Survivorship Bias in Historical Market Data

Survivorship bias removes the companies that failed before the test even starts, and it inflates a backtest's return in a specific, measurable way.

Survivorship bias is what happens when a backtest's universe of tradeable instruments is built from a list of companies that exist today rather than a list of companies that existed on each date being tested. Every firm that went bankrupt, was delisted for falling below a listing threshold, or was acquired at a depressed price is absent from the test, and a strategy tested only against the winners will report a return no live trader could ever have earned.

The most common route to the bias is downloading a current index membership list, such as the current constituents of a broad equity benchmark, and applying that list to a decade of historical prices. A stock that was in the index in 2016 and dropped out in 2019 after a bankruptcy filing never appears in the current list, so the backtest never trades it and never records the loss. The strategy looks like it consistently avoided disaster, when in fact it was never given the chance to encounter one.

Point in time index membership files, which record which constituents were in an index on each historical date rather than only the current set, are the standard fix. Several data vendors sell these files specifically because the mistake is common enough that a dedicated product exists to prevent it.

A related problem sits inside the price series itself rather than the universe list. When a company is delisted, many data providers simply stop the price series on the last trading day rather than recording the often severe loss that shareholders took in the delisting process, whether through a bankruptcy filing, a forced sale, or a squeeze out at a low price. A backtest that treats a stopped series as a neutral exit, closing the position at the last available price with no further loss, understates the true cost of holding positions that later fail. Academic studies of delisting returns have found the average loss in the month a stock is delisted for financial distress to be substantial…

Survivorship bias does not distribute evenly across the market. Large, well capitalized companies rarely disappear, so a strategy trading exclusively in names like the largest technology or bank stocks is affected only modestly by the bias. A strategy trading small capitalization or low priced stocks, where the annual rate of delisting for cause is meaningfully higher, can see its backtested return overstated by several percentage points a year once dead constituents are added back. Value and distressed credit strategies are particularly exposed, since the entire premise of the strategy is holding companies that are statistically closer to failure than average, which is exactly the…