A stock screener usually shows only the companies that still exist today, which means every firm that was delisted, merged, or went bankrupt has silently dropped out of the list. Because the failures are missing, historical averages, win rates, and backtests look better than the market ever actually delivered. To read a screener honestly you have to remember the companies that are no longer there.
Key takeaways
- Survivorship bias is the distortion caused by studying only the survivors and ignoring the companies that dropped out.
- Delisted, merged, and bankrupt firms disappear from most screeners, so the visible universe is healthier than the real one was.
- Historical averages and backtests are inflated because the worst outcomes were removed before you ever saw the data.
- A point-in-time universe reconstructs which companies actually existed on a past date, which is the honest basis for any study of history.
- The fix is not a better filter but an awareness of what the list cannot show you.
What is survivorship bias in plain terms?
Survivorship bias is the error you make when you study only the things that survived and forget the ones that did not. The classic version comes from aircraft that returned from missions covered in bullet holes: reinforcing the parts that were hit would have been a mistake, because the planes hit in other places never came back to be counted at all.
In investing, the survivors are the companies still listed today. When you open a stock screener, you almost always see the current universe: the firms that exist right now. Every company that was delisted, absorbed in a merger, or wound up in bankruptcy has quietly left the list. You are studying the planes that came back.
How do failed companies drop out of a screener?
A company leaves the visible universe through several ordinary routes. It can be delisted for failing to meet exchange requirements, it can be acquired and cease to exist as a separate name, or it can collapse and be removed entirely. On the NSE and BSE, names disappear this way year after year, and a screener built from today's listed securities simply will not contain them.
The result is a list that is quietly curated by outcome. The companies that failed are not marked as failures, they are just absent. Because nothing on the screen tells you they are missing, it is easy to treat the surviving list as if it were the whole market, when it is really the market with its worst outcomes deleted.
Compare companies that trade today on Artha Terminal, with clear framing that the list reflects the current universe.
Why does this inflate historical averages?
Imagine the market a decade ago contained a hundred companies, and today seventy of them still trade while thirty were delisted, merged away, or went bankrupt. A screener that computes the ten-year average return over "these companies" is averaging over the seventy survivors, not the original hundred. The thirty worst stories were removed before the average was ever taken.
This is why backtested strategies so often look stronger on paper than in reality. A rule tested only on companies that survived to today has an unfair advantage: it never had to hold the ones that went to zero. The same distortion inflates CAGR figures, dividend track records, and "companies that grew earnings every year" lists. The averages are not lies, but they answer a different question than the one you think you asked.
What is a point-in-time universe and why does it fix this?
The honest way to study history is to reconstruct the universe as it actually existed on each past date, including the companies that later disappeared. This is called a point-in-time universe. A study run on it holds the firms you really could have owned then, failures included, so its averages reflect the outcomes an investor actually faced.
Building this correctly is demanding, because it requires knowing which companies were listed on each historical date and what happened to the ones that left. That is the same discipline described in Point-in-time data, and it is why serious historical work insists on it. When a dataset cannot tell you what its universe looked like in the past, its long-run averages should be read with caution.
How should you read a screener knowing all this?
The goal is not to distrust screeners but to read them for what they are: a snapshot of the companies that exist now, useful for finding and comparing them today. A screener is a good tool for filtering the present. It is a poor tool for claiming what the average company "historically" returned, because the historical losers are not in it.
Treat any average, ranking, or backtest that leans on history with the question "which companies fell out before this was calculated". Pair the screen with a check of the underlying figures, as covered in Reliable vs unreliable data, so you are not compounding a coverage gap with an accuracy gap.
On Artha Terminal, the screener is built for comparing companies that trade today, and the wiki is explicit that its filters describe the current universe rather than a survivorship-free history, so you know exactly what the list can and cannot tell you.