AI Backtesting: Test Any Trading Strategy in Plain English

You have a strategy in your head: "long when the 9 EMA crosses the 21, but only above the 200 EMA, stop under the swing low, 2R target." Proving whether it actually makes money used to cost you either a weekend of manual replay or a crash course in Pine Script. AI backtesting removes both: you type the strategy in plain English, and a full backtest report comes back in seconds.

This guide explains what AI backtesting actually is, the one architectural detail that separates a real backtest from a hallucination, and how to run one properly.

What is AI backtesting?

AI backtesting is a two-step pipeline:

1. Translation. An AI model reads your plain-English description and converts it into precise, machine-readable rules: entry condition, direction, stop placement, target, session filter, trade limits.
2. Execution. A backtest engine runs those rules over real historical candles, bar by bar, and records every trade the rules would have produced.

The output is the same report a coded backtest gives you: win rate, profit factor, expectancy, maximum drawdown, an equity curve, and the full trade list. The difference is the input: a sentence instead of a script.

The trap: an AI chatbot cannot backtest

Here's the part most traders miss. If you paste your strategy into a chatbot and ask "backtest this on EURUSD for me," the numbers you get back are invented. A language model doesn't have candle data, doesn't execute rules, and will happily produce a plausible-looking win rate that corresponds to nothing. It's the fastest way to build false confidence in an untested idea.

The AI should only ever translate your strategy. A deterministic engine, running on real price data, must execute it. If you can't re-run the test and get the identical numbers twice, it wasn't a backtest.

That's the architecture behind RB Trading's AI Strategy Backtest: one AI call turns your description into a rule spec you can see and confirm, then a deterministic engine executes those rules over the actual chart candles. Same strategy plus same data equals the same report, every single time. The AI never touches the results.

What an honest AI backtest report includes

• Win rate with the exact W/L split
• Profit factor (gross wins divided by gross losses)
• Expectancy in R, your average edge per trade in risk units
• Maximum drawdown of the equity curve, in R
• The equity curve itself, so you can see the losing streaks you'd have to sit through
• Every individual trade: direction, entry, exit, R result, date
• Edge by time: net R broken down by day of week and session, flagging where the strategy actually earns and where it bleeds

Two honesty details matter more than any headline number:

1. Declared assumptions. A candle-based engine has to decide what happened inside a bar that touched both your stop and your target. The industry standard is TradingView's broker-emulator assumption: price is treated as visiting the extreme nearer the bar's open first. RB Trading's engine uses the same rule (with exact ties counted as losses), so results line up with what a TradingView strategy test would show at the same resolution.

2. Declared blind spots. Plain English can describe rules no candle engine can test: "only trade when the news is quiet," "wait for a clean order block," "trail the stop by feel." A trustworthy AI backtester lists every clause it could NOT convert, instead of silently ignoring it. In RB Trading these show up as amber "not testable" chips before you run, so you always know exactly what was and wasn't in the test.

How to run an AI backtest, step by step

1. Load a chart. In the journal's Backtest tab, pick your symbol, timeframe and start date, and load the candles. Forex, stocks, indices, metals and crypto all work.
2. Describe the strategy. Hit AI Strategy Test and write it the way you'd tell a friend: "Short when price breaks the 20-bar low during New York, stop above the last 8-bar swing high, take profit at 2R, max 3 trades a day."
3. Confirm the rules. The AI shows its interpretation as rule chips: entry, stop, target, filters. Anything it couldn't map is flagged. If a chip is wrong, reword and re-parse.
4. Run it. The engine walks every candle: signals confirm on the close, entries fill at the next bar's open, one position at a time. Tick the deep history option to automatically pull in extra months of older data before the run.
5. Read the report like a sceptic. A 40% win rate with 2R winners is a good system. A 75% win rate over 9 trades is noise. Look at max drawdown and the losing streaks before you look at anything else.
6. Put the trades on the chart. One click draws every entry and exit on the candles, so you can eyeball whether the fills look like trades you'd genuinely have taken. Save the strategy and re-run it on other pairs and timeframes for free.

What AI backtesting cannot do

It can't test discretion. If your entries depend on reading context, "this looks like accumulation," "the move feels exhausted," no rule engine can replicate your eye. For discretionary systems, use bar-by-bar replay with the future hidden and log each decision manually. Our guide to backtesting without lying to yourself covers that workflow in depth.

It can't stop you overfitting. Because AI backtests take seconds, it's tempting to tweak one number at a time until the report glows. That's curve-fitting at machine speed. Decide the rules first, test once, and treat any post-hoc tweak as a brand-new hypothesis that needs fresh, out-of-sample data.

It works at candle resolution. Fills are modelled on OHLC candles without slippage or spread. Real results will run a little worse. Treat every backtest as an upper bound, then forward-test before risking capital.

How the tools compare

The workflow that actually works

1. AI backtest the mechanical core of your idea in seconds. Kill weak ideas cheaply.
2. Replay-drill the survivors bar by bar, so your eyes and hands learn the setup the engine validated.
3. Forward-test small for two weeks, journaling every trade with the same tags.
4. Compare your live numbers against the backtest. If live expectancy holds within reason, size up.

RB Trading Pro Journal covers all four steps in one tool: AI backtesting, replay, and the journal your live trades land in, at $29.99/month with a 30-day money-back guarantee. Describe your strategy, and know in sixty seconds whether it deserves your money.