PROOF · proof · Updated 2026-11-05 · Derek William Frazier

What Is Survivorship Bias, and How Does It Fake a Trading Track Record?

Survivorship bias is what happens when a track record only counts the accounts, strategies, or traders that survived long enough to be counted — and quietly drops the ones that didn't. The record you're shown isn't a lie about any single call. It's a lie about the population the calls were drawn from.

PLATE · WHO'S LEFT TO COUNTEleven accounts start together. The average only ever describes the ones still here.started togetherwhat gets countedonly the 3 still here get averaged8 of 11 quietly dropped, never countedRemoving the losers before the count isn't editing the record — it's rigging it.CYCLICAL MARKETS · THE CYCLITECNICAL METHODEducational. Not advice. No performance promise.

What is survivorship bias, exactly?

It's a measurement error, not a deception in any one data point. It happens whenever a group gets filtered down to "whoever's still standing" before anyone runs the numbers, and the filtering itself is never mentioned.

The classic version: a fund company runs 20 strategies. Five do well, fifteen don't. Over a few years the fifteen quietly get shut down, merged into other funds, or renamed. Five years later, the marketing page shows the average return of "our strategies" — which is now just the average of the five survivors, dressed up as if it always described all twenty. Nobody edited a number. They just stopped counting the ones that lost.

Where does it show up in trading specifically?

In three places, most often, and all three share the same mechanism — something disappears from view before the count is taken.

Why does it inflate the numbers so much?

Because the accounts, strategies, or traders that get dropped are disproportionately the losers — that's usually why they got dropped — so removing them doesn't just shrink the sample, it skews it. A fair coin flipped by a thousand people will, by chance alone, produce some people with long, lucky winning streaks. Publicize only the streak-holders and quietly retire everyone else, and the group average looks like skill. It isn't. It's what happens to any large starting population once you filter it down to "whoever happened to win."

What gets shownWhat's missing from the count
"Our top traders beat the market by a wide margin"The traders who no longer trade for the firm
"This exact parameter set backtests the best of anything we tried"The dozens of similar parameter sets tested and discarded
"My track record starts in March"Whatever the record looked like before March

How is this different from a cherry-picked track record?

They're related but not identical. Cherry-picking is choosing which calls to show, after the fact, from one continuous record. Survivorship bias is choosing which records — whole accounts, whole strategies, whole traders — get to exist in the comparison at all. A single honest trader can still cherry-pick individual calls from their own real history. A firm with survivorship bias can be scrupulously honest about every call each surviving trader made, and the bias would still be there, baked into which traders made it into the room being measured.

What's the one question that catches it?

Ask what the group looked like at the start, not just who's left now: "How many strategies/traders/accounts did you begin with, and what happened to the rest?"

A record with nothing to hide answers this easily and specifically — a number, a reason, a date. A record built on survivorship bias either can't answer it, answers with a vague "some didn't work out," or answers a different question than the one you asked (how the current group is doing, instead of how the original group did). The tell isn't a wrong answer. It's the absence of any count for the group that no longer exists.

The same "what got dropped from the count" question applies one step earlier, to ideas rather than accounts — see what data-mining bias is and why testing many strategies and keeping only the winner inflates the result.

Questions traders ask

Does survivorship bias only apply to funds and firms, not individual traders?

No — a single person can build a survivorship-biased record of their own without ever intending to. Someone who tries five different trading approaches over two years, keeps trading the one that happened to work, and then shows you only that one's track record has quietly dropped the other four from the count. The honest version states all five and what happened to each, the same way a record should state its losses per how a trading educator should score their own calls.

Is a backtest that only shows the best-performing version of a strategy always dishonest?

Not necessarily dishonest — sometimes it's just incomplete, and the person presenting it may not realize how much the selection process itself inflated the result. But the effect on you, the reader, is the same either way: the number you're looking at describes the winner of a search, not the expected performance of a strategy. Ask how many variations were tested before landing on this one.

If a track record just started recently, is that automatically survivorship bias?

Not automatically — every honest record has to start somewhere. It becomes a survivorship-bias question when there's a reason to suspect an earlier record existed and was dropped, replaced, or renamed rather than simply not existing yet. A brand-new trader with a brand-new record isn't hiding anything; a trader on their third "new" account in two years likely is.

How does this connect to sample size?

They compound. A small sample is already noisy on its own — see how many calls a track record needs before it means anything. Survivorship bias makes a small sample worse by ensuring the noise all points one direction: toward the lucky outcomes, because the unlucky ones were the ones removed. A large N doesn't fix this either — a thousand calls drawn only from surviving accounts is a bigger version of the same skew, not a cure for it.

The full picture of evaluating any claimed record: start with how to spot a cherry-picked trading track record for what gets left out of a single ledger, how many calls a track record needs before it means anything for whether the count is large enough to trust, and why every hot streak cools off for the sibling mistake — not a filtered population, but a single hot stretch pulled from an honest one. How to verify a trading guru's credentials is the five-minute background check. My own graded record, survivors and losers together, is at the four-year scoreboard. The method behind the calls there is documented at the methodology hub.

Keep reading

I Graded Four Years of My Own Public Calls. Here's the Scoreboard.How a Trading Educator Should Score Their Own CallsHow to Verify a Trading Guru's CredentialsMy 2022 Posts Beat My 2026 Posts. Badly.
Get our AI tools and everything we send out New AI trading tools and our Claude Code builds, plus session invites. All free.
Get them free