Why You Can't Copy Someone Else's Trading Bot

It's a reasonable question, and I've asked it myself: somewhere out there, people have built automated strategies that work. Why not find out which ones, and start from there rather than from nothing?

The instinct is sound — learning from what works is how most skills are acquired. Trading is unusual in that the normal mechanism breaks down. It's worth understanding precisely why, because the reasons also tell you what is worth doing.

1. You only ever see the survivors

Strategies that made money get posted. Strategies that quietly lost money, or broke in a way that was embarrassing, do not. This isn't dishonesty; it's just what people choose to share.

The consequence is that any collection of "winning bots" has been filtered by outcome rather than by process. Given enough attempts, some strategies will show excellent historical results through chance alone. Those are exactly the ones that get shared. A list of winners tells you nothing about the base rate — how many similar attempts failed — and the base rate is the number you actually need.

2. A shared edge tends to stop being an edge

Market inefficiencies exist because relatively few participants are exploiting them. As more capital pursues the same signal, the opportunity is competed away — the price moves earlier, the spread tightens, the fill gets worse.

This produces an awkward selection effect. A genuine, durable edge is the kind of thing people tend not to publish. What circulates widely is more likely to be either a decayed edge or one that never had statistical support. Popularity is weak evidence of validity, and may be mild evidence against it.

3. Backtested performance and live performance are different things

The gap between a backtest and a live account is wide and consistently underestimated. Some of the reasons:

• Overfitting. Parameters tuned until history looks good are fitted to noise as much as signal. The more variants tested, the more certain this becomes.
• Costs. Spread, commission, funding and slippage compound with turnover. A high-frequency strategy with a small theoretical edge can be comfortably profitable on paper and unprofitable in practice.
• Execution. Backtests assume you got the fill. Live, you may not — particularly around the fast moves the strategy depends on.
• Regime. A strategy tested through one volatility environment may behave very differently in another.

I ran into a plain version of this in my own testing. A well-documented level-trading method appeared reasonable until the timeframe was examined: the 12-tick stop it specifies only makes sense at 1–5 minute resolution. On 30-minute bars, the stop sits inside normal noise and first-touch trades won just 28% of the time. Same rules, same instrument — a detail in the implementation reversed the result. The full study is here.

4. The published base rates are sobering

Regulatory disclosures from retail brokers consistently indicate that a large majority of active retail accounts lose money over time — figures in the 70–90% range are common. Automation doesn't alter the underlying distribution. It removes emotional interference from execution, which is genuinely valuable, while also removing the hesitation that sometimes stops a bad strategy from compounding losses.

What does transfer

What survives the objections above isn't a strategy — it's a way of working:

• Test on the timeframe the signal actually lives on. Convenience of data is not a reason to change resolution.
• State the sample size, and treat small ones as uninformative. A 60% hit rate over 40 trades is close to meaningless.
• Decide the rule before looking at the result. Pre-set triggers are harder to rationalise away afterwards.
• Model costs pessimistically. If the edge disappears under realistic slippage, it wasn't an edge.
• Keep the failures visible. They carry more information than the successes, and they're the part nobody else will show you.

This is why the research on this site reports hit rates that are barely better than a coin flip, and says so. Publishing the studies that didn't work is less impressive than publishing a curve that goes up and to the right. It is also the only version that's any use to somebody trying to judge whether the process is sound.

For educational purposes only. Nothing here is investment advice or a recommendation to trade any strategy or instrument. Figures on retail loss rates vary by broker and jurisdiction — check current disclosures directly.

← All reports