Survivors only
Testing on the companies that are still around today, and missing the ones that died.
Most price databases you can download today contain the companies that exist today. The ones that went bankrupt, were delisted or were bought years ago are quietly missing. A test run on such data is a test on survivors only, and survivors, by definition, did not go to zero.
The result is an optimistic bias in almost every direction. Buying losers looks safer than it was, because the losers that died are not in the sample. Shorting hated names looks worse than it was, because the names that vanished were often the ones being shorted. The logbook measured the size of that bias on its own data and found it inflates the apparent quality of a strategy by more than half.
The cure is to rebuild the universe date by date, including the dead. Every entry says whether that was done, and results that could not do it carry a discount.
Explore the diagram (scroll horizontally on small screens)
Where this word comes up
First needed in the log entry H-042 · The post-earnings bounce did not survive our cost assumptions.
From the same entry: Backtest, Basis point, Buy the dip, Correction for looking twice, Gross edge, Index membership, Large caps, Net return, Post-earnings drift, Pre-registration, Published anomaly, Quarterly results, Real costs, RSI indicator, Short side, Terciles.