Bitcoin is the perfect example.
Today, we can look at its history and observe that a purchase made at a few dollars would have generated a fortune. We can find statements from those who predicted its demise in 2011, 2014 or 2018 and calculate how much an investment made when each of those criticisms was voiced would be worth today.
The calculation is interesting.
The conclusion can become dangerous.
Why? Because we already know that Bitcoin survived.
We do not perform the same exercise with the thousands of cryptocurrencies launched at the same time or during subsequent cycles that disappeared, lost almost all their liquidity, were abandoned or replaced by other projects.
The problem is even more striking with meme coins. CoinGecko studied more than 18.67 million tokens created on Pump.fun between January 2024 and June 2026. The result: 68.67% recorded their last trade on the very day they were created. Including those that disappeared the next day, 80.37% had not exceeded one day of activity under this methodology. Only 4.55% lasted beyond 90 days, although there is an important methodological limitation: some tokens that migrated to external DEXs were no longer tracked in the same way.
Another CoinGecko study estimates that 53.2% of the cryptocurrencies active on GeckoTerminal between July 2021 and December 2025 were no longer trading, with 11.56 million failures in 2025 alone.
These millions of dead projects rarely appear in threads titled “here’s how to turn $1,000 into $1 million.”
That is precisely where survivorship bias begins.
Survivorship bias in crypto makes the losers disappear
The concept has existed for a long time in statistics and finance. The CFA Institute describes it as a problem that arises when the data being studied contains only the entities that survived until the end of the period. The resulting statistics can then become misleading.
In crypto, this mechanism is particularly severe because the number of tokens and projects is exploding, while the most visible survivors gradually concentrate attention, liquidity and collective memory.
We remember Bitcoin, but much less of its contemporaries
Consider an extremely common statement:
“If you had invested $1,000 in Bitcoin fifteen years ago, you would be rich today.”
That is true in its context.
The problem begins if that observation becomes:
“So you simply need to buy a new cryptocurrency very early and hold it.”
The second claim absolutely does not follow from the first.
Bitcoin has endured several bear markets, bans, platform bankruptcies, internal divisions, regulatory criticism and periods when its value fell by more than 70%. Today, it has global liquidity, an extremely secure network, institutional infrastructure and a market capitalization of more than one trillion dollars.
That outcome was far from obvious at the beginning.
The particular way Bitcoin works also explains why it is difficult to take its history and copy it onto a randomly selected altcoin. A maximum supply, Proof of Work, decentralization, liquidity and the absence of a founding team capable of unlocking 30% of new BTC create a very different economic model.
Looking at Bitcoin today is like observing the winner of a competition after the finish line.
The bias appears when we imagine that its current characteristics were just as easy to identify before the race.
Crypto graveyards are much larger than today’s rankings
Open CoinGecko or CoinMarketCap today.
You see Bitcoin.
Ethereum.
XRP.
BNB.
Solana.
Dogecoin.
Cardano.
Chainlink.
Liquid assets that have been tracked for several years.
It is tempting to take this group and study how “cryptocurrencies” have performed historically.
But this group already contains information from the future: these assets have survived until today.
The cryptocurrencies that disappeared are no longer in your selection.
Tokens whose final trading pairs were removed from exchanges are less visible.
Abandoned projects no longer post on X.
Telegram communities have disappeared.
Websites are no longer renewed.
Developers have left.
Investors who lost 98% generally do not talk about it every morning anymore.
Your screen gradually cleans up the past.
CoinGecko provides a particularly stark illustration of this disappearance. Its study, updated in April 2026, counted approximately 13.4 million cryptocurrencies that had become inactive among the assets tracked in its sample since 2021. For 2025 alone, the figure reached 11,564,909.
By looking only at the assets still visible in 2027, we therefore observe a selected version of history.
The dead no longer produce attractive charts.
The meme coin market pushes the phenomenon to extremes
The ease with which a token can now be created amplifies survivorship bias.
There is no longer any need to develop a blockchain.
A few operations are enough to create an asset on an existing infrastructure.
Then comes the marketing.
A name.
A logo.
An X account.
A Telegram community.
A liquidity pair.
Some tokens disappear a few hours later.
CoinGecko’s Pump.fun study is instructive because it provides the denominator that gain screenshots usually leave out.
Of the 18,675,645 tokens studied, 12,825,175 had recorded their last transaction on the day of launch. Another group of 2.18 million disappeared after just one additional day. The population lasting beyond 90 days represented 850,180 tokens, or 4.55% under the methodology.
Even this final figure should be read cautiously, since CoinGecko notes that some tokens that migrated off the bonding curve to other DEXs may be misclassified.
The broader trend remains difficult to ignore.
When an investor explains how they made 100x on a meme coin, they are telling the story of a survivor.
To understand the real probability of reproducing that result, we would also need to observe all the tokens that resembled the winner at launch and subsequently disappeared.
That is much less spectacular.
It is statistically much more useful.
The winner looks obvious only after winning
Another trap appears when we explain in hindsight why a cryptocurrency succeeded.
“Solana was fast.”
“Ethereum had smart contracts.”
“Dogecoin had a community.”
“Bitcoin was scarce.”
These explanations may contain a great deal of truth.
The problem is forgetting that several projects that disappeared also had highly convincing arguments at the time.
A prestigious team.
An ambitious roadmap.
An active community.
Announced partnerships.
Technology supposedly faster.
A new narrative.
Significant funding.
An inexpensive token.
Many failures had characteristics that seemed excellent before they failed.
In hindsight, the human mind easily reconstructs a coherent story around the winner.
Success appears almost inevitable.
Risk disappears from the narrative.
That is exactly what survivorship bias produces: it replaces an uncertain distribution of outcomes with the neat story of the few players that made it to the present.
“You simply had to hold” is often a survivor’s statement
HODLing is an extremely simple strategy.
Buy.
Do not sell.
With Bitcoin, this approach has historically been remarkably powerful for many investors who made it through several cycles.
It becomes much more dangerous when generalized to all cryptocurrencies.
A cryptocurrency that falls 95% does not merely need to rise 95% to recover its high.
It needs to rise 20x.
A 99% decline requires a 100x gain.
And when liquidity disappears completely, even that arithmetic becomes secondary.
The token may no longer have a sufficiently active market.
Saying that “those who held Bitcoin for ten years made money” therefore does not prove that holding any cryptocurrency for ten years is rational.
Survivors make HODLing visible.
Dead projects make the cost of stubbornness much less visible.
That is why cryptocurrency evaluation must be updated over time. A thesis may be sound at the time of purchase and become false three years later.
Holding is not a virtue in itself.
It depends on what you are holding.
Old ATHs create the same trap
A cryptocurrency once traded at $20.
It is now worth $2.
Some investors conclude that it automatically has 10x potential because it will “return one day” to its previous high.
This reasoning is closely linked to survivorship bias.
We have seen Bitcoin recover and then exceed several previous highs.
Ethereum has done the same in some cycles.
These examples become mentally available.
They create the impression that quality assets naturally recover eventually.
But many altcoins never recover their ATH.
Between the peak and today, their economics may have changed completely.
Circulating supply has increased.
Retail investors have received their tokens.
Competition has developed.
Users have left.
A new blockchain has replaced the old narrative.
Revenue has declined.
Exchanges have reduced the number of trading pairs.
The past price is therefore not a claim on the market.
The previous ATH shows what someone was willing to pay under a different market configuration.
Nothing more.
Survivorship bias also favors influencers
The selection does not concern cryptocurrencies alone.
It also affects people.
Suppose 10,000 social media accounts publish predictions every week.
Some are bullish on Bitcoin.
Others are bearish.
Some buy Solana.
Others choose meme coins.
With enough participants and enough bets, a few people will inevitably produce a spectacular series of good decisions.
These accounts then attract more followers.
Their old posts are reshared.
“They called the breakout.”
“They bought before everyone else.”
“They predicted the 20x.”
Meanwhile, thousands of accounts that made poor predictions become much less visible.
Some stop posting.
Others change their usernames.
Their posts receive little engagement.
The platform itself therefore helps select the survivors.
After several years, users may feel that a group of traders has an extraordinary ability to anticipate the market.
To test that hypothesis, we would need to study all their predictions, not the screenshots that went viral.
Survivorship bias works very well with a retweet button.
Even wealth stories are selected
“I quit my job thanks to crypto.”
This story naturally attracts attention.
“I bought fifteen altcoins, lost 70% and then stopped trading” produces much less viral content.
Both experiences exist.
Only one easily becomes aspirational.
This asymmetry affects our perception of average returns.
If one hundred people each buy a different meme coin and one makes 100x, that person may become an influencer.
The other 99 remain almost invisible.
By observing only the influencer, we may believe that a 100x result is far more common than it is.
To estimate the true probability, the denominator matters as much as the winner.
How many tried?
How many lost?
How many could not exit?
How many recovered their capital?
How many actually realized the advertised gain?
These questions are less exciting.
They are the difference between marketing and statistics.
A public portfolio can hide several dead portfolios
The same problem appears in wallet screenshots.
A trader shows an address that turned $5,000 into $500,000.
The blockchain makes it verifiable.
The story seems solid.
But does this person have only one address?
Or fifty?
If they experimented with fifty wallets and share only the one that succeeded, the visible result does not represent their true performance.
The same logic applies to exchange accounts.
A screenshot shows a PnL of +2,000%.
It does not necessarily show the previous portfolio that was liquidated.
The additional capital injected.
Other strategies.
Positions that are still open.
The selection may be deliberate.
It may also be unintentional.
We naturally prefer to show what worked.
Over time, the internet therefore becomes a gigantic database containing a great many survivors and relatively few graveyards.
The investor must reconstruct the missing data themselves.
Survivorship bias distorts crypto returns, backtests and strategies
The bias becomes even more dangerous when it leaves social media and enters the numbers.
A table may look scientifically rigorous while using a universe already filtered by the future.
This is particularly problematic in crypto, where assets disappear quickly and historical data for small tokens can sometimes be difficult to reconstruct.
Testing today’s cryptocurrencies on the past already gives you the answer
Imagine an analyst wants to know whether buying the ten largest altcoins and holding them for five years is a good strategy.
Today, they open the list of large-cap assets.
They select ETH, XRP, BNB, SOL and a few other survivors.
Then they retrieve their history where available.
They conclude that owning “large cryptocurrencies” was an excellent strategy.
The problem is enormous.
In 2021, they did not know that these ten cryptocurrencies would still be important in 2026.
To correctly simulate the decision of a 2021 investor, they would need to use the list that was actually available in 2021, then keep in the calculation the assets that subsequently declined or disappeared.
Otherwise, the analysis silently uses information from the future.
The CFA Institute specifically cites survivorship bias among the issues analysts must consider in a backtest. It recommends an approach that attempts to reproduce the investment process actually available in each period rather than reconstructing the past with knowledge of the present.
A good backtest should travel into the past without taking a list of winners in its luggage.
Today’s top 100 is not the historical top 100
This mistake is extremely easy to make.
Take the 100 largest cryptocurrencies in September 2026.
Calculate their performance since 2022.
Then use that average to measure what a “diversified altcoin” strategy would have returned.
You have not calculated that.
You have calculated the past performance of the cryptocurrencies that performed well enough or proved resilient enough to be in the top 100 in September 2026.
Assets that were in the top 100 in 2022 but subsequently left it are absent.
That is selection by outcome.
A cleaner methodology must reconstruct the ranking point in time.
What could an investor have bought on the exact date?
What was the supply?
What was the liquidity?
Which exchanges listed the asset?
Then keep the failures in the database.
The difference may seem technical.
It can radically change the calculated return.
In equities, the problem already exists with companies that went bankrupt or were delisted.
In crypto, the speed of disappearances amplifies it.
A crypto index can look better after expelling the losers
Indexes can also create a selection effect.
Suppose an index consists of the 20 largest cryptocurrencies and is rebalanced regularly.
When a token collapses and drops out of the top 20, it is replaced by a better-performing asset.
That is exactly what a dynamic index is designed to do.
The problem arises when we take the current composition and calculate its performance retrospectively as though those assets had always been constituents.
We then eliminate the historical cost of exits.
The retrospective index appears much more robust than a genuinely investable strategy.
This does not mean that all crypto indexes are poorly constructed.
It means that methodology matters.
Creation date.
Selection rules.
Rebalancing.
Delisting treatment.
Minimum liquidity.
Price used at exit.
All of this influences the result.
A clean chart does not guarantee a clean sample.
Bots can be trained on an impossible universe
The problem becomes even more important with bots and artificial intelligence.
A developer wants to train a strategy on “the best liquid cryptocurrencies.”
Today, they download five years of data on BTC, ETH, SOL, BNB, XRP, DOGE, LINK and other assets that remain important.
The model learns perfectly.
Then the backtest shows a superb return.
The trading bot appears ready.
Except that the dataset was built with surviving winners.
The model never learned what happens when a token loses its liquidity.
When it is delisted.
When a protocol shuts down.
When a stablecoin loses its peg.
When an asset becomes practically impossible to sell.
Real-world risk is broader than the history supplied to the algorithm.
The CFA Institute also notes that financial returns often exhibit asymmetry, fat tails and extreme dependencies that standard backtests capture imperfectly.
A strategy can therefore be doubly optimistic: it studies only survivors while also underestimating the extreme events that affected them.
A Bitcoin backtest may answer the wrong question
Bitcoin itself can produce another version of the bias.
Suppose a technical strategy is tested only on BTC since 2012.
It appears highly profitable.
Can we conclude that it will work on “cryptocurrencies”?
No.
Part of the result may simply come from the fact that the underlying asset experienced an extraordinary historical rise.
A strategy that buys almost all the time may look brilliant on an asset that rose by thousands of percent.
To measure its real value, it would need to be compared with a simple buy-and-hold strategy.
Then with different assets.
Then with bearish periods.
Then with assets that did not survive.
Otherwise, the bot may simply be a complicated way to stay long on the sector’s historical winner.
Automation does not correct a biased sample.
It only speeds up the calculations.
The average return of traders can also be misleading
CoinGecko provides an excellent real-world case here.
In its study published in May 2026 on Pump.fun traders, the platform observed that in April 2026 approximately 73.3% of the traders studied showed a positive realized profit. The figure could easily become an extremely optimistic headline: “three out of four meme coin traders make money.”
Then comes the methodology.
CoinGecko specifies that the study counts only realized PnL.
Bagholders who never sold their token despite its collapse toward zero are therefore not fully reflected.
The study itself notes that this underestimates losses.
It also specifies that trades are aggregated at wallet level and that price data for illiquid assets may be imperfect.
This is a valuable example.
The figure is not false.
But the selected population profoundly changes what the figure means.
An investor should therefore always ask:
who is included?
Who is missing?
Sometimes that matters more than the percentage itself.
A platform that disappeared also disappears from some comparisons
Imagine that we want to compare the best exchanges of the past ten years.
We take only the platforms still active in 2027.
Then we calculate their historical liquidity, fees and security.
The conclusion may be flattering.
Yet a real user was not guaranteed to choose only exchanges that would survive.
Some platforms went bankrupt.
Others were hacked.
Some shut down.
Others lost their market.
If these players are removed from the sample, the historical risk of using an exchange appears artificially low.
The same logic applies to:
crypto funds;
stablecoins;
DeFi pools;
lending protocols;
validators;
staking services;
wallets.
The survival of the provider is itself a variable.
In traditional finance as in crypto, selecting only structures that are still standing mechanically reduces the visibility of disasters.
Fund performance can also be embellished
Suppose one hundred crypto funds are launched during a bull market.
After three years, fifty close.
The worst performers disappear mainly because they no longer have enough assets or clients.
We then calculate the average performance of the fifty remaining funds.
It will appear better than that of the original cohort.
The bias is exactly the one historically documented in the traditional fund industry.
The mechanism does not depend on blockchain.
It depends on data selection.
This is why industry-published returns should ideally include funds that were liquidated, closed or merged.
Otherwise, survivors become the statistical norm even though they are precisely the ones that resisted better.
“Bitcoin is dead” stories are true and yet easy to overinterpret
Bitcoin provides a particularly subtle case.
Many media outlets, investors and executives have publicly criticized BTC or predicted its demise at different times.
Comparing their statements with the current price shows how wrong some predictions were.
That is perfectly legitimate.
The problem arises when this result becomes a general rule:
“When everyone says a cryptocurrency is dead, you should buy.”
No.
Bitcoin is precisely the survivor.
Thousands of other cryptocurrencies that appeared “dead” were actually dying.
The contrarian signal works only if the asset still has sufficiently strong fundamentals to survive the crisis.
Buying Bitcoin after a historic panic and buying a token with no developers after a 95% fall are not the same strategy.
Survivorship bias can easily turn BTC’s exceptional resilience into generic advice to catch falling knives.
Even the famous 100x gains are selected from the finish line
A list titled “the cryptocurrencies that made 100x” contains, by definition, cryptocurrencies that made 100x.
It does not identify what should have been bought beforehand.
To be useful, it would need to review all the tokens available at the starting point.
Then measure how many made 100x.
How many made 10x.
How many lost money.
How many fell to zero.
How many had exactly the same visible characteristics as the future winners.
The comparison between Bitcoin and meme coins becomes much more interesting when we examine this full distribution.
Saying that a winning token had “a very strong community” is useful only if the tokens that failed did not also have apparently strong communities.
Otherwise, that criterion may explain nothing.
Hindsight analysis easily confuses the winner’s characteristics with the causes of victory.
Survivorship bias amplifies the illusion of talent
Imagine 10,000 traders each launching ten independent bets.
Even without any particular skill, a few people will achieve extraordinary streaks purely by chance.
If only those people continue publishing their results, their track record will appear remarkable.
The market may then attribute talent to them.
They may eventually come to believe that their success came solely from their method.
The problem becomes particularly acute when a bull market lifts almost every asset in a sector.
The trader chooses five meme coins.
Four explode.
They conclude that they have an exceptional method.
Then the market regime changes.
The model stops working.
Real skill often becomes visible when conditions differ from those that created the initial success.
That is why analyzing returns alone is not enough.
Risk.
Drawdown.
Duration.
Number of trades.
Market regimes.
Capital actually committed.
All this data helps separate luck from skill.
Strategies from 2021 should not be judged using the winners of 2027
This rule summarizes much of the problem.
To evaluate a past decision, we must use the information available at that date.
Not information revealed six years later.
In 2021, no one could know exactly which Layer 1 networks would retain a significant community in 2027.
No one knew with certainty who the future DeFi winners would be.
No one knew which exchange would survive each crisis.
Serious analytical work consists of recovering the uncertainty that existed at the time.
Survivorship bias does the opposite.
It rewrites the past as though the present had always been obvious.
Correcting survivorship bias in crypto completely changes how you invest
It is impossible to eliminate every bias entirely.
We can nevertheless change the method.
The first change is to stop asking only, “which project won?”
The question becomes:
among all the projects that initially appeared comparable, what proportion won?
This simple wording brings the denominator back into the analysis.
Always look for the starting population
Suppose an article claims:
“Eight of the ten best cryptocurrencies in this category outperformed Bitcoin.”
Before concluding that the category has a structural advantage, we must ask how the ten cryptocurrencies were selected.
Top 10 today?
Top 10 at the start of the period?
Still-liquid tokens?
Tokens listed on a single exchange?
If the list is built today, survivorship bias becomes likely.
A more robust analysis should ideally start with the population available at the time of the decision.
For meme coins, that could mean all tokens meeting certain criteria at launch, not only those still active six months later.
For Layer 1 networks, all projects sufficiently capitalized on the date being studied.
For exchanges, all platforms used at the time.
For a bot, all tradable pairs according to the historical rules.
The principle is simple.
Never begin the past with a list built by the future.
Include dead tokens with a value close to zero
When an asset has disappeared, removing it from the dataset is generally the wrong instinct.
That is precisely its outcome.
If a token was worth $10 and then became practically impossible to sell, its economic return is not “missing data.”
For many holders, it is close to a total loss.
The difficulty comes from the data.
The final prices may be unreliable.
Liquidity may be zero.
The exchange may have removed the pair.
We must then define methodologically how to treat the exit.
Last liquid price.
Delisting price.
Zero value after complete disappearance.
The convention may be imperfect.
It is often less misleading than simply removing the asset.
The dataset must preserve the memory of failures.
Use historical snapshots rather than a current list
For a quantitative strategy, the ideal approach is to reconstruct the universe at regular intervals.
January 1, 2023: which assets actually met the criteria?
February 1: which ones?
March 1: which ones?
Then apply the rebalancing rules without looking at the future ranking.
This point-in-time approach requires more work.
It brings the backtest much closer to a genuinely investable situation.
Listing dates must also be taken into account.
A token launched in 2025 obviously cannot appear in a portfolio simulated in 2023.
That seems obvious.
Yet database errors make this kind of anachronism surprisingly common.
The future must remain outside the past.
Compare a complex strategy with a simple benchmark
A bot returns +120% over four years.
Very well.
How much did Bitcoin return over the same period?
A portfolio split 50% BTC and 50% ETH?
A simple DCA strategy?
A broad index built without survivorship bias?
The comparison shows whether sophistication actually added anything.
A bot that gains 100% while Bitcoin gains 300% is not automatically useless if it has much lower volatility.
But that reduction in risk must then be demonstrated.
Complexity must earn its place.
Otherwise, it may simply mask passive exposure to the sector’s dominant survivor.
Test several market regimes
Survivorship bias is often accompanied by period bias.
A strategy designed during a bull market looks excellent.
A breakout bot loves trends.
A grid strategy loves ranges.
A short-selling system can look brilliant during a crisis.
The market changes.
For 2027, a serious backtest should therefore include several environments where the data allows.
Rising market.
Falling market.
Range-bound market.
Liquidity crisis.
High volatility.
Calm market.
The CFA Institute specifically recommends supplementing backtests with scenario analysis and simulations because the available history does not represent every possible future configuration.
The goal is not to prove that a strategy wins everywhere.
An honest strategy can perfectly well have bad periods.
The goal is to know which ones.
Delistings should become data, not omissions
For altcoins, delisting is fundamental information.
A cryptocurrency removed from Binance, Coinbase or other major platforms can lose a great deal of liquidity.
That does not automatically mean it falls to zero.
The risk of exiting the market nevertheless increases.
For a backtest, ignoring delistings amounts to assuming that an investor has access to the same market forever.
That is not realistic.
For a real investment, tracking liquidity is therefore just as important as tracking price.
A token down 60% but still traded across several liquid markets is not in the same situation as a token down 60% for which almost all major platforms are reducing the number of pairs.
The second may be gradually leaving the investable universe.
Liquidity helps identify zombies before they disappear
Not every dead project falls abruptly to zero.
Some become zombies.
The website still exists.
The token is still technically transferable.
A DEX displays a pair.
But volume is almost zero.
The spread is enormous.
Developers post little.
The community has emptied out.
In a purely nominal ranking, the cryptocurrency “still exists.”
Economically, its survival is already questionable.
Volume, order-book depth, the number of active markets and holder concentration allow us to go beyond mere technical existence.
This is particularly important when assessing the real value of a cryptocurrency.
A stated market cap is not necessarily convertible into money.
A survivor without liquidity sometimes looks more like an accounting corpse than an investable asset.
Unlocks create another form of survivorship bias
Young tokens create an additional problem.
Imagine that a project performs brilliantly during its first year.
Its token rises 8x.
It is compared with other launches and presented as a model.
But only 15% of the supply is circulating.
Private investors, the team and the foundation still hold enormous allocations that will be unlocked over four years.
The token has survived the first stage.
That does not mean its economics have made it through its entire dilution cycle.
The bias can therefore appear over time, even before the project disappears.
Assessing a winner too early is sometimes like declaring a marathon runner the winner at kilometer ten.
For 2027, market cap, FDV and the unlock schedule must therefore accompany performance figures.
Survival must be observed over a sufficiently long period to be meaningful.
Survivorship bias does not mean ignoring winners
Beware of the opposite excess.
The fact that Bitcoin is a survivor does not make its history useless.
Quite the opposite.
Longevity itself contains information.
A network that has operated for more than fifteen years, maintained strong liquidity, attracted institutional capital and withstood several crises has demonstrated something that a token created yesterday has not yet demonstrated.
Survival can become part of the fundamental case.
We simply need to avoid two opposite mistakes.
First mistake: treating the survivor as representative of all the projects that once existed.
Second mistake: treating its survival as pure chance and assigning it no value.
Bitcoin is not interesting despite its survival.
Its ability to survive is part of what distinguishes it.
The bias appears when that survival is used to rewrite the probabilities of the past or predict the fate of every future asset.
An older asset may sometimes deserve a premium for its survival
This idea matters when comparing cryptocurrencies.
A project that has operated for eight years has already gone through far more scenarios than a project that is eight months old.
Bear markets.
Outages.
Attack attempts.
Regulatory changes.
Developer departures.
Attention cycles.
Time guarantees nothing.
It nevertheless reduces some unknowns.
This is close to the “Lindy effect” often discussed for technologies or institutions: something that has already survived for a long time can sometimes be considered to have demonstrated greater robustness.
Caution is still required.
An old technology can become obsolete.
An old token can slowly lose its community.
Historical survival is data.
Not insurance.
The best defense remains probabilistic thinking
Survivorship bias becomes particularly powerful when an investor thinks in binary terms.
“This project will be the next Solana.”
“This cryptocurrency will die.”
“This trader is a genius.”
“This strategy works.”
The real world is much more probabilistic.
A small token may have a 5% chance of producing an enormous return and a 70% chance of losing almost all its value.
A large cryptocurrency may have less extreme upside and a greater probability of remaining liquid.
The investment choice then depends on price, position size and risk tolerance.
This reasoning prevents every opportunity from being turned into a certainty.
It also explains why someone can achieve a 50x return without their decision having been rational ex ante.
An improbable event sometimes happens.
Someone always wins the lottery.
The winner does not prove that the ticket had exceptional expected value.
Position sizing helps you survive selection errors
Survivorship bias ultimately offers a lesson in risk management.
If we accept that identifying future survivors is difficult, concentrating all our capital in a young project becomes much more aggressive.
An investor may allocate much smaller amounts to bets whose distribution looks like this:
high probability of failure;
low probability of a huge gain.
Core capital can be placed in more established assets or other asset classes.
This logic guarantees no return.
It prevents a single selection error from ending the entire experiment.
In a universe where millions of tokens can disappear, the ability to keep investing in the next cycle becomes a resource.
Survival matters for the investor too.
Diversification does not mean buying fifty similar tokens
There is, however, a trap.
Buying fifty meme coins does not automatically turn a portfolio into a prudent investment.
If they all depend on the same narrative, the same blockchain and the same appetite for speculation, they can collapse together.
Useful diversification seeks different sources of risk.
Even within crypto, Bitcoin, stablecoins, Layer 1 infrastructure, DeFi and meme coins do not have exactly the same drivers.
And asset diversification obviously extends beyond crypto.
Survivorship bias can lead investors to overweight a sector because we observe its most spectacular winners.
Looking at the losers as well reminds us why it is generally unnecessary to build our entire financial life around a single asset class.
Write the thesis before knowing the outcome
A very simple method helps counter hindsight reconstruction.
Before buying, write down:
why I am buying;
which data supports the idea;
which risks have been identified;
which event would invalidate the thesis;
what time horizon is being considered;
what maximum loss is acceptable.
Six months later, reread the document.
The result may be excellent even though the initial reasoning was poor.
Or the reverse.
This separation between process and outcome prevents the brain from automatically rewriting the story.
A 10x gain does not turn every initial assumption into genius.
A loss does not automatically mean that the entire analysis was absurd.
Uncertainty sometimes produces results that differ from the initial probabilities.
The goal is to improve the method over many decisions.
Not to turn every winning trade into proof of skill.
Social media should be read as an editorial selection
When an impressive screenshot appears, a few questions are enough.
How long has this person been trading?
How many accounts do they have?
How many other positions?
Has the profit been realized?
Is the capital injected known?
Is leverage visible?
Are losses published with the same frequency?
Is the wallet verifiable?
These questions are not meant to automatically accuse the creator of lying.
They simply remind us that what is displayed is a selection.
Even a completely honest person naturally chooses which experiences to share.
Investors should therefore treat success stories as case studies.
Not as statistics about the population.
“Best cryptocurrency” articles should also show disappearances
Crypto media is not immune to the problem.
A media outlet may write about fifty projects every year.
Three become enormous.
A few years later, historical coverage naturally focuses on those three successes.
The other articles grow old at the back of the archives.
This creates a selective editorial memory.
A more useful practice is to revisit old selections.
What became of the project?
Is the token still liquid?
Was the roadmap executed?
Could investors actually have exited?
This culture of follow-up is much less attractive than a new prediction.
It greatly improves journalistic quality.
Survivorship bias is therefore not only an investor problem.
It is also a financial reporting problem.
The survivor must always be compared with the graveyard
This rule summarizes almost everything.
Bitcoin succeeded?
Let us also look at the other assets available at different times.
A trader made 100x?
Let us look at how many people tried comparable strategies.
A bot shows five positive years?
Let us look at the assets that left its database.
A fund is outperforming?
Let us look at the closed funds.
A blockchain has ten years of activity?
Very well. That survival has value, but it does not automatically make the blockchain representative of the thousands of projects launched around it.
The winner tells you what was possible.
The graveyard helps you understand what was probable.
Both pieces of information are necessary.
What the Pump.fun data really says
The 18.67 million tokens studied by CoinGecko may provide the clearest illustration in this article.
Social media allows us to see the few meme coins that became enormous.
The blockchain allows us to see the millions that existed.
CoinGecko estimates that 68.67% of the tokens in its Pump.fun sample recorded their last trade on the day they were created and 80.37% no later than the following day.
That is the denominator.
It does not mean that making money on a meme coin is impossible.
It means that using a few successes to estimate the odds of the next launch is statistically dangerous.
For the speculator, the lesson is not necessarily “never touch meme coins.”
It may be much more practical:
treat every small token as a position that can suffer a total loss;
avoid allowing a single bet to destroy the portfolio;
never confuse current virality with future survival;
and ask whether liquidity will actually make it possible to sell when attention moves elsewhere.
In 2027, the bias could become even more important
Artificial intelligence, launchpads and automation are making token creation ever cheaper.
This is increasing the supply of projects much faster than the attention available to analyze them.
The more candidates there are, the more a few survivors can display extraordinary results.
And the more attractive their stories become.
Imagine 100 million tokens being launched.
Even if 99.99% fail, 0.01% still represents 10,000 potential survivors.
The internet would then have thousands of success stories to tell.
The number of winning stories can therefore increase at the same time as the average probability of success declines.
That is counterintuitive.
And very important.
A larger market produces more absolute winners.
At the same time, it may become harder for each individual new project to survive.
The number of 100x screenshots is therefore not enough to measure opportunity.
We must always return to the total number of attempts.
Survivorship bias also changes how we read Bitcoin
Bitcoin deserves special consideration.
Its survival since 2009 is a major fact.
The network has built liquidity, security, infrastructure and recognition that no new token can reproduce instantly.
This is important information for an investor.
Survivorship bias therefore does not require us to downplay Bitcoin until it becomes ordinary.
It requires us to formulate the conclusion correctly.
Bitcoin has demonstrated an exceptional ability to survive so far.
That strengthens its historical record.
It does not prove that every future criticism will be wrong.
It does not prove that it will never decline again.
Above all, it does not prove that the next token superficially resembling early Bitcoin will have the same fate.
Something may be rare because it is extraordinarily robust.
Not because its trajectory is easy to reproduce.
That is the whole difference.
Conclusion: survivors show what worked, not how many failed
Survivorship bias in crypto is dangerous because it produces stories that are entirely true.
Bitcoin turned small historical investments into fortunes.
Ethereum created a vast programmable ecosystem.
Solana survived periods when many people doubted its future.
Some meme coins generated spectacular multiples.
Some traders genuinely changed their lives through the market.
None of this is necessarily false.
The problem appears when these cases become our main sample.
CoinGecko estimates that more than half of the cryptocurrencies included in its GeckoTerminal universe studied since 2021 had stopped trading actively by the end of 2025. On Pump.fun, nearly seven out of ten tokens studied had recorded their last transaction on the day they launched.
These figures do not mean that crypto is doomed.
They simply indicate that the selection of winners is extremely brutal.
For investing in 2027, looking at survivors remains useful.
We simply need to look at those that did not survive at the same time.
Before a 100x gain, ask how many comparable tokens disappeared.
Before following a trader, ask how many of their previous calls failed.
Before trusting a backtest, check whether delisted assets are still in the data.
Before buying a cryptocurrency down 95% because Bitcoin has already rebounded after major declines, check whether the project still has users, developers, liquidity and an economic reason to exist.
Before concluding that a strategy works, reconstruct the information that was actually available at the time.
That is less spectacular than a portfolio screenshot.
It is much closer to reality.
The crypto market produces a huge number of extraordinary survivors because it produces an even greater number of attempts.
The winner teaches you what was possible.
The number of losers teaches you what was probable.
For 2027, this second piece of information probably deserves much more attention.
