Point-in-Time Data
Point-in-time data records what was actually known on each historical date, including later-restated figures as originally reported, so that backtests do not use information from the future. It is closely related to Look-Ahead Bias, Survivorship Bias, Data Leakage and Walk-Forward Backtesting, each defined in the hedgewing.ai glossary.
Most financial databases store the latest version of every number. Earnings get restated, index memberships change, and tickers are reassigned, and the database quietly overwrites history with the corrected values. A point-in-time (PIT) database instead keeps every value with the date it became public, so that a query for 'what did we know about this company on 15 March 2019' returns exactly that, not the version revised in 2021.
The difference matters because a backtest that uses restated fundamentals is trading on knowledge nobody had at the time. A strategy that buys stocks with rising earnings can look brilliant if the 'rising earnings' were only recorded months later. The same applies to index constituents, analyst estimates, and even corporate actions such as splits that are applied retroactively.
Price-and-volume data is naturally closer to point-in-time than fundamentals, which is one reason hedgewing.ai's engineered features are built mainly from market data available at the prediction timestamp. The general principle for any investor evaluating a model is simple: every feature must be computable using only information that existed at the moment of the prediction, and the burden of proof is on the tool to show that it was.
FAQ
What is Point-in-Time Data?
Point-in-time data records what was actually known on each historical date, including later-restated figures as originally reported, so that backtests do not use information from the future.
Which terms are related to Point-in-Time Data?
Point-in-Time Data is closely related to Look-Ahead Bias, Survivorship Bias, Data Leakage and Walk-Forward Backtesting. Each has its own entry in the hedgewing.ai glossary.
Is this definition of Point-in-Time Data investment advice?
No. This glossary entry explains what Point-in-Time Data means. It is educational reference material, not personalised investment advice and not a recommendation to buy or sell any security.
Create a free Hedgewing account to try hedgewing.ai yourself.