Most historical datasets quietly describe the past as we understand it today, not as anyone experienced it at the time. Earnings get restated. Databases get cleaned. Bankrupt companies get dropped. Test a strategy on that polished version of history and you're handing it information no real investor had — which makes results look better than they honestly were.
Point-in-time data fixes this by answering a stricter question: not "what were this company's earnings for Q1?" but "what did investors believe this company's Q1 earnings were on the specific day the strategy would have traded?" Those are different questions, and the difference is where fake performance hides.
A company's quarter ends March 31, but the numbers aren't filed until weeks later. A backtest that lets an April 1 trade use those numbers is trading on information that didn't exist yet — look-ahead bias. It sounds small; compounded across thousands of simulated trades, it can manufacture several points of annual return out of thin air.
Restatements make it worse. If a company later corrects its revenue downward, the modern database remembers the corrected number — but investors at the time saw the original, rosier one. An honest simulation has to see what they saw, mistakes included. Point-in-time storage keeps every value stamped with the date it became public knowledge, so the simulation can only ever use what was truly known.
Screen today's market and every company you see is, by definition, a survivor. The delisted, the bankrupt, and the acquired-at-a-loss have vanished from the list. Backtest only on survivors and your strategy gets to skip most of history's disasters — a silent, massive tailwind.
The classic trap is testing on an index's current membership. Run a strategy on today's large-cap list back through twenty years and you've cheated twice: you excluded the companies that failed out, and you "discovered" the future winners decades early. Honest testing uses the full universe as it stood on each historical date — graveyard included, future stars absent until they actually arrived.
Well Street's historical data is stored so that every fundamental value carries its as-of date — the date the market actually learned it. When the backtester evaluates a rule on a simulated day, it queries the snapshot of the world as of that day: the reported (not restated) fundamentals, the universe of stocks that were actually listed and tradable, and prices adjusted correctly for the splits and dividends that had occurred by then.
That's the whole trick, and it's less glamorous than it sounds: mostly it means refusing shortcuts. It's the difference between a time machine and a highlight reel.
Because the point of sharing strategies is learning what actually works, and inflated backtests teach the wrong lessons. Strategies validated on hindsight-polished data routinely fall apart in live markets — not because markets changed, but because the edge never existed. It was an artifact of the test.
When you compare ideas on Well Street, you're comparing them on a level field where nobody — including us — gets to peek at the future. A strategy that survives that standard has cleared a real bar. It still isn't a guarantee of anything; it's simply evidence you can take seriously.
← All methodology pages · Investing concepts library