Most trading backtests are quietly optimistic. They run a strategy over the stocks that exist today — and silently leave out every company that was delisted, went bankrupt, or got acquired along the way. That single omission, called survivorship bias, can turn a losing strategy into a winning-looking one.
What survivorship bias is
Survivorship bias is the error of analysing only the things that "survived" a selection process and ignoring the ones that didn't. In markets, the survivors are the tickers still trading today. The casualties — the pump-and-dumps that went to zero, the reverse-split penny stocks, the bankruptcies — vanish from most datasets.
If your strategy buys low-float momentum names, many of those names eventually collapse. A backtest that only sees the ones that survived will dramatically overstate your edge, because it never experienced the blow-ups.
Why it inflates results
- Losers are removed: the worst outcomes are missing entirely, so average returns look higher.
- Risk looks lower: drawdowns and tail losses shrink because the catastrophic names aren't there.
- Patterns look more reliable: a setup that 'always works' may simply have had its failures deleted.
How to test honestly
The fix is a point-in-time, survivorship-bias-free dataset: the universe must include every stock that was tradeable on each historical date, including ones later delisted. TradeScaner's backtest spans roughly 277,000 setups from 2016–2026 with delisted tickers kept in — so the engine is measured against the failures, not just the survivors.
Pair that with walk-forward, out-of-sample testing (tune on the past, measure on data the model has never seen) and you get expectancy numbers that don't fall apart the moment you trade them live.
The takeaway
When you evaluate any scanner or strategy, ask one question: does the backtest include the stocks that died? If the answer is no — or if nobody can tell you — treat the results as marketing, not evidence.