What AI Can and Can’t Do for Retail Traders

Can a machine tell where EUR/USD will be next Friday? The US Commodity Futures Trading Commission has already answered that one. In its warnings about trading bots and signal services, it says plainly that artificial intelligence cannot predict the future or sudden market changes.

None of that has slowed the spread of AI trading tools, and some of them are useful. The hard part for a retail trader is telling those apart from the ones built mainly to sell.

The research is more balanced than either side of the argument tends to admit. Studies of machine learning in financial markets have found that these models can pick out patterns in large datasets and, under the right conditions, improve forecasts of returns, risk and volatility. Better forecasts on historical data are not the same thing as profitable trading, though. A relationship a model finds may disappear when conditions change, or turn out to be too small to cover spreads and commissions once real money is involved.

Part of the confusion is that three different things get sold under the same label. Research assistants summarise information and help traders work through data. Machine-learning models study historical patterns to produce forecasts or signals. Automated trading systems carry out instructions they have been given, with or without any AI inside them. They do different jobs, and none of them can ensure a profitable outcome.

Split the work into tasks and the picture gets clearer:

The jobs on the left are structured and repetitive, which is exactly why software handles them well. They still need checking. Language models can produce confident statements that turn out to be wrong, and the US Financial Industry Regulatory Authority tells firms to test for that kind of error and keep a person reviewing the output.

Risk rules are a slightly different case. Software follows a loss limit every time, including on the days a person would talk themselves out of it, and for many traders that alone is worth having. It is protection, not prediction.

The jobs on the right are the ones that are new, high-stakes or hard to undo. A model trained on years of calm markets has very little to say about something it has never seen.

Backtests are where this bites hardest. One widely cited bond-forecasting study had to publish a correction after information that was not available at the time of each forecast found its way into the data. The corrected results still held up, but they were weaker. If researchers working carefully can make that mistake, a trading system sold with a flawless historical chart should be met with some hard questions.

Live results are usually more sobering. One AI-driven US equity ETF beat the S&P 500 in its fiscal year to September 2025, returning 19.68% against 17.60%. Over the full period from its 2017 launch to that same date, it returned 9.60% a year against the index’s 14.73%. One good year told investors very little.

And then there is the marketing. In 2024, the US Securities and Exchange Commission fined two investment advisers for misleading claims about their use of AI, a practice now known as AI washing. The CFTC has warned separately that fraudsters lean on AI language to sell bots and signals promising high or certain returns.

None of this means automation is off limits. ABHFX, a multi-asset broker, allows Expert Advisors and other automated systems on MetaTrader 5 for supported accounts, may review a system before it runs on a live account, and does not warrant the performance of any automated strategy, whether built in-house or by a third party.

“Software is very good at following a plan,” said a member of the ABHFX Market Research Desk. “It cannot tell you whether the plan was any good, and it will follow a bad one just as faithfully.”

Before trusting any AI trading tool with real money, it is worth asking four questions:

1. Does it promise profit, or a high or certain return? No model can support that claim.
   
   
2. Are its results from live trading, or only from a backtest?
   
   
3. Do those results include spreads, commissions and other costs?
   
   
4. Can its output be checked against an original source?
   
   

If a tool cannot answer those clearly, that is an answer in itself. The ones that can are worth using for what they do well, which is gathering, sorting and enforcing, while the decisions about risk stay with the trader.

The ABHFX Market Research Desk covers the macro and policy events shaping currency, commodity and index markets.

ABH Forex LTD is a multi-asset broker offering forex, commodities, indices, and cryptocurrency CFD trading through MetaTrader 5. Trading CFDs carries a high level of risk and may not be suitable for all investors. Past performance is not indicative of future results.