Take a hundred people, give each one a different automated strategy, and let them run for a year. A few will finish well ahead. Those few will have screenshots. The other ninety-something will quietly stop posting. A year later the survivors are selling courses, and the question "do trading bots work?" has been answered by the only people still in the conversation.
That is the shape of almost every claim you will read on this subject. So here is the honest version, which is neither yes nor no.
"Does it work" is three different questions wearing one coat, and they have three different answers.
Conflating those three is how the marketing works. A vendor demonstrates the first, implies the second, and lets you assume the third.
A bot is an execution mechanism, not a source of profit. It does not find an edge, it repeats one. If the rules inside it have no advantage, automating them just produces losses faster and more consistently than you could by hand.
There is a real and measurable benefit here, and it is not the one usually advertised. Automation removes the gap between the plan and the execution.
Research on individual investors has been unkind on this point for decades. Barber and Odean's work on retail brokerage accounts found that the households that traded most actively earned the lowest net returns, with costs and behavior, not stock picking, doing most of the damage. Nothing about a machine improves a bad plan. What it removes is the part of the loss that comes from a good plan being abandoned halfway: the skipped entry, the widened stop, the trade taken because the last one lost. That is the failure mode described in why most traders lose, and it is the one automation genuinely addresses.
A bot does not turn a bad plan into a good one. It turns a good plan into one that actually gets followed.
It cannot predict. It cannot adapt to a market regime it was never shown. It cannot protect you from your own sizing decisions, and it cannot make an unprofitable set of rules profitable through discipline alone. It also cannot stop you switching it off, which turns out to be the most common way an otherwise functional system fails.
Almost everything published about bot performance is affected by survivorship. Failed strategies do not get marketed, discontinued products disappear from the internet, and the screenshots that circulate are the good months of the accounts that lived. You are looking at the winners of a selection process whose losers were deleted, which is why reading a track record is a skill of its own.
The second filter is worse, because it happens before the product exists. Test enough parameter combinations against the same history and some of them will look excellent by chance alone. That is not a hypothetical: it is a documented statistical problem, and the more configurations you try the more certain you are to find a fantastic-looking result with no predictive content. A backtest presented without saying how many variants were tried is missing the number that would tell you what it is worth. The mechanism is explained in overfitting and curve fitting.
| Ask this | Why it settles the question |
|---|---|
| Real account or backtest? | Both are legitimate. Only one is often presented as the other. |
| What account size and how many contracts? | A profit figure without a denominator and a size is decoration. |
| How long, and how many trades? | Seven good months is a sample, not a track record. |
| What was the worst drawdown? | A record with no losing stretch has been curated or curve-fitted. |
| Can I reproduce it myself? | If the answer is no, it is a claim rather than evidence. |
A system that answers all five is not thereby proven. It is simply the only kind that can be checked. Everything else is a picture of a number.
The full backtest is published with its account size, contract count, profit factor and worst drawdown attached, plus the steps to run it yourself in NinjaTrader 8.
Here is the case this site actually makes, with everything attached. Algomyth is two rule-based strategies for NinjaTrader 8. The Algomyth engine takes one trade a day in the 9:30 AM ET window when the US cash market opens and the day's volume arrives, places the stop and the target in the market at the same instant as the entry with the target at three times the risk, and never holds a position overnight.
Take the first one. From January 2022 to September 2026 in backtest, on a mid-size account: $87,359 net, 1,164 trades, a 1.46 profit factor and a worst drawdown of $5,273. Net means net: the commission and a tick of slippage come off every trade before it is counted, and there is no gross figure published anywhere to compare it against, because a gross figure is the easiest number on a report to flatter. With the target at three times the risk, break-even sits at a 25% win rate, so it does not need to be right most of the time. It earns roughly $75 a trade after costs, which is a small statistical advantage repeated a thousand times rather than anything resembling a jackpot. The arithmetic behind that is in how much a trading bot actually makes.
Hypothetical performance. Every figure above comes from a backtest over historical data, not from a live account. Simulated results are prepared with hindsight, carry no financial risk, and cannot fully reflect real execution, slippage or liquidity. Past performance, real or simulated, does not guarantee future results.
It is not proof the system will make money next year. A backtest is a description of the past under a set of assumptions, and the honest way to read one is set out in what a backtest actually is. What the numbers do provide is something a screenshot cannot: a labeled, complete result you can regenerate on your own machine and disagree with.
That is the highest standard available in this business. Anyone claiming a higher one is claiming to know the future, which is where the regulators, and common sense, draw the line.
Yes, in the narrow sense that automation reliably executes a set of rules and removes the human errors that destroy otherwise sound plans. No, in the broad sense the marketing intends, because no bot creates an edge, guarantees a result or protects you from your own decisions about size and patience.
The useful reframe is to stop asking whether trading robots work as a category and start asking whether this set of rules has an edge, whether the evidence for it is checkable, and whether you could sit through its worst stretch without interfering, a question the risk page takes seriously. Those three questions have answers. "Do bots work" does not.
An automated strategy can be profitable if the rules inside it have a genuine edge and the costs do not consume it. The automation itself contributes reliability, not profitability. The realistic version is a small average gain per trade repeated thousands of times, with losing months included, rather than a steady monthly income. Past performance, real or simulated, does not guarantee future results.
Some were never working: they were parameter sets tuned to fit past data and had no predictive content to begin with. Others were built for a market condition that ended, such as a particular volatility regime around the open. And some are abandoned by their owners during an ordinary drawdown, which looks identical to failure from the outside. Distinguishing those three is the whole skill.
Yes, when you know what they are. A backtest tells you how a fixed set of rules would have behaved over a period, under stated assumptions about fills and costs. That is genuinely useful for understanding a strategy's shape, its drawdowns and its trade frequency. It is not a forecast, and a backtest you cannot reproduce yourself is worth less than one you can.
No, and it is often the opposite. A system can win well under half its trades and still make money if the average winner is much larger than the average loser: with a target at three times the risk, break-even arrives at a 25% win rate. A very high advertised win rate usually means small wins and rare large losses, a profile that looks wonderful right up until the large loss arrives.
Ask whether you can reproduce the result, whether the worst drawdown is stated as plainly as the profit, and whether every figure is labeled as real or hypothetical with the account size and contract count attached. Then ask yourself whether you could sit through the published worst stretch without switching it off. If the seller passes the first three and you fail the fourth, the system is not the problem.
Everything we write comes back to one idea: verify, then decide. The live terminal exists for exactly that.
Unlock Algomyth Live