Point-in-Time Backtest
A point-in-time backtest simulates a strategy using only the information that was actually available on each historical date. It is the discipline of not letting the test know things the investor could not have known.
How it is measured
The distinction matters because economic data is revised. A figure published today for last quarter is not the figure that was available last quarter, and testing against the revised number credits a decision that could not have been made.
The same applies to index membership. Testing a strategy on the companies that are in an index today, over a period when they were not, quietly excludes everything that failed and left.
How to read it
A point-in-time result will almost always look worse than one built on final data, and that gap is the size of the illusion rather than a defect in the method.
Transaction costs matter for the same reason. A strategy that trades frequently can look excellent before costs and unremarkable after them, so costs modelled on actual turnover are part of an honest test.
A worked example
A screen that buys companies on a particular reported metric can be tested two ways. Using data as revised today, it looks strong. Using the numbers as first published, some of those companies did not qualify at the time. The second result is the one an investor could actually have obtained.
The most common mistake
Trusting a backtest because it is detailed. Precision is not accuracy. A test with hourly resolution built on revised data is more confidently wrong than a rough test built on point-in-time data.
How CORVIX uses it
CORVIX's backtester runs point-in-time with no look-ahead, models transaction costs on actual turnover, and adds a timing-luck stress test comparing best, worst and neutral entry timing against staying in cash. That last one exists because the same strategy produces very different outcomes depending on when deposits happened to land.
Common questions
What is Point-in-Time Backtest?
A point-in-time backtest simulates a strategy using only the information that was actually available on each historical date. It is the discipline of not letting the test know things the investor could not have known.
How is point-in-time backtest measured?
The distinction matters because economic data is revised. A figure published today for last quarter is not the figure that was available last quarter, and testing against the revised number credits a decision that could not have been made. The same applies to index membership. Testing a strategy on the companies that are in an index today, over a period when they were not, quietly excludes everything that failed and left.
What is the most common mistake when using point-in-time backtest?
Trusting a backtest because it is detailed. Precision is not accuracy. A test with hourly resolution built on revised data is more confidently wrong than a rough test built on point-in-time data.