How survivorship bias works
Suppose 100 small companies go public in the same year. Ten years later, 40 are still listed, 30 were acquired, 20 were delisted after running out of money and 10 went bankrupt. If you study the 40 survivors to learn what a good small-cap IPO looks like, you are studying the winners and the lucky. The 30 companies that failed are not in your data, and the 30 that were acquired may be recorded at their final price or not at all.
The bias does not require anyone to hide anything. It happens because databases, indexes and memories are built from what exists now. Losers are removed for good reasons: a delisted stock no longer trades and a merged fund no longer reports. The effect is that the sample is no longer the group that existed at the start.
Where it hides in investment research
| Where it appears | What drops out | How it distorts the picture |
|---|---|---|
| Fund databases | Funds merged or closed for poor results | Category returns look higher, dispersion looks smaller |
| Stock screeners used for backtests | Delisted, bankrupt and acquired companies | Strategy returns look higher, drawdowns look shallower |
| Index histories | Companies removed from the index | Index constituents look like a stable group of winners |
| Famous investor stories | Investors with the same approach who failed | A style looks more reliable than it is |
| Case studies of turnarounds | Turnarounds that ended in bankruptcy | Recovery looks more likely than it is |
| Hedge fund and newsletter records | Products that closed and stopped reporting | Track records of the category look better |
Fund data is the textbook case. Fund families often merge a poor performer into a better one, after which the poor record can stop appearing in many summaries. Hedge fund databases rely on voluntary reporting, so funds tend to stop reporting when results turn bad.
Stock data has the same problem in a more dangerous form. A screen for cheap stocks run today shows only companies that survived. If the screen is then used to ask "how would this have done over ten years", the companies that were cheap for a good reason and then failed are no longer in the list.
Why special situations are especially exposed
Many of the situations Dombey covers have outcomes that remove the stock from view:
- Bankruptcies. In many Chapter 11 cases the old equity is cancelled. Our explainer on what happens to stock in Chapter 11 covers why the shares often go to zero and then leave the exchange.
- Failed biotechs. A company whose only drug fails a phase 3 trial may wind down or reverse-merge. Its history often survives only in old filings.
- Broken deals. Targets of failed mergers sometimes recover and sometimes do not. The ones that do not tend to be forgotten.
- Net-nets and stocks below cash. Some cheap stocks are cheap because cash is about to be spent or lost. Those that burned through cash and delisted vanish from screens. Our guide on screening for net-net stocks covers the checks that separate cash that stays from cash that leaves.
- Delisted and going-dark companies. Companies that file a Form 15 and stop reporting leave a gap in the data. See what going dark means.
The result is that a casual review of past special situations, built from what is easy to find, will mostly show the ones that worked. That is a reason to be skeptical of any claim that a category of situations "usually" works out.
How to correct for it
There is no single fix, but a few habits remove most of the distortion.
- Define the sample at the start. List every company, fund or idea that met the criteria on the start date, not the ones that still meet them now.
- Use data that includes the dead. Survivorship-free datasets keep delisted securities with their final prices and the reason they left. When a vendor says its backtests are survivorship-free, check how delisting returns are recorded, since a missing final price can be as misleading as a missing stock.
- Track your own ideas, including the ones you passed on. A decision journal records each idea when it is considered, so the losers cannot quietly fall out of memory.
- Ask what the base rate is. Before trusting a story about winners, look for the full population that looked similar at the start. Our explainer on base rates and the outside view shows how to frame that question.
- Read the failures. For any approach, look for well-documented cases where it failed and why. A pre-mortem applies the same thinking forward to a live idea.
Where this goes wrong
Correcting for survivorship bias has its own traps.
- Overcorrecting. Not every successful example is luck. The goal is a fair sample, not a belief that nothing works.
- Bad delisting data. A dataset can include delisted stocks but record them at their last traded price, which may be far above what holders actually received. The final return matters as much as the presence of the stock.
- Look-ahead bias in the fix. Defining the sample using information that was not available at the start, such as a later restatement, introduces a different error.
- Mixing time periods. A survivorship-free sample from one decade may not describe another, especially when listing standards or market structure changed.
- Ignoring acquisitions. Acquired companies are survivors in a sense, since holders were paid. Treating every departure as a failure understates the results of approaches that find takeover targets.
Common mistakes
- Learning only from famous winners. The investors who became well known for a style are a sample of survivors. Many others used the same style and are not known.
- Backtesting on today's index members. The current constituents are, by construction, companies that did well enough to stay in.
- Comparing a fund to its category average. If the category average excludes closed funds, the benchmark itself is inflated.
- Assuming a stock that disappeared was acquired. Check the filings. A delisting can mean a sale, a bankruptcy or a decision to stop reporting, each with very different results for holders.
- Forgetting your own discarded ideas. Ideas you dropped are part of your record. Without them, your sense of your own record is a survivor's view.
Research tools and survivorship
Most research platforms show currently listed securities by default, which suits idea generation. Fewer let you see the companies that left. When choosing tools, check whether they keep delisted companies, how they record final prices and whether historical screens use point-in-time data. Our list of the best stock research tools for individual investors and our review of backtesting tools for stock screens note which tools handle this well.
SEC filings are a reliable fallback. A company's final 8-K, its Form 15 or its bankruptcy filings remain on EDGAR after the stock stops trading, which makes it possible to reconstruct what happened to holders. Our guide on how to search SEC EDGAR filings explains how to find them.
The bottom line
Survivorship bias makes the past look better than it was because the failures are missing from the data. It shows up in fund databases, backtests, index histories and the stories investors tell about winners. The fix is to define the sample as it existed at the start, include the companies and funds that left, and read the failures as carefully as the successes. In special situations, where losers often disappear entirely, that discipline is the difference between a real base rate and a flattering one.

