Survivorship Bias in Signal Providers: Why the Leaderboard Shows the Survivors

Survivorship bias is the distortion you get when you only see the signal providers still standing: the ones who blew up an account, quit, or deleted their channel are no longer on the leaderboard, so the records that remain look far better than the group’s real results. The providers you are shown are not a random sample; they are the survivors of a filter that removed most of the bad outcomes.

In short: if many people trade with no skill at all, chance alone will give some of them excellent records. Remove the ones who lost, rank the rest by return, and the top of the list looks like proof of talent. The academic work on fund performance documented this decades ago, and signal and copy-trading markets are more exposed to it than funds, because nobody is required to report the accounts that failed.

The pitch you have probably seen

“Top 10 signal providers this year.” “Our best trader: +180% in 12 months.” A copy-trading app shows a leaderboard of traders sorted by return. Each record may be entirely real. The problem is everything that isn’t on the page: the traders who were on last year’s leaderboard and aren’t now, the accounts that were closed after a bad quarter, and the channels that were renamed and relaunched.

What the research found

Survivorship bias was studied first in mutual funds, where the data is far better than anything available for signal providers.

Stephen Brown, William Goetzmann, Roger Ibbotson and Stephen Ross, “Survivorship Bias in Performance Studies” (Review of Financial Studies, 1992, vol. 5, no. 4). They showed that when a sample is truncated by survival (managers who do badly drop out), the surviving records can show apparent performance persistence even when there is none. In other words, the filter itself can make past winners look like they keep winning.
Edwin Elton, Martin Gruber and Christopher Blake, “Survivorship Bias and Mutual Fund Performance” (Review of Financial Studies, 1996, vol. 9, no. 4). Tracking funds that later merged or disappeared, they estimated that studying only surviving funds overstated risk-adjusted performance by roughly 0.9 percentage points a year.

Mutual funds have to file reports with regulators, and databases keep the dead ones. A signal provider’s failed Telegram channel or closed copy-trading account usually leaves no trace at all, so the bias in what you see can be much larger.

How the filter works in signals and copy trading

Survivorship bias gets into signal marketing in several ways:

Leaderboards. Copy-trading platforms rank active accounts by recent return. Accounts that lost heavily stop trading or are closed and drop off. How these platforms work is covered in how copy trading works.
Relaunches. A channel with a bad record is abandoned and a new one starts with a clean history.
Multiple accounts. A provider runs several accounts or strategies and advertises the best one. Each is real; the selection is not.
Start dates. A record that “starts” just after a big losing period.
Marketing best months. “We made 40% in March” says nothing about April.

The fraud version: the stock-tip letter

Bailey, Borwein, López de Prado and Zhu describe a classic scheme in their 2014 Notices of the American Mathematical Society paper on backtest overfitting. A manager sends a forecast to a large mailing list, telling half that the market will rise and half that it will fall. Next month he writes only to the half who received the correct call, and splits them again. After n rounds, the remaining recipients have seen n correct calls in a row.

The arithmetic: to end with 10 people who have seen 10 straight correct calls, you need to start with 10 x 2 to the power 10 = 10,240 people. For 6 straight calls, 640. A signal group can do the same thing with free channels, sending opposite calls to different audiences and promoting whichever channel got lucky. Every person at the end has a perfect, real record in front of them. The paper’s point is that the number of attempts is the missing information.

Worked example: a leaderboard built from zero skill

We simulated 1,000 signal providers in Python, none of whom has any skill. Each month, each account’s return is random, centred on zero with a standard deviation of 6% (a volatile but not extreme forex account). Over 24 months, any provider whose account falls 25% below its starting value gives up and disappears. Then we built a leaderboard from the survivors.

Group	Number	Average 2-year return
All 1,000 providers (including those who quit, at the point they quit)	1,000	-0.1%
Survivors still on the leaderboard	692	+12.2%
Survivors’ median		+7.2%

The top 10 survivors were up between 83% and 266% over two years. None of them had any skill; the whole population averaged roughly zero, as it was designed to. The leaderboard’s average looks like a modest edge, and its top looks like genius, purely because the losers left.

A related check: if 1,000 providers each have a 50% chance of a winning month, the number who show 9 or more winning months out of 12 is, on average, 73, and about 19 would show 10 or more. Those 73 would make an impressive “top providers” page.

A longer record helps, but doesn’t cure it

Length is the best single defence, because luck has to keep repeating. Out of 1,000 zero-skill providers, each with a 50% chance of a profitable year, about 125 would still show three profitable years in a row, about 31 would show five, and about 8 would show seven. Long records thin the field quickly, but a large enough starting crowd still leaves a few spotless survivors. That is why the number of accounts and strategies a provider has run matters as much as the length of the one they show you, and why a long record is a necessary condition for trust rather than a sufficient one.

Why leaderboards also reward risk

Survivorship bias interacts with risk-taking. The traders at the top of a return-ranked list are disproportionately the ones taking the biggest risks, because high risk produces both the biggest gains and the biggest losses, and the losses are filtered out. In an experiment published as “Copy Trading” (Management Science, 2020), Jose Apesteguia, Jörg Oechssler and Simon Weidenholzer found that giving people information about others’ success and the ability to copy them led to excessive risk taking, with participants much more likely to copy the riskiest choices.

The base rate matters too. EU regulators required CFD providers to warn retail clients after national regulators’ analyses found that 74-89% of retail CFD accounts typically lose money (ESMA, 2018). A leaderboard of winners is drawn from a population in which most accounts lose. The evidence on retail results is collected in why most retail forex traders lose.

How to correct for it

You can’t see the missing accounts, but you can ask questions that expose the filter:

How many accounts, channels or strategies have you run, including closed ones? One success out of one is very different from one out of twenty.
When did this account start, and was there a previous one? A record that begins after a reset is a survivor by construction.
How long is the record, and how many trades? Luck fades with time and trade count. Statistical significance of a track record shows how many trades it takes.
On a leaderboard, how long has this trader been ranked, and what happened to last year’s top 10? If last year’s leaders are gone, expect this year’s to follow.
Is the return driven by a few huge months? A record made of two lucky months and twenty flat ones is the signature of risk-taking that happened to pay.
What is the maximum drawdown? High return with a deep drawdown is often the profile of a trader who got lucky after nearly failing.

What can go wrong

Survivors stop surviving. The traders who reached the top through risk are the most likely to fall out next.
You join at the peak. People subscribe after a great run, which is when a lucky streak is most likely to end.
Real records, false conclusion. Every number on the leaderboard can be accurate and the inference (“these people have skill”) still wrong.
Backtests have the same problem. A strategy chosen as the best of many variations is a survivor too. See backtests vs live results.

The short version

The providers you see are the ones who haven’t failed yet. Research on funds found that ignoring the dead ones overstates performance and can create the illusion that winners keep winning; signal markets, with no reporting of closed accounts, are more exposed. A simulation of 1,000 zero-skill traders produces a leaderboard averaging +12% and topping out above +250%. Ask how many accounts a provider has run, how long the record is, and what happened to the last set of leaders.