A number moved through crypto media this week: tokens endorsed by celebrities are "nearly five times" more likely to be scams. Inside forty-eight hours it had been quoted, screenshotted, reposted into group chats and newsletters and rendered into three languages. The study behind it has been quoted exactly zero times, because the study cannot be located. No institution. No sample size. No time window. No operational definition of the word "scam." No base rate.
Four information points reached the reader. One was a number. The other three were opinions wearing the costume of findings.
I have spent six weeks of my professional life reading smart contracts that other people were hours away from funding. In 2017, working as a quantitative analyst for a Singapore-based VC, I audited the liquidity pool logic of a top-10 ICO and found three integer overflow vulnerabilities in the allocation math. I wrote the report. The investment committee rejected it, because the pitch deck was better designed than the contract. What I carried out of that room was a rule that has held for nine years since: numbers travel, denominators do not. Data doesn"t get quoted in full. It gets quoted in halves.
What follows is an attempt to restore the denominator, and to separate what this brief proves from what it merely echoes.
The Celebrity Endorsement Cycle Has a Public Record, and the Record Is Not Ambiguous
Celebrity promotion of crypto assets is not a new phenomenon and it is not a mystery. It has a documented legal history, which makes the current wave of commentary feel oddly amnesiac — as though each generation of investors discovers the same structure and files it under a new name.
The first cycle ran from 2017 into 2018. Centra Tech raised tens of millions on the claim of a crypto debit card backed by Visa and Mastercard; the card did not exist and the founders were prosecuted. Floyd Mayweather and DJ Khaled promoted token sales on social media without disclosing compensation and settled with the SEC, disgorging gains and paying penalties on top. Steven Seagal paid a civil penalty for failing to disclose that he had been paid to promote an offering. None of these outcomes required a novel legal theory. They required a rule that has existed since 1933 — Section 17(b) of the Securities Act, which prohibits promoting a security without disclosing the consideration received.
The second cycle ran from 2021 into 2022, when celebrity attention migrated to NFTs. The distribution mechanism changed. The structure did not. Attention replaced the product, and when the ice age arrived the attention evaporated first.
The third cycle is the one we are living in. Political tokens. Memecoins launched by accounts with tens of millions of followers. KOL rounds with allocations that are never published. Launchpads that will mint a token for anyone with a wallet and an audience, taking the fee at the top and leaving the residual risk at the bottom.
Three cycles, one structure. Celebrity attention is a liquidity event, not a product event. It manufactures a bid. It does not manufacture cash flow.
Now consider the information chain that produced the number currently circulating:
original research, unattributed -> press release or institutional PR -> crypto-native brief -> reader in a group chat.
Every hop strips uncertainty language. A preprint becomes "a study." A correlation becomes "a finding." A relative risk becomes "five times more likely." By the final hop, the claim has the texture of a measurement and the substance of a rumor. Crypto Briefing published the brief. Crypto Briefing is not the problem. The format is the problem: a brief compresses uncertainty because uncertainty does not fit inside a headline, and the compression is invisible to the reader.
The prior cycles matter because they provide a calibration instrument. We do not have to accept or reject the five-times claim in a vacuum. We can ask whether its direction is consistent with what enforcement records and token structures already demonstrate. It is. That is the entire usable value of the claim — direction, not magnitude.
Relative Risk Without a Base Rate Is Not Information
If the baseline scam rate across launched tokens is one percent, then five times is five percent. If the baseline is five percent, five times is twenty-five percent. If the baseline is twelve percent, five times is sixty percent. Those are three different worlds, and the decision implications are not adjacent — they are orthogonal. At five percent absolute risk a speculative position might be sized. At twenty-five percent it should not exist.
A relative risk figure is the ratio of two incidence rates. It says nothing about the absolute probability of the event an investor actually cares about, which is narrower and simpler: the thing I am about to buy goes to zero and stays there. Those are not the same question. Confusing them is the base rate fallacy, and it does not announce itself. It feels like precision. That is what makes it dangerous — the error arrives dressed as rigor.
I have watched this failure mode in real portfolios. In 2020, managing a two-million-dollar book for a family office in Ho Chi Minh City through DeFi Summer, the metric everyone quoted was APY. A farm advertised four hundred percent and capital moved accordingly, because the number was large and continuously visible on a dashboard. What was not visible was the emission schedule, the sell pressure it implied, and the fact that the yield was denominated in a token the protocol was printing from nothing. When bZx was exploited in April, my pre-defined exit rules mattered more than any yield figure. Stability, I learned, is its own narrative, and it is quieter than four hundred percent. I started writing about sustainable yield versus emission subsidy precisely because the ratio always outshouts the denominator.
"Scam" Is a Definition, and the Definition Is the Conclusion
The most consequential undisclosed item in the entire brief is how the study operationalized the word "scam." Consider the plausible options:
Definition A: regulatory action or judicial finding. The sample shrinks to a few dozen names and the finding becomes conservative, because formal adjudication is rare and slow.
Definition B: token price approaches zero, or the team goes silent. This sweeps in the entire low-quality attention-token category. The conclusion then becomes close to tautological — attention tokens with no product mostly died, therefore attention tokens are scams.
Definition C: liquidity removal, the literal rug pull. Narrow, mechanically detectable, and concentrated in the small-cap long tail where depth is thin enough to be pulled by a single address.
Under Definition B, "celebrity tokens are five times more likely to be scams" is very close to restating "celebrity tokens are low-quality tokens." Under Definition A, the confidence interval becomes a canyon. The study does not tell us which definition it used, which means the conclusion cannot be falsified, and an unfalsifiable claim is not a finding. It is a mood with a decimal point.
Three Confounders That Can Produce the Five-Times Result on Their Own
Self-selection. A project that intends to extract value from retail buyers has a strong incentive to purchase distribution. Celebrity promotion is a distribution purchase, priced like advertising. A project with a working product and a treasury has weaker incentive to pay a public figure to say its name. The causal arrow may therefore run from scam-proneness to endorsement, not from endorsement to scam. The brief assumes the latter because the latter makes a better warning.
Celebrity screening preference. Promoters and their teams select for asymmetric upside, fast liquidity, and minimal legal paperwork. That describes memecoins and pre-launch tokens. It does not describe cash-flowing protocols with compliance departments. The endorsement is a symptom of the selection process, not its cause.
Detection bias — the one almost nobody discusses. Celebrity-endorsed tokens receive more media coverage, more regulator attention, more on-chain forensics, and more aggrieved holders posting publicly. Their failures are legible; their collapses are documented in real time. A failed anonymous token with nine hundred holders dies in silence and never enters a dataset. This is detection bias: the category with more eyes looks sicker, not because it is sicker, but because it is scanned. Part of the five-times ratio may be a difference in discovery rate rather than a difference in fraud rate. Without a methodology note that fraction is unknowable, and it could be large. It could be most of the effect.
If the study had controlled for market capitalization, chain, listing venue, and launch date, and the celebrity coefficient survived at meaningful significance, that would be a genuine contribution to the literature. My estimate is that it would not survive intact. The stronger underlying variable is almost certainly size — small-cap, thin-liquidity tokens are more likely to be scams, because they are cheap to launch and cheap to abandon. Celebrity endorsement correlates heavily with small cap and thin liquidity, so it inherits the effect and collects the credit.
Volume lies. Liquidity speaks. A token with forty thousand dollars of real depth and a famous face is not a market. It is a stage.
Category Structure Explains More Than the Statistic Does
Celebrity tokens do not usually fail because a famous person lied. They fail because the cap table is designed to fail. The category's recurring structure:
Insider and team-linked wallets: frequently twenty to fifty percent of supply or more, with no cliff and no vesting, or a vesting schedule that exists only in the marketing thread.
Sniper allocations: five to twenty percent bought in the first blocks by addresses with privileged RPC access. Sometimes those addresses belong to the same people. Sometimes they belong to people who were told, in advance, that the promotion was coming.
Market maker and exchange reserves: undisclosed, unverifiable, and frequently the actual source of what chart-watchers describe as organic volume.
Retail: the remainder, fully unlocked, positioned at the top of the curve, holding the exit liquidity.
Add one more property, and the structure closes: zero real revenue. No fees. No product. No cash flow. When a token generates no income, price is a pure function of new buyers arriving. After gas, exchange fees, and insider distribution are subtracted, that is a negative-sum game by construction. It does not require malice. It requires arithmetic, and arithmetic does not negotiate.
This is why I stopped treating "is the founder malicious?" as the useful question. Intent is unobservable and legally contested. Structure is legible in a block explorer in under ten minutes. The useful questions are: who is unlocked, who is insulated, and who is being sold to.
Celebrity endorsement and structural defect are also collinear. That is the statistical problem in one sentence. Promotion is associated with no product, no disclosure, high insider concentration, and absent vesting — because the projects that need to buy promotion are precisely the projects that cannot sell themselves. If you regress failure outcomes on "has paid celebrity promotion" and omit "share of insider supply," the celebrity variable absorbs credit that belongs to the tokenomics. The five-times figure may be a proxy wearing a face.
The Regulatory Ledger Is the Most Reliable Part of This Story
Here the record becomes concrete, and it is the reason I would put the direction of the finding at high confidence while holding the magnitude at low confidence.
Under the Howey test, several elements are typically satisfied in a celebrity token. Money is invested. A common enterprise exists. Expectation of profit is generated directly by the endorsement. And delivery depends on the efforts of others. The contested elements are usually the last two — which is exactly where celebrity promotion is most corrosive, because the endorsement exists to manufacture the expectation it then relies on.
But the sharpest legal exposure is not the securities classification question. It is disclosure.
Section 17(b) of the Securities Act prohibits touting a security without disclosing the compensation received. Separately, and frequently overlooked by crypto readers, the FTC's Endorsement Guides require disclosure of material connections between an endorser and a seller — and those guides apply regardless of whether the promoted asset is classified as a security at all. Two independent regulatory surfaces exist: securities law and consumer protection. A promoter who slips past one often walks into the other, and the second one does not require the token to be a security.
In Europe the surface is widening. The UK financial promotion regime imposes risk warnings and cooling-off periods for first-time investors, with criminal liability available for serious breaches. MiCA folds marketing communications into its perimeter, requiring them to be fair, clear, and not misleading, with disclosure obligations attached to the promotion itself. Singapore and Hong Kong have signaled comparable expectations for token marketing.
The net effect is that a compliance path already exists. Disclose the compensation. Add the risk warning. Observe the cooling-off period. The failure combination is not "celebrity." It is celebrity plus non-disclosure plus no product. That is more precise than the popular version of the story, and precision is the only thing that makes a warning actionable.
Note also the timing asymmetry. Enforcement is ex post. It arrives after losses have already been realized. A penalty on a promoter does not return capital to a holder who bought the top of the curve; it prices reputational risk for the next promoter. Regulation here is recovery, not protection.
I have used that asymmetry deliberately. Before the US spot Bitcoin ETF approvals in 2024, I spent three months reading SEC litigation history and prior registration precedents and produced a two-hundred-page internal memo on approval probability. Colleagues were trading memecoins. The memo was not glamorous. It was correct, and the fund outperformed by roughly twenty-five percent because we positioned before clarity arrived. Regulatory clarity is the most durable narrative in this asset class. Warning briefs are its mirror image: they arrive after the repricing, when the losses have already been distributed.
What the Study Should Have Measured
The most useful data in this domain is not price and it is not the word "scam." It is retention and distribution.
Buyer retention at thirty days: what share of unique buyers still hold the token? A category with sub-thirty-percent retention is a churn machine, and that is measurable without any legal definition.
Wallet concentration: what share of supply sits in the top twenty addresses, and how does that concentration change over the first month? Concentration that increases while price falls is insider distribution captured in real time.
Unlock schedule versus price: does the decline track the vesting calendar with uncomfortable fidelity?
Holder quality: does volume come from two hundred addresses cycling, or twenty thousand distinct wallets behaving independently?
In 2022, during the NFT ice age, I systematically reviewed more than five hundred collections looking for anything with marginal utility and a live development team. The collections that held floor value were the ones with recurring revenue — game economies, fractionalized real estate — not the ones with famous names attached. Celebrity drops faded fastest. The metric that separated survivors was user retention, not market capitalization, and it was measurable the entire time. Nobody wanted the chart because it did not move.
A study showing that celebrity tokens have lower thirty-day buyer retention and higher insider concentration would be far more persuasive than a five-times scam ratio. It would be checkable. It would survive replication. It would tell an investor which field to open first.
How I Would Actually Use the Finding
Strip the number. Keep the filter.
A negative screen is a rule that removes an entire population from consideration at close to zero analytical cost. "Any token where a paid promoter has not disclosed compensation" is a legitimate screen, and it is cheap to apply. It does not require the five-times figure to be accurate. It only requires that the removed population has a fatter left tail than the population you keep. On the available evidence — enforcement records, supply concentration, absence of revenue — that assumption is safe.
Here is the discipline in plain terms. Treat a disclosed, compensated endorsement as a pricing signal about distribution spend, not a quality signal about product. Then ask the questions that do not care about the celebrity at all. Who is unlocked. Who is vested. Who is the market maker. What happens to the price curve if the KOL wallets sell in the same hour. What does liquidity depth look like at three in the morning on a weekend, when nobody is watching the order book and the depth is the only thing standing between a holder and zero.
If the answers are bad, the identity of the promoter is irrelevant. That is the whole operational content of this brief, and it survives the absence of a methodology.
The Number Is Useless, and the Warning Still Works
Two contrarian conclusions follow, and they will annoy opposite audiences.
First, the five-times figure is unusable, and that does not matter. It was never a measurement. It was a pointer. The debunking instinct — no methodology, no base rate, therefore ignore entirely — is correct as epistemology and wrong as risk management. The correct move is to downgrade the claim's precision to zero while retaining its direction. That direction is corroborated by three sources the study never cites: enforcement records, the structural arithmetic of attention tokens, and the observed behavior of launchpads and venues that internalize fees at the top of the funnel and externalize losses at the bottom.
Second, and harder to accept: celebrity endorsement may not be an independent risk factor at all. If you control for supply concentration, absence of vesting, absence of revenue, and market capitalization, the celebrity coefficient may collapse toward insignificance. That would mean the entire discourse about celebrity coins is a proxy conversation about something more boring and far more useful — cap table hygiene. The celebrity is the visible surface. The vesting schedule is the cause. One of these gets screenshotted. The other gets ignored.
There is a distributional point worth stating coldly. A bear market for attention tokens benefits the compliance complex. Every brief like this becomes narrative ammunition for regulated venues, custody providers, and spot products. That does not make the brief wrong. It does make the brief useful to someone, and it is worth knowing who is holding the megaphone at the end of the chain.
And a final, unwelcome observation about warning narratives generally: their behavior-change efficiency is close to zero. Attention-to-risk decay in retail crypto is measured in weeks, not years. Risk education does not stop speculation; it delays it by one news cycle. The people who most need this analysis will read the headline and the number. The people who finish it already hold nothing in the category — which is exactly why the category keeps refilling, and why the next brief will quote the next number.
What to Watch, and the Question That Survives the Headline
Signal one: the original study surfaces. If it exists with a disclosed sample, a stated scam definition, and a base rate, the five-times claim can be upgraded or discarded in an afternoon. Until then it is a rumor with a citation-shaped hole in the middle, and it should be handled like one.
Signal two: enforcement moves upstream. The most fragile layer in the celebrity token stack is not the celebrity. It is the marketing and PR intermediary that packages the promotion and invoices for it. If regulators begin charging the intermediaries, the cost of distribution rises and the category contracts at the source rather than at the retail end.
Signal three: exchanges add a disclosure field. If major venues start requiring paid-promotion disclosure as a listing condition, risk reprices before the losses do. That has never happened in this sector. It would be the first genuine structural improvement produced by a study nobody can find.
Meanwhile, the question that survives every headline about this research is not how many times more likely a scam is. It is narrower and more personal. When the next famous name launches a token, will you read the vesting schedule, or will you count the followers?
Code is law, until it isn"t. The five-times number is not a fact. It is a filter. Use it as one.
