Anatomy of a memecoin pump: what being early pays, and what chasing costs
We keep a running record of tokens that printed a +100% day. Two experiments on that cohort, from opposite sides of the same trade: what did a follower make copying the earliest wallets in, at realistic delays — and what did a buyer make chasing the pump once it was on the chart? Same tokens, same weeks, opposite answers.
What was measured — and the selection bias you must read it through
| Pump definition | close-over-close ≥ +100% in a day, with ≥ $150k volume |
| Pumps on record | 275 · 265 measurable with full bar data |
| Early entries measured | 4,769 across 193 pumped tokens |
| Price basis | 1-hour bars, per token, from the pump-cohort record |
| Cost assumption | 4% round-trip haircut on every follower outcome |
| Follower delays priced | entry-hour close (≤1h late), +1 hour, +6 hours |
Hindsight selection is inherent to the question
Every token in this cohort did pump — that is how it entered the record. The early-side numbers answer “was following these wallets into these tokens profitable,” not “will following their next entry be.” For the chasing side the same selection is the favourable case: these are the winners, and chasing even the winners produced the losses below. Wherever these numbers travel, this box travels with them.
The early side: an edge that dies by the hour
For every persisted early entry — a wallet that was among the first distinct buyers of a token that later pumped — we price a follower entering at the close of the entry hour, one hour later, and six hours later:
| Follower delay | Median +24h | Median +7d | Median to pump day | Positive net of costs (24h) | Positive net (pump day) |
|---|---|---|---|---|---|
| ≤1h late | +62% | +79.1% | +387% | 65% | 88% |
| +1 hour | +15.4% | +37% | +246.6% | 57% | 82% |
| +6 hours | +6.6% | +18.9% | +170% | 52% | 73% |
Median return per entry. “Positive net” = share of entries still up more than the 4% round-trip haircut.
One hour of delay costs three quarters of the 24-hour edge. Six hours costs nearly all of it. The window is not a block — but it is about an hour.
You don’t need to be first — you need to be early
| Buyer queue position | Entries | Median +24h | Median to pump day |
|---|---|---|---|
| #1–5 | 963 | +65.3% | +421.9% |
| #6–15 | 1,912 | +65% | +387% |
| #16–25 | 1,894 | +53.1% | +370.2% |
Entries at ≤1h follower delay, split by the entry’s rank among the token’s first distinct buyers.
Being fifteenth pays the same as being first. The edge is not a race for block position against snipers — it is presence inside a window that stays open for roughly an hour and, on median, opens 9.9 days before the pump day itself.
The chasing side: buying the close of a +100% day
Same cohort, opposite entry: a chaser buys at the close of the pump day — the moment the move is undeniable on every screener. 265 pumps measured:
| Metric | Value |
|---|---|
| Chaser return, 24h after buying the pump close | -18.2% median · 64% negative |
| Chaser return, +72h | -28.9% median · 66% negative |
| Median retrace from peak by +72h | 60.0% |
| Pumps fully round-tripped within 3 days | 15% — the entire move given back to below the pre-pump close |
| Median time to peak (from pump-day start) | 22h · 63% peak inside the pump day itself |
| Tokens that pumped again (+100% day) within 30 days | 37% of 188 · median gap 3 days |
Nearly two thirds of the move’s peak is typically surrendered within three days. The one number that partially redeems chasing — 37% of tokens print another +100% day within a month — is also the reason the losses are only −18%: some chases get bailed out by a second pump, on a median gap of just three days.
Two sides of the same trade
The wallets that were early kept a median +58% in 24 hours net of costs (copied an hour or less behind them). The people who bought the same tokens at the pump’s close lost a median 18% by the next day — and 64% of them lost something. Same tokens. Same week. The trade exists — the only question is which side of it you are on, and that is decided almost entirely by when you act, not what you pick.
Reading it honestly
- •Entries cluster. Up to 25 wallets ride the same token, so entry-level medians overstate the effective sample. The per-token view (one row per token, median across its entries) is the honest unit and is included in the dataset download — it is noisier, as single price paths are.
- •Sample sizes are stated per cell in the raw JSON. Small cells are suggestive, not conclusive, and the numbers must be regenerated as the cohort grows.
- •One regime. The cohort spans a drawdown market. A regime where pumps extend further could flatter chasing; the delay decay, being intra-day structure, is less regime-exposed but not immune.
- •“To pump day” columns carry the strongest hindsight — they are conditioned on the pump happening. Treat them as descriptive of the cohort, not as expected returns.
What we built from this: our copy-trading surfaces gate on entry recency — an alert that arrives six hours late is, by these numbers, not worth acting on, and the product says so instead of pretending otherwise. Live, scored calls are on the track record.