copy trading bots, and how they lose money

strip the branding off any crypto copy trading bot and you find the same four components. they are worth naming, because vendors compete loudly on the first one and the other three are where your outcome is decided.

the four parts

1. detection. how it learns that the trader did something.
2. sizing. how it converts their trade into your trade.
3. execution. how it gets your order filled.
4. exit. how, and whether, it gets you out.

a bot that is excellent at detection and careless about the other three will reliably lose money faster than doing nothing. this is not a hypothetical failure mode, it is the modal one.

detection: where the latency actually goes

there are two places to learn that a trader has traded, and they are nowhere near each other in time.

a bot that reads notifications or polls an api is standing in the same queue as every human follower, slightly nearer the front. a bot subscribed to the chain is not in the queue at all — it sees the fill as a settled fact at the block, before any notification has been delivered to anyone.

whether that matters depends on how fast the edge decays. we measured 29 first buys from a large fomo wallet: median +39% at one minute, +43% at fifteen, and a fall back toward the middle of the hour's range by sixty minutes. against a curve like that, polling every thirty seconds is not a slightly worse version of the same strategy. it is a different strategy with a worse entry price.

sizing: the part that actually ruins people

this is the largest risk in copy trading and it gets a fraction of the attention that latency does.

a leader who opens a $49,900 position may be risking two percent of their account. copy that proportionally to your $2,000 and you have a rounding error. copy it as a flat ticket you chose while feeling optimistic and it can be a third of everything you have. the ticker is identical and the bet is not remotely the same.

two settings fix most of this, and any bot missing either is unfinished:

• a fixed ticket per copied trade, denominated in your money, not a percentage of theirs.
• a hard cap per trade that overrides the ticket. the ticket bounds the normal trade. the cap bounds the abnormal one, which is the one that matters.

exits: the half that is usually missing

an enormous number of copy tools mirror entries and stop there. this is a defensible engineering decision and an indefensible product one, because on a decaying move the sell carries more of the outcome than the buy.

the arithmetic is unforgiving. if the median move peaks at fifteen minutes and reverts through the hour, a copied entry without a copied exit converts a good trade into a coin flip resolved by whether you happened to be looking at your phone. you have automated the exciting half and kept the hard half manual, which is precisely backwards.

ask any tool one question before funding it: does it copy the sell? if the answer is a paragraph rather than a yes, it is a no.

six ways these things lose money

1. slippage that is never shown. your fill lands after theirs, worse, and on a violent candle much worse. tools that report performance without per-trade slippage are reporting the trader's results, not yours.
2. gas on both legs. every entry and every exit is a transaction. an active leader making several trades a day makes this a real line item rather than a rounding error.
3. copying adds as if they were entries. a trader topping up a two-week-old position is not making the same event as one opening a name they have never held. bots that treat both identically dilute the thing that worked.
4. partial exits that do not translate. the trader sells 30% of a position. at your scale that is below a viable trade size, so it resolves to all or nothing. your risk profile silently stops matching theirs.
5. manual overrides. reviewing each copy before it fires reintroduces every millisecond of the latency you paid to remove, and adds the delay of your own deliberation.
6. tax volume. automation multiplies your trade count. in most jurisdictions each one is a disposal. the paperwork scales with the activity, and nobody plans for it in month one.

the custody question

before any of the above matters, there is one binary question, and it dominates the others:

can this thing move funds out of my wallet, or only trade them in place?

an allowance granted from your own wallet, revocable by you at any moment, means the worst realistic outcome is bad trades. a deposit into an address someone else controls means the worst realistic outcome is a counterparty event, and you are now making two bets where you intended to make one. this distinction is checkable on-chain rather than a matter of trusting a landing page, which is what makes it the first thing to check.

the checklist before you fund anything

• does it copy exits, not just entries?
• does it detect on-chain, or is it reading notifications and polls?
• can i set a flat ticket and a hard cap, independently?
• does it show me realised slippage per trade, or only aggregate performance?
• is there a two-stage kill switch — stop new copies, and close what is open? these are different emergencies.
• can it withdraw? if yes, stop and price that separately.
• who pays gas, and is it disclosed before the first trade or after it?

the honest summary

a copy trading bot does not give you somebody else's judgment. it gives you somebody else's judgment, delayed, at your size, minus costs. the whole engineering problem is shrinking each of those three terms, and the whole marketing problem is that only the first one is fun to talk about.

which is also why the sensible expectation is not that copying turns a bad account into a good one. it is that copying lets you participate in a strategy you could not execute manually — because it happens at 4am, or in fifteen-minute windows, or forty times a week — with the costs made visible instead of hidden.

if the trader you have in mind is on fomo specifically, the mechanics are in how to copy a fomo trader, and the reason latency dominates there is in the first-buy study.