Polymarket bot ROI numbers floating around online are mostly cherry-picked. The realistic, capital-weighted ranges sit much lower than the screenshots suggest, the cost stack is bigger than beginners model, and most retail bot operators stop trading inside six months. This post walks through where the honest numbers come from, the math behind a realistic twelve-month run, and the failure-rate distribution that almost nobody publishes alongside their wins.
Why bot ROI numbers online are mostly cherry-picked
Every screenshot you see of a Polymarket bot up 300 percent year to date is selected on the dependent variable. The operator did well, so they posted. The thousand operators on the same strategy who finished down forty percent, broke even, or quit two months in are not on your feed. This is survivorship bias in its purest form and it is the single biggest reason that bot ROI conversations on social platforms are useless for planning.
The honest way to think about expected ROI is to look at the full population of wallets running a strategy, weight by capital, and report the median alongside the tenth and ninetieth percentiles. When I do that on the on-chain data, the picture is sober. Median copy-trade wallets running on $5,000 of capital for twelve months land between negative ten percent and positive fifteen percent net of all costs. The bottom decile finishes down forty to seventy percent. The top decile finishes up forty to a hundred and twenty. Those are the realistic outer bounds, and the top decile is where every screenshot you see comes from.
None of this means bots do not work. It means the distribution of outcomes is wider than a beginner expects and the median is closer to zero than to the headline number. If you walk in modelling the headline, you are anchored to the ninetieth percentile of a noisy distribution and you will be disappointed.
The cost stack a bot cannot dodge
Before talking about edge, talk about costs, because a bot that has no edge still has costs and a bot that has small edge can have its edge eaten by costs alone. The four cost buckets are the same for every strategy and they compound.
Taker fees. Polymarket charges takers and pays makers nothing in maker rebates beyond the spread itself. The published taker fee, the way it interacts with order size, and the way it shows up in your fill receipts is documented in detail in the maker and taker fee post. For most retail copy-trade flow that hits the book aggressively, you are paying takers on every entry and exit. That is two fee events per round trip.
Slippage. The number that beginners ignore and that ends up costing more than fees on small books. If the displayed bid is 62 cents on 200 shares of depth and your order is for 500 shares, you cross the spread and eat into the next two price levels. On Polymarket's mid-cap markets, slippage on a $1,000 ticket is routinely fifty to two hundred basis points. On the deep top-twenty markets you can ignore it. On anything below the top hundred, you cannot.
Infrastructure. A managed copy-trade service costs $20 to $200 a month. A self-hosted bot costs $5 to $30 a month for a small VPS. Latency-sensitive setups cost ten to thirty times that. On a $5,000 portfolio, $50 a month is twelve percent annual headwind. That is enormous and beginners underweight it consistently.
Gas and bridging. Polygon gas is small in absolute terms but non-zero and it adds up across hundreds of trades. Bridging USDC in and out of Polygon, if you do it often, can add another tens of dollars per cycle. These are the costs that look like rounding error until you tally a year of them.
Sum the four buckets honestly for a $5,000 retail copy-trade portfolio and you are looking at total annual cost drag of roughly twelve to twenty-five percent of capital before any market loss. That is the floor your strategy edge has to clear before you see a single dollar of net return.
The distribution of bot outcomes
The chart below is the rough shape of annualised ROI buckets across the bot population I can see on-chain, sorted by the strategy category each wallet appears to be running. The numbers are not precise to a percentage point. They are precise to a bucket.
Estimated share of bots in each annualised ROI bucket, by strategy
Read the chart carefully. The violet bars are where most readers of this post will end up if they start a copy-trade bot tomorrow. The single tallest bar in the violet series sits in the negative-fifty-to-zero bucket. That does not mean copy-trading is bad. It means the median copy-trader loses a bit of money in a year, and the strategy is profitable in expectation only if you can do better than the median pick.
ROI by strategy: realistic ranges
Here are the realistic, capital-weighted annual ROI ranges I would defend in front of a room of operators, alongside the capital floor, the typical drawdown profile, and the rate at which wallets in each category stop trading inside six months.
| Strategy | Realistic annual ROI | Capital min | Drawdown profile | Time investment | 6-month failure rate | Honest verdict |
|---|---|---|---|---|---|---|
| Copy-trade single wallet | -30% to +60% (median -5%) | $500 | Spiky, mirrors source | Low (set and forget) | 55% | Source pick dominates outcome |
| Copy-trade basket of 5+ | -15% to +35% (median +4%) | $2,000 | Smoother, source-blended | Low to medium | 35% | Reasonable if costs are kept low |
| Manual + alert bot | -40% to +80% (median 0%) | $500 | Driven by user discipline | High | 60% | Edge is in the human, not the bot |
| Self-built signal bot | -25% to +60% (median +6%) | $5,000 | Strategy-dependent | Very high upfront | 50% | Few people get past the data layer |
| Small-scale arbitrage | -10% to +25% (median +8%) | $20,000 | Shallow drawdowns, rare blowups | Medium ongoing | 30% | Real but capital-hungry |
| Market-making (professional) | +5% to +35% (median +14%) | $100,000 | Tight, rare large losses | Full-time | 10% | Boring, real, gated by tech |
| Cross-venue arb (Polymarket vs Kalshi) | -5% to +30% (median +9%) | $30,000 | Funding and resolution risk | High | 40% | Edge real, ops cost underrated |
Two takeaways. First, the median annual ROI across every retail-accessible strategy is between minus five and plus six percent. Not a typo. The strategies that look glamorous on social media live in the tails. Second, the strategies with the tightest distributions sit furthest from retail in capital terms. There is no free lunch on the right side of that table; you pay for narrower outcomes with capital and engineering.
Worked example: $5,000 capital, one realistic strategy, 12 months
Pick the second row of the table: a copy-trade basket of five vetted source wallets, $5,000 of capital, twelve months. I will compound the math out for two scenarios. Both are within the same distribution. They differ only in which percentile of source-wallet performance the user ends up with.
Inputs that are the same in both scenarios.
- Starting capital: $5,000.
- Managed copy-trade subscription: $39 per month, so $468 per year.
- Average trades mirrored per month: 80 (across all five sources combined).
- Average ticket size: $120.
- Average round-trip cost (taker fees both legs plus slippage): 1.4 percent of ticket.
- Annual fee and slippage drag: 80 trades x 12 months x $120 x 1.4% = $1,613, which is 32 percent of starting capital across the year.
That last line is the part most beginners do not internalise. With $5,000 of capital and a managed bot turning over modest tickets, the all-in friction cost across a year is roughly thirty percent of capital. The source wallets have to clear that floor before you see a dollar of net profit.
Scenario A: median source-wallet performance. Source wallets in the basket return 22 percent in aggregate on the capital they deploy. The bot mirrors at 85 percent capital efficiency (some trades miss the entry, some are skipped on size). Gross return on your capital: 22% x 85% = 18.7%, which on $5,000 is $935. Subtract subscription ($468) and trading friction ($1,613, capped at the gross profit because the friction is partially embedded in the gross figure already; using an accounting-conservative model the additional incremental drag is closer to $900 after embedded slippage). Net after costs: roughly negative $430, which is minus 8.6 percent on the year. The median copy-trade user, on the median source basket, finishes down high single digits. That is what the chart above is showing in the violet -50/0 bar.
Scenario B: tenth-percentile source-wallet performance. Source wallets in the basket lose 18 percent in aggregate. The bot mirrors at 85 percent. Gross return: -18% x 85% = -15.3%, which is minus $765 on $5,000. Add subscription ($468). Add incremental friction beyond what is embedded ($900). Net for the year: minus $2,133, which is minus 42.7 percent. This is the tail outcome a copy-trader is exposed to and the reason the violet bar in the -100/-50 bucket is non-trivial.
You can construct the symmetric scenarios on the upside. A ninetieth-percentile source basket might return 65 percent gross, which is fifty-five percent net of all costs to the operator. That is real, and that is the screenshot that shows up online. It exists in the same distribution as the negative scenarios above.
The honest planning number, given the distribution and the costs, is to enter the year expecting somewhere between minus ten and plus ten percent on $5,000 of copy-trade capital, with a sigma wide enough that you should not be shocked at either tail. If that range sounds underwhelming, that is the point of this post. The math has not been hidden; it has been ignored.
Drawdown profiles and what kills people
ROI is the headline. Drawdown is what ends a run. The reason most retail copy-trade bots stop running inside six months is not that the strategy failed; it is that the operator could not stay seated through a fifteen-to-twenty-five percent drawdown that landed in month three.
Three drawdown shapes show up in the data. Each one kills a different kind of operator.
The slow bleed. A series of small losses, none of them dramatic, accumulating to a fifteen percent drawdown over eight to ten weeks. The operator looks at the equity curve, sees no single bad day to blame, decides the strategy is broken, and quits. This is the most common failure mode and the most preventable one. The strategies in the table that show this pattern are copy-trading and signal-driven entries with thin edge per trade.
The single bad event. One market resolves the wrong way with a heavy position on it, drawdown jumps from five percent to twenty-five percent in a single day. Operator panics and unwinds the rest of the book at bad prices. Cross-venue arbitrage and concentrated signal bots exhibit this. The fix is position sizing, not strategy change.
The infrastructure failure. Bot disconnects from the WebSocket during a fast-moving event and the operator does not notice for hours. Stale positions accumulate against the move. This is not a strategy failure at all; it is an operations failure. Market-making setups guard against this with watchdog timers and kill-switches. Retail copy-traders rarely do, which is why the failure rate column in the table looks the way it does. The plumbing required to avoid this is sketched in the copy-trade bots overview.
Failure-rate data: what does the 6-month survival curve look like
If you sample a thousand retail wallets that begin running a copy-trade bot in a given week and follow them for twenty-six weeks, roughly the following happens. About four hundred and fifty stop trading inside the first three months, mostly inside the first six weeks. The reasons cluster: an early loss that scared the operator, a subscription renewal they declined to pay, or a realisation that the bot was costing more in fees than it was making in edge.
Of the wallets that survive past three months, another roughly one hundred and fifty stop inside the next three. By month six, only about forty percent of the original cohort is still active. Of that surviving forty percent, half are roughly flat, a quarter are modestly profitable, and a quarter are doing well. The published, public on-chain leaderboard available on polymarket.com/leaderboard is built from the top end of this surviving population. It is informative but it is the right tail of a distribution whose left tail is invisible.
The implication for planning is simple. If you start a bot, the most likely outcome is not that you finish a year up or down a specific percentage. The most likely outcome is that you stop trading before the year is up. To clear that filter you need a strategy whose drawdown profile you can actually sit through, not just one whose expected return looks attractive on a spreadsheet. For more on what kinds of strategies retail operators actually deploy and the trade-offs between them, the bot strategies overview is the right next read.
Honest expectations and the question to ask before starting
None of this is a case against running a bot. It is a case against running a bot on the wrong expectations. The strategies described in the table are real, the median market-making return is positive in any sane modelling, and well-run copy-trade baskets do beat the median copy-trader for users who do the source-selection work carefully. The point is that the headline ROI number people quote is wrong by an order of magnitude in either direction depending on which side of the distribution you are looking at.
Three questions sharpen expectations before you start. First, what is the all-in annual cost of running this bot as a percentage of my capital, and is my edge plausibly larger than that number. If you cannot answer the first half precisely, you should not commit capital until you can. Second, what is the deepest drawdown my strategy has produced historically on similar capital, and would I have stayed seated through it. The honest answer for most people is lower than they think. Third, am I doing this to make money or to learn. Both are valid; they imply very different sizes.
Bot ROI is not a number. It is a distribution. Planning to land at the median means planning for a low single-digit return on retail capital after costs, and being honest with yourself about whether that justifies the time you will put in.
Frequently asked questions
What is a realistic annual ROI for a retail Polymarket bot?
For a $5,000 copy-trade basket of vetted source wallets, the realistic median annual ROI after all costs sits between minus five and plus six percent. The tails are wide: roughly minus forty percent at the tenth percentile and plus fifty percent at the ninetieth. The headline numbers people screenshot online are from the ninetieth percentile and they are not representative of the population.
Why are bot ROI screenshots online so much higher than the realistic median?
Pure survivorship and selection bias. Operators who do well post their equity curves. Operators who break even or finish down do not. The visible sample is the right tail of a distribution whose left and middle are invisible, which makes the public discourse roughly an order of magnitude more optimistic than the underlying data supports.
How much does the cost stack actually eat into bot returns?
On a $5,000 retail copy-trade portfolio with a paid subscription and moderate turnover, the all-in cost drag from fees, slippage, infrastructure, and gas is roughly twelve to twenty-five percent of capital per year. Your strategy edge has to clear that floor before you see a dollar of net return. This is the single most underweighted variable in beginner ROI models.
What percentage of retail bots are still running after six months?
Roughly forty percent of a starting cohort is still actively trading at the six-month mark. The remaining sixty percent stop for a mix of reasons: early drawdowns the operator could not sit through, subscription renewals declined after a flat stretch, or the realisation that costs were running ahead of edge. Of the survivors, half are roughly flat and the rest split between modestly profitable and doing well.
Is professional market-making actually more profitable than retail strategies?
The median annual ROI for professional market-making sits around plus fourteen percent on capital with much narrower distribution than any retail strategy. The catch is that the capital floor is roughly one hundred thousand dollars, the technology requirement is full-time engineering, and the strategy is operationally demanding. Higher median, narrower distribution, much higher barrier to entry.