What is survivorship bias? The problem with today’s stock list

Survivorship bias arises when analysis includes only subjects that remain later, omitting some that existed at the time. Backtesting today’s listed stocks can exclude securities that disappeared through delisting or corporate reorganization.

A long price history does not establish complete historical universe coverage. Before reading performance, check which date’s information was used to select the securities.

In this fictional teaching example, three securities present at the start, A, B and C, return +10%, 0% and −80% over one interval. An equal-money average is (10+0−80)÷3 = −23.33%. These are not actual securities or results.

If C alone disappears from a later list and that list is used to study the past, A and B average +5%. Selection now depends on later information. This example illustrates changing samples; it does not reproduce settlement for an actual delisted security.

Delisting can follow an acquisition or reorganization as well as business failure. Excluded securities need not always have low returns. The definition alone cannot determine the bias’s magnitude or direction.

The key question is whether later outcomes influenced historical selection. When names or codes change, document whether histories are linked or treated as separate securities, and why.

Reconstructing the historical universe may require listing and delisting dates, category eligibility, constituent changes and identifier histories. Copying current constituents onto older dates does not reconstruct that universe.

Do not invent final prices when the final trade, acquisition consideration, suspension or cash-settlement timing is unknown. Check the availability of price history separately from whether a security belonged in the historical universe.

Record the universe reference date, inclusion rules, missing securities and periods with unverified membership. A long dataset or file hash does not guarantee historically complete coverage.

KabuForge’s public course leaves historical information availability unverified. A completed run or screenshot does not establish that survivorship bias has been resolved. Examine dataset coverage and the timing of available information. Related explanations: point-in-time data, look-ahead bias, out-of-sample testing.

Workflow and display source: public GUI research course (reviewed October 6, 2026). Fictional examples illustrate calculations and are not course results.

Public course results are research-only; historical information availability is unverified. Explaining a general definition does not establish full software support. No securities or trades are recommended. Read the disclaimer.

Explore the backtesting learning path

• Historical Paper Replay and Forward Testing: Different Evidence
• What Is Backtesting? What Historical Tests Can and Cannot Tell You
• What are transaction costs? Why fees change backtest results