matter: which trades are included (regular session only, or extended hours) and whether prices are adjusted for splits and dividends.
Order book data
An order book shows resting buy and sell orders at each price level. Level 1 data provides the best bid and offer; deeper levels show more of the queue. Order book data is central to market making and execution research, and is especially important in crypto markets where venues publish full depth.
A model trained on adjusted prices but traded on unadjusted ones is a bug waiting to happen.
Why timestamps matter
Timestamps determine what was knowable when. Exchange timestamps, vendor receipt times and local arrival times can differ by milliseconds or more. For daily strategies the difference is negligible; for intraday research it can create artificial predictability. Fundamental data has the same problem at a slower pace: figures must be stamped with their publication date to qualify as point-in-time data.
Choosing a source
- Check coverage of delisted instruments to avoid survivorship bias.
- Confirm how corporate actions are adjusted and whether raw data is also available.
- Read the licence: redistribution and commercial use are often restricted.
- Test a sample against an independent source before committing.
Aggregators such as OpenBB make it easy to compare providers side by side, which is a sensible first step before building on any single feed.
Corporate actions in practice
Splits, dividends, spin-offs and symbol changes are the most common source of silent data errors. A two-for-one split halves the share price overnight; unadjusted, it looks like a 50% crash. Back-adjusted series correct historical prices so returns are continuous, which is right for return calculations but wrong for questions about the actual price level traded, such as whether a stock was below a minimum price threshold at the time.
Good practice is to store raw prices and adjustment factors separately, and to compute adjusted series as needed. That way both questions can be answered correctly from the same data.
Fundamental and alternative data
Financial statements, estimates and alternative datasets carry their own timing issues. Companies restate results; analysts revise forecasts; vendors backfill history when they add coverage. Ask every vendor whether their history is truly point-in-time or reconstructed — and, if reconstructed, from what date the data was actually collected live.
Costs of data
Exchange fees and vendor licences can be significant, and terms frequently distinguish between personal and professional use. Budgeting for data — and reading the licence before building a product on it — avoids unpleasant surprises later.