# Do trading bots work? What the evidence says, and a live record

Bots execute rules, they don't create an edge. Firms' algos earn most of the profit in some markets, but retail bots lack good data. A live record to check.

Source: https://rizon.markets/learn/do-trading-bots-work (updated 2026-11-04)

> **Short answer:** They work in the sense that they do exactly what they're told, fast and without emotion. Whether they make money depends entirely on the strategy. Large firms' algorithms clearly do: in India, [99% of foreign investors' and proprietary firms' profits](https://www.sebi.gov.in/sebi_data/attachdocs/aug-2026/1787233506209.pdf) came from entities using them. For retail bots we could find no reliable study either way, only sellers' claims and [regulators' warnings](https://www.cftc.gov/LearnAndProtect/AdvisoriesAndArticles/AITradingBots.html). [We run a few in public](https://rizon.markets/bots) so you can check for yourself.

## What the evidence says

| Source | Finding |
|---|---|
| [SEBI, India](https://www.sebi.gov.in/sebi_data/attachdocs/aug-2026/1787233506209.pdf), August 2026 | 87.7% of individual derivatives traders lost money in FY26; 99% of foreign investors' and proprietary firms' profits went to "algo entities" |
| [Saguillo and others](https://arxiv.org/abs/2508.03474), 2025 | About $40M of arbitrage profit on Polymarket in a year; the top account made $2M |
| [Bloomberg](https://news.bgov.com/crypto/prediction-market-users-suffer-broad-losses-as-bots-reap-gains), April 2026 | "Prediction market users suffer broad losses as bots reap gains" |
| [Wiecki and others](https://hedgefundalpha.com/strategies/comparing-backtest-sample-performance-large-cohort-trading-algorithms/), 2016, 888 strategies | A backtest's Sharpe ratio predicted live results poorly: R² under 0.025 |

So automation clearly earns money for professionals with speed, data and capital. That says little about a retail bot, since those firms are the competition. "Algo entities" in SEBI's data means firms with at least one algo trade. And the backtest result is a warning about nearly every performance chart you'll see on a bot's sales page.

## What regulators warn about

The [CFTC's advisory](https://www.cftc.gov/LearnAndProtect/AdvisoriesAndArticles/AITradingBots.html) is titled "AI Won't Turn Trading Bots into Money Machines" and says "AI technology can't predict the future or sudden market changes". It cites Mirror Trading International, which took "more than $1.7 billion in bitcoin from at least 23,000 people". [The SEC, NASAA and FINRA](https://www.investor.gov/introduction-investing/general-resources/news-alerts/alerts-bulletins/investor-alerts/artificial-intelligence-fraud) warn against claims like "Our proprietary AI trading system can't lose!"

A simple test: if a bot really made steady money, its owner would run it, not sell it to you.

## A live record, losers included

There's so little honest data on simple bots that we started our own. [The bot record](https://rizon.markets/bots) tracks these around the clock, with paper money, through the same public API as anyone:

| Bot | Rule | What you'd expect |
|---|---|---|
| `COINFLIP-BOT` | Picks a side at random | Loses the spread, slowly |
| `MOMENTUM-BOT` | Follows the last minute's move | The same, unless Bitcoin trends at that scale |
| `REVERSAL-BOT` | Fades the last minute's move | The same, unless Bitcoin mean-reverts at that scale |
| `SHARP-BOT` | Takes quotes a price move has left stale | Makes money, from whoever's slow |
| The market maker | Quotes both sides | Makes money from random flow, which overstates what a real maker earns |

[The code is open](https://github.com/PlathsOven/prediction-market-maker), and the table updates with every trade. Rizon's flow is mostly bots and the money is paper, so this isn't evidence about any real market. What it does show is where edge comes from: speed and structure, not a clever indicator.

## How to tell if your own bot works

1. **Run it live on paper,** on a venue where someone can actually trade with it: [the options](https://rizon.markets/learn/test-a-trading-bot-without-real-money).
2. **Beat a coin flip, after costs.** A rule with no edge loses the spread, so the coin flip is your baseline.
3. **Count enough trades.** Over 50 trades, a coin flip wins 60% or more about 10% of the time. Over 500 it almost never does.
4. **Test it on data you didn't tune it to.** That's what the backtest study is warning you about.
5. **Be honest about who's on the other side.** If your bot's edge depends on slow, random flow, it won't survive a real market.

[You can build one in minutes with Claude Code](https://rizon.markets/learn/how-to-build-a-trading-bot-with-claude-code). Our own blind test got a working bot in 17 minutes, and its first version was run over at once.

## Is a trading bot a good side hustle?

It's a good project. As income, there's no evidence for it at retail size, and the risk of a bug is real. It's "passive" only until something breaks at 3 a.m. with your money attached. For a developer, it's a great way to learn about markets, and it can make [a strong project for a resume](https://rizon.markets/learn/quant-project-for-resume).

## FAQ

### Are trading bots profitable?

Some are, mostly at firms with speed, data and capital. For retail bots there's no reliable public data, and claims mostly come from people selling them.

### Do AI trading bots work?

AI can write and run a bot, but it doesn't give you an edge. [The CFTC warns](https://www.cftc.gov/LearnAndProtect/AdvisoriesAndArticles/AITradingBots.html) that it can't predict the future, and that promises otherwise are a common sign of fraud.

### Are trading bots legal?

Generally yes, and exchanges including Polymarket and Kalshi publish APIs for them. Check each venue's rules, and be careful with market manipulation rules. Selling a bot with false promises is a different matter.

### Can a bot make $500 a day?

A strategy could, with enough capital and an edge. You should assume anyone advertising that number is selling something, and test it yourself.

### How much do you need to start?

Nothing, to find out if it works: paper accounts are free. [Rizon's key takes one call](https://rizon.markets/docs).