Point-in-time data
Point-in-Time Data: The Difference Between a Test and a Time Machine
A historical value is usable only if the simulation knows when that value became available.
Working definition
Point-in-time data represents what a decision maker could have known at each historical moment, including original releases, revisions, publication lags, and historical entity membership.
01
Observation time and availability time are different
A quarter-end fundamental describes a period that has ended, but the company may publish it weeks later. A macro series may be revised several times. A news event can carry a timestamp from the event, the publisher, the vendor, and the ingestion system. The backtest must use the timestamp that governs actual availability.
Joining by the date the value describes rather than the date it was known creates look-ahead bias even when every source row is historically accurate.
02
As-of joins need declared lag policy
The safe join selects the latest value available before the decision cutoff, applies publication and operational lags, and respects market calendars and time zones. When exact historical delivery times are unavailable, the policy should be conservative and the uncertainty stress-tested.
- Retain observation, publication, revision, and ingestion timestamps.
- Use original vintages when the strategy could not know revisions.
- Apply exchange calendars and decision-time cutoffs explicitly.
- Test longer lags to measure how much edge depends on timing precision.
03
Point-in-time is a property of the pipeline
Buying a point-in-time dataset does not guarantee a point-in-time backtest. Transformations, entity mappings, universe filters, and feature stores can reintroduce future information. Validation should trace the availability contract through every layer consumed by the strategy.
Practical takeaways
- Model when information became usable, not only what it described.
- Use original vintages and historical mappings where required.
- Make as-of joins, calendars, and lags explicit.
- Audit the full pipeline for reintroduced future knowledge.
Want an independent read?
Test the claim, the data, and the implementation together.
Validraft scopes the hypothesis, checks feasibility, and delivers a descriptive validation report with visible evidence and limitations. Research and simulation only; never investment advice.